repo
stringlengths
2
99
file
stringlengths
13
225
code
stringlengths
0
18.3M
file_length
int64
0
18.3M
avg_line_length
float64
0
1.36M
max_line_length
int64
0
4.26M
extension_type
stringclasses
1 value
NeuroKit
NeuroKit-master/neurokit2/hrv/intervals_utils.py
# -*- coding: utf-8 -*- import numpy as np import scipy def _intervals_successive(intervals, intervals_time=None, thresh_unequal=10, n_diff=1): """Identify successive intervals. Identification of intervals that are consecutive (e.g. in case of missing data). Parameters ---------- intervals : list or ndarray Intervals, e.g. breath-to-breath (BBI) or rpeak-to-rpeak (RRI) intervals_time : list or ndarray, optional Time points corresponding to intervals, in seconds. Defaults to None, in which case the cumulative sum of the intervals is used. thresh_unequal : int or float, optional Threshold at which the difference between time points is considered to be unequal to the interval, in milliseconds. Defaults to 10. n_diff: int, optional The number of times values are differenced. Can be used to check which values are valid for the n-th difference assuming successive intervals. Defaults to 1. Returns ---------- array An array of True/False with True being the successive intervals. Examples ---------- .. ipython:: python import neurokit2 as nk rri = [400, 500, 700, 800, 900] rri_time = [0.7, 1.2, 2.5, 3.3, 4.2] successive_intervals = nk.intervals_successive(rri, rri_time) successive_intervals rri = [400, 500, np.nan, 700, 800, 900] successive_intervals = nk.intervals_successive(rri) successive_intervals """ # Convert to numpy array intervals = np.array(intervals) if intervals_time is None: intervals_time = np.nancumsum(intervals / 1000) else: intervals_time = np.array(intervals_time).astype(float) intervals_time[np.isnan(intervals)] = np.nan diff_intervals_time_ms = np.diff(intervals_time, n=n_diff) * 1000 abs_error_intervals_ref_time = abs( diff_intervals_time_ms - np.diff(intervals[1:], n=n_diff - 1) ) successive_intervals = abs_error_intervals_ref_time <= thresh_unequal return np.array(successive_intervals) def _intervals_time_uniform(intervals_time, decimals=3): """Check whether timestamps are uniformly spaced. Useful for determining whether intervals have been interpolated. Parameters ---------- intervals_time : list or array, optional List or numpy array of timestamps corresponding to intervals, in seconds. decimals : int, optional The precision of the timestamps. The default is 3. Returns ---------- bool Whether the timestamps are uniformly spaced """ return len(np.unique(np.round(np.diff(intervals_time), decimals=decimals))) == 1 def _intervals_sanitize(intervals, intervals_time=None, remove_missing=True): """**Interval input sanitization** Parameters ---------- intervals : list or array List or numpy array of intervals, in milliseconds. intervals_time : list or array, optional List or numpy array of timestamps corresponding to intervals, in seconds. remove_missing : bool, optional Whether to remove NaNs and infinite values from intervals and timestamps. The default is True. Returns ------- intervals : array Sanitized intervals, in milliseconds. intervals_time : array Sanitized timestamps corresponding to intervals, in seconds. intervals_missing : bool Whether there were missing intervals detected. Examples --------- .. ipython:: python import neurokit2 as nk ibi = [500, 400, 700, 500, 300, 800, 500] ibi, ibi_time, intervals_missing = intervals_sanitize(ibi) """ if intervals is None: return None, None else: # Ensure that input is numpy array intervals = np.array(intervals) if intervals_time is None: # Impute intervals with median in case of missing values to calculate timestamps imputed_intervals = np.where( np.isnan(intervals), np.nanmedian(intervals, axis=0), intervals ) # Compute the timestamps of the intervals in seconds intervals_time = np.nancumsum(imputed_intervals / 1000) else: # Ensure that input is numpy array intervals_time = np.array(intervals_time) # Confirm that timestamps are in seconds successive_intervals = _intervals_successive(intervals, intervals_time=intervals_time) if np.all(successive_intervals) is False: # Check whether intervals appear to be interpolated if not _intervals_time_uniform(intervals_time): # If none of the differences between timestamps match # the length of the R-R intervals in seconds, # try converting milliseconds to seconds converted_successive_intervals = _intervals_successive( intervals, intervals_time=intervals_time / 1000 ) # Check if converting to seconds increased the number of differences # between timestamps that match the length of the R-R intervals in seconds if len(converted_successive_intervals[converted_successive_intervals]) > len( successive_intervals[successive_intervals] ): # Assume timestamps were passed in milliseconds and convert to seconds intervals_time = intervals_time / 1000 intervals_missing = _intervals_missing(intervals, intervals_time) if remove_missing: # Remove NaN R-R intervals, if any intervals_time = intervals_time[np.isfinite(intervals)] intervals = intervals[np.isfinite(intervals)] return intervals, intervals_time, intervals_missing def _intervals_missing(intervals, intervals_time=None): if len(intervals[np.isfinite(intervals)]) < len(intervals): return True elif intervals_time is not None: successive_intervals = _intervals_successive(intervals, intervals_time=intervals_time) if not np.all(successive_intervals) and np.any(successive_intervals): # Check whether intervals appear to be interpolated if not _intervals_time_uniform(intervals_time): return True return False def _intervals_time_to_sampling_rate(intervals_time, central_measure="mean"): """Get sampling rate from timestamps. Useful for determining sampling rate used to interpolate intervals. Parameters ---------- intervals_time : list or array, optional List or numpy array of timestamps corresponding to intervals, in seconds. central_measure : str, optional The measure of central tendancy used. Either ``"mean"`` (default), ``"median"``, or ``"mode"``. Returns ---------- bool Whether the timestamps are uniformly spaced """ if central_measure == "mean": sampling_rate = float(1 / np.nanmean(np.diff(intervals_time))) elif central_measure == "median": sampling_rate = float(1 / np.nanmedian(np.diff(intervals_time))) else: sampling_rate = float(1 / scipy.stats.mode(np.diff(intervals_time))) return sampling_rate
7,239
34.145631
103
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_badchannels.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import pandas as pd import scipy.stats from ..signal import signal_zerocrossings from ..stats import hdi, mad, standardize def eeg_badchannels(eeg, bad_threshold=0.5, distance_threshold=0.99, show=False): """**Find bad channels** Find bad channels among the EEG channels. Parameters ---------- eeg : np.ndarray An array (channels, times) of M/EEG data or a Raw or Epochs object from MNE. bad_threshold : float The proportion of indices (for instance, the mean, the SD, the skewness, the kurtosis, etc.) on which an observation is considered an outlier to be considered as bad. The default, 0.5, means that a channel must score as an outlier on half or more of the indices. distance_threshold : float The quantile that defines the absolute distance from the mean, i.e., the z-score for a value of a variable to be considered an outlier. For instance, .975 becomes ``scipy.stats.norm.ppf(.975) ~= 1.96``. The default value (.99) means that all observations beyond 2.33 SD from the mean will be classified as outliers. show : bool Visualize individual EEG channels with highlighted bad channels. Defaults to False Returns ------- list List of bad channel names DataFrame Information of each channel, such as standard deviation (SD), mean, median absolute deviation (MAD), skewness, kurtosis, amplitude, highest density intervals, number of zero crossings. Examples --------- .. ipython:: python import neurokit2 as nk eeg = nk.mne_data("filt-0-40_raw") bads, info = nk.eeg_badchannels(eeg, distance_threshold=0.95, show=False) """ if isinstance(eeg, (pd.DataFrame, np.ndarray)) is False: try: import mne except ImportError as e: raise ImportError( "NeuroKit error: eeg_badchannels(): the 'mne' module is required for this function" " to run. Please install it first (`pip install mne`).", ) from e selection = mne.pick_types(eeg.info, eeg=True) ch_names = np.array(eeg.ch_names)[selection] eeg, _ = eeg[selection] else: ch_names = np.arange(len(eeg)) results = [] for i in range(len(eeg)): channel = eeg[i, :] hdi_values = hdi(channel, ci=0.90) info = { "Channel": [i], "SD": [np.nanstd(channel, ddof=1)], "Mean": [np.nanmean(channel)], "MAD": [mad(channel)], "Median": [np.nanmedian(channel)], "Skewness": [scipy.stats.skew(channel)], "Kurtosis": [scipy.stats.kurtosis(channel)], "Amplitude": [np.max(channel) - np.min(channel)], "CI_low": [hdi_values[0]], "CI_high": [hdi_values[1]], "n_ZeroCrossings": [len(signal_zerocrossings(channel - np.nanmean(channel)))], } results.append(pd.DataFrame(info)) results = pd.concat(results, axis=0) results = results.set_index("Channel") z = standardize(results) results["Bad"] = (z.abs() > scipy.stats.norm.ppf(distance_threshold)).sum(axis=1) / len( results.columns ) bads = ch_names[np.where(results["Bad"] >= bad_threshold)[0]] if show: _plot_eeg_badchannels(eeg, bads, ch_names) return list(bads), results def _plot_eeg_badchannels(eeg, bads, ch_names): # Prepare plot fig, ax = plt.subplots() fig.suptitle("Individual EEG channels") ax.set_ylabel("Voltage (V)") ax.set_xlabel("Samples") bads_list = [] for bad in bads: channel_index = np.where(ch_names == bad)[0] bads_list.append(channel_index[0]) # Prepare colors for plotting colors_good = plt.cm.Greys(np.linspace(0, 1, len(eeg))) colors_bad = plt.cm.autumn(np.linspace(0, 1, len(bads))) # Plot good channels for i in range(len(eeg)): if i not in bads_list: channel = eeg[i, :] ax.plot(np.arange(1, len(channel) + 1), channel, c=colors_good[i]) # Plot bad channels for i, bad in enumerate(bads_list): channel = eeg[bad, :] ax.plot(np.arange(1, len(channel) + 1), channel, c=colors_bad[i], label=ch_names[i]) ax.legend(loc="upper right") return fig
4,410
32.930769
100
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_source.py
def eeg_source(raw, src, bem, method="sLORETA", show=False, verbose="WARNING", **kwargs): """**Source Reconstruction for EEG data** Currently only for mne.Raw objects. Parameters ---------- raw : mne.io.Raw Raw EEG data. src : mne.SourceSpace Source space. See :func:`mne_templateMRI()` to obtain it from an MRI template. bem : mne.Bem BEM model. See :func:`mne_templateMRI()` to obtain it from an MRI template. method : str Can be ``"sLORETA"``, ``"MNE"`` or ``"dSPM"``. See :func:`.mne.minimum_norm.apply_inverse_raw()`. show : bool If ``True``, shows the location of the electrodes on the head. See :func:`.mne.viz.plot_alignment()`. verbose : str Verbosity level for MNE. **kwargs Other arguments to be passed to ``mne.make_forward_solution()`` and ``mne.minimum_norm.make_inverse_operator()`` and ``mne.minimum_norm.apply_inverse_raw()``. See Also -------- mne_templateMRI """ # Try loading mne try: import mne except ImportError as e: raise ImportError( "NeuroKit error: mne_templateMRI(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e if "trans" not in kwargs.keys(): trans = "fsaverage" # MNE has a built-in fsaverage transformation else: trans = kwargs.pop("trans") # Setup source space and compute forward fwd = mne.make_forward_solution( raw.info, trans=trans, src=src, bem=bem, verbose=verbose, **kwargs ) # Get noise covariance matrix noise_cov = mne.compute_raw_covariance(raw, tmin=0, tmax=None) # Get inverse solution inverse_operator = mne.minimum_norm.make_inverse_operator( raw.info, fwd, noise_cov, verbose=verbose, **kwargs ) src = inverse_operator["src"] snr = 1.0 # use smaller SNR for raw data # Compute inverse solution stc = mne.minimum_norm.apply_inverse_raw( raw, inverse_operator, lambda2=1.0 / snr ** 2, method=method, # sLORETA method (could also be MNE or dSPM) verbose=verbose, **kwargs ) # Plot if show is True: # Check that the locations of EEG electrodes is correct with respect to MRI # requires PySide2, ipyvtklink and mayavi mne.viz.plot_alignment( raw.info, src=src, eeg=["original", "projected"], trans=trans, mri_fiducials=True, dig="fiducials", verbose=verbose, ) return stc, src
2,658
31.036145
109
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_crop.py
import numpy as np def mne_crop(raw, tmin=0.0, tmax=None, include_tmax=True, smin=None, smax=None): """**Crop mne.Raw objects** This function is similar to ``raw.crop()`` (same arguments), but with a few critical differences: * It recreates a whole new Raw object, and as such drops all information pertaining to the original data (which MNE keeps, see https://github.com/mne-tools/mne-python/issues/9759). * There is the possibility of specifying directly the first and last samples (instead of in time unit). Parameters ----------- raw : mne.io.Raw Raw EEG data. path : str Defaults to ``None``, assuming that the MNE data folder already exists. If not, specify the directory to download the folder. tmin : float See :func:`mne.Raw.crop()`. tmax : float See :func:`mne.Raw.crop()`. include_tmax : float See :func:`mne.Raw.crop()`. smin : int Cropping start in samples. samx : int Cropping end in samples. Returns ------- mne.io.Raw a cropped mne.Raw object. Examples --------- .. ipython:: python import neurokit2 as nk raw = nk.mne_data(what="raw") raw_cropped = nk.mne_crop(raw, smin=200, smax=1200, include_tmax=False) len(raw_cropped) """ # Try loading mne try: import mne except ImportError as e: raise ImportError( "NeuroKit error: eeg_channel_add(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e # Convert time to samples if smin is None or smax is None: max_time = (raw.n_times - 1) / raw.info["sfreq"] if tmax is None: tmax = max_time if tmin > tmax: raise ValueError(f"tmin ({tmin}) must be less than tmax ({tmax})") if tmin < 0.0: raise ValueError(f"tmin ({tmin}) must be >= 0") elif tmax > max_time: raise ValueError( f"tmax ({tmax}) must be less than or equal to the max time ({max_time} sec)." ) # Convert time to first and last samples new_smin, new_smax = np.where( _time_mask(raw.times, tmin, tmax, sfreq=raw.info["sfreq"], include_tmax=include_tmax) )[0][[0, -1]] if smin is None: smin = new_smin if smax is None: smax = new_smax if include_tmax: smax += 1 # Re-create the Raw object (note that mne does smin : smin + 1) raw = mne.io.RawArray(raw._data[:, int(smin) : int(smax)].copy(), raw.info, verbose="WARNING") return raw def _time_mask(times, tmin=None, tmax=None, sfreq=None, raise_error=True, include_tmax=True): """Copied from https://github.com/mne-tools/mne-python/mne/utils/numerics.py#L466.""" orig_tmin = tmin orig_tmax = tmax tmin = -np.inf if tmin is None else tmin tmax = np.inf if tmax is None else tmax if not np.isfinite(tmin): tmin = times[0] if not np.isfinite(tmax): tmax = times[-1] include_tmax = True # ignore this param when tmax is infinite if sfreq is not None: # Push to a bit past the nearest sample boundary first sfreq = float(sfreq) tmin = int(round(tmin * sfreq)) / sfreq - 0.5 / sfreq tmax = int(round(tmax * sfreq)) / sfreq tmax += (0.5 if include_tmax else -0.5) / sfreq else: assert include_tmax # can only be used when sfreq is known if raise_error and tmin > tmax: raise ValueError(f"tmin ({orig_tmin}) must be less than or equal to tmax ({orig_tmax})") mask = times >= tmin mask &= times <= tmax if raise_error and not mask.any(): extra = "" if include_tmax else "when include_tmax=False " raise ValueError( f"No samples remain when using tmin={orig_tmin} and tmax={orig_tmax} {extra}" "(original time bounds are [{times[0]}, {times[-1]}])" ) return mask
4,041
32.966387
105
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_gfp.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd from ..signal import signal_filter from ..stats import mad, standardize def eeg_gfp( eeg, sampling_rate=None, method="l1", normalize=False, smooth=0, robust=False, standardize_eeg=False, ): """**Global Field Power (GFP)** Global Field Power (GFP) constitutes a reference-independent measure of response strength. GFP was first introduced by Lehmann and Skrandies (1980) and has since become a commonplace measure among M/EEG users. Mathematically, GFP is the standard deviation of all electrodes at a given time. Parameters ---------- eeg : array An array (channels, times) of M/EEG data or a Raw or Epochs object from MNE. sampling_rate : int The sampling frequency of the signal (in Hz, i.e., samples/second). Only necessary if smoothing is requested. method : str Can be either ``l1`` or ``l2`` to use the L1 or L2 norm. normalize : bool Normalize GFP. smooth : float Can be either ``None`` or a float. If a float, will use this value, multiplied by the sampling rate. robust : bool If ``True``, the GFP extraction (and the data standardization if requested) will be done using the median/MAD instead of the mean/SD. standardize_eeg : bool Standardize (z-score) the data across time prior to GFP extraction using ``nk.standardize()``. Returns ------- gfp : array The global field power of each sample point in the data. Examples --------- .. ipython:: python import neurokit2 as nk eeg = nk.mne_data("filt-0-40_raw") eeg = nk.eeg_rereference(eeg, 'average') eeg = eeg.get_data()[:, 0:500] # Get the 500 first data points * **Example 1:** Compare L1 and L2 norms .. ipython:: python l1 = nk.eeg_gfp(eeg, method="l1", normalize=True) l2 = nk.eeg_gfp(eeg, method="l2", normalize=True) @savefig p_eeg_gfp1.png scale=100% nk.signal_plot([l1, l2]) @suppress plt.close() * **Example 2:** Compare Mean-based and Median-based .. ipython:: python gfp = nk.eeg_gfp(eeg, normalize=True) gfp_r = nk.eeg_gfp(eeg, normalize=True, robust=True) @savefig p_eeg_gfp2.png scale=100% nk.signal_plot([gfp, gfp_r]) @suppress plt.close() * **Example 3:** Standardize the data .. ipython:: python gfp = nk.eeg_gfp(eeg, normalize=True) gfp_z = nk.eeg_gfp(eeg, normalize=True, standardize_eeg=True) @savefig p_eeg_gfp3.png scale=100% nk.signal_plot([gfp, gfp_z]) @suppress plt.close() References ---------- * Lehmann, D., & Skrandies, W. (1980). Reference-free identification of components of checkerboard-evoked multichannel potential fields. Electroencephalography and clinical neurophysiology, 48(6), 609-621. """ # If MNE object if isinstance(eeg, (pd.DataFrame, np.ndarray)) is False: sampling_rate = eeg.info["sfreq"] eeg = eeg.get_data() # Normalization if standardize_eeg is True: eeg = standardize(eeg, robust=robust) # Compute GFP if method.lower() == "l1": gfp = _eeg_gfp_L1(eeg, robust=robust) else: gfp = _eeg_gfp_L2(eeg, robust=robust) # Normalize (between 0 and 1) if normalize is True: gfp = gfp / np.max(gfp) # Smooth if smooth is not None and smooth != 0: gfp = _eeg_gfp_smoothing(gfp, sampling_rate=sampling_rate, window_size=smooth) return gfp # ============================================================================= # Utilities # ============================================================================= def _eeg_gfp_smoothing(gfp, sampling_rate=None, window_size=0.02): """Smooth the Global Field Power Curve""" if sampling_rate is None: raise ValueError( "NeuroKit error: eeg_gfp(): You requested to smooth the GFP, for which ", "we need to know the sampling_rate. Please provide it as an argument.", ) window = int(window_size * sampling_rate) if window > 2: gfp = signal_filter(gfp, method="savgol", order=2, window_size=window) return gfp # ============================================================================= # Methods # ============================================================================= def _eeg_gfp_L1(eeg, robust=False): if robust is False: gfp = np.sum(np.abs(eeg - np.mean(eeg, axis=0)), axis=0) / len(eeg) else: gfp = np.sum(np.abs(eeg - np.median(eeg, axis=0)), axis=0) / len(eeg) return gfp def _eeg_gfp_L2(eeg, robust=False): if robust is False: gfp = np.std(eeg, axis=0, ddof=0) else: gfp = mad(eeg, axis=0) return gfp
4,869
29.061728
102
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_data.py
# -*- coding: utf-8 -*- def mne_data(what="raw", path=None): """**Access MNE Datasets** Utility function to easily access MNE datasets. Parameters ----------- what : str Can be ``"raw"`` or ``"filt-0-40_raw"`` (a filtered version). path : str Defaults to ``None``, assuming that the MNE data folder already exists. If not, specify the directory to download the folder. Returns ------- object The raw mne object. Examples --------- .. ipython:: python import neurokit2 as nk raw = nk.mne_data(what="raw") raw = nk.mne_data(what="epochs") """ # Try loading mne try: import mne except ImportError: raise ImportError( "NeuroKit error: mne_data(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) old_verbosity_level = mne.set_log_level(verbose="WARNING", return_old_level=True) # Find path of mne data if path is None: try: path = str(mne.datasets.sample.data_path()) except ValueError: raise ValueError( "NeuroKit error: the mne sample data folder does not exist. ", "Please specify a path to download the mne datasets.", ) # Raw if what in ["raw", "filt-0-40_raw"]: path += "/MEG/sample/sample_audvis_" + what + ".fif" data = mne.io.read_raw_fif(path, preload=True) data = data.pick_types(meg=False, eeg=True) # Epochs elif what in ["epochs", "evoked"]: raw = mne.io.read_raw_fif(path + "/MEG/sample/sample_audvis_filt-0-40_raw.fif").pick_types( meg=False, eeg=True ) events = mne.read_events(path + "/MEG/sample/sample_audvis_filt-0-40_raw-eve.fif") event_id = {"audio/left": 1, "audio/right": 2, "visual/left": 3, "visual/right": 4} data = mne.Epochs(raw, events, event_id, tmin=-0.2, tmax=0.5, baseline=(None, 0)) if what in ["evoked"]: data = [data[name].average() for name in ("audio", "visual")] else: raise ValueError("NeuroKit error: mne_data(): the 'what' argument not recognized.") mne.set_log_level(old_verbosity_level) return data
2,308
28.602564
99
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_rereference.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd def eeg_rereference(eeg, reference="average", robust=False, **kwargs): """**EEG Rereferencing** This function can be used for arrays as well as MNE objects. EEG recordings measure differences in electrical potentials between two points, which means the signal displayed at any channel is in fact the difference in electrical potential to some other recording site. Primarily, this other recording site is the ground electrode, which picks up electrical noise that does not reach the other scalp electrodes. Consequently, the voltage difference between ground and EEG scalp electrodes is also affected by this noise. The idea behind re-referencing is to express the voltage at the EEG scalp channels with respect to another, new reference. It can be composed of any recorded channel or an average of several channels. Parameters ----------- eeg : np.ndarray An array (channels, times) of M/EEG data or a Raw or Epochs object from MNE. reference : str See :func:`.mne.set_eeg_reference()`. Can be a string (e.g., 'average', 'lap' for Laplacian "reference-free" transformation, i.e., CSD), or a list (e.g., ['TP9', 'TP10'] for mastoid reference). robust : bool Only applied if reference is ``average``. If ``True``, will substract the median instead of the mean. **kwargs Optional arguments to be passed into ``mne.set_eeg_rereference()``. Returns ------- object The rereferenced raw mne object. Examples --------- .. ipython:: python import neurokit2 as nk raw = nk.mne_data("filt-0-40_raw") eeg = raw.get_data() * **Example 1:** Difference between robust average .. ipython:: python avg = nk.eeg_rereference(eeg, 'average', robust=False) avg_r = nk.eeg_rereference(eeg, 'average', robust=True) @savefig p_eeg_rereference1.png scale=100% nk.signal_plot([avg[0, 0:1000], avg_r[0, 0:1000]], labels=["Normal", "Robust"]) @suppress plt.close() * **Example 2:** Compare the rereferencing of an array vs. the MNE object .. ipython:: python avg_mne = raw.copy().set_eeg_reference('average', verbose=False) @savefig p_eeg_rereference2.png scale=100% nk.signal_plot([avg[0, 0:1000], avg_mne.get_data()[0, 0:1000]]) @suppress plt.close() * **Example 3:** Difference between average and LAP .. ipython:: python lap = nk.eeg_rereference(raw, 'lap') @savefig p_eeg_rereference3.png scale=100% nk.signal_plot( [avg_mne.get_data()[0, 0:1000], lap.get_data()[0, 0:1000]], standardize=True ) @suppress plt.close() References ----------- * Trujillo, L. T., Stanfield, C. T., & Vela, R. D. (2017). The effect of electroencephalogram (EEG) reference choice on information-theoretic measures of the complexity and integration of EEG signals. Frontiers in Neuroscience, 11, 425. """ # If MNE object if isinstance(eeg, (pd.DataFrame, np.ndarray)): eeg = eeg_rereference_array(eeg, reference=reference, robust=robust) else: eeg = eeg_rereference_mne(eeg, reference=reference, robust=robust, **kwargs) return eeg # ============================================================================= # Methods # ============================================================================= def eeg_rereference_array(eeg, reference="average", robust=False): # Average reference if reference == "average": if robust is False: eeg = eeg - np.mean(eeg, axis=0, keepdims=True) else: eeg = eeg - np.median(eeg, axis=0, keepdims=True) else: raise ValueError( "NeuroKit error: eeg_rereference(): Only 'average' rereferencing", " is supported for data arrays for now.", ) return eeg def eeg_rereference_mne(eeg, reference="average", robust=False, **kwargs): eeg = eeg.copy() if reference == "average" and robust is True: # Assigning "custom_ref_applied" to True throws an error with the # latest MNE. If this error goes away in the future, we might able to # restore this feature. # > eeg._data = eeg_rereference_array(eeg._data, reference=reference, robust=robust) # > eeg.info["custom_ref_applied"] = True raise ValueError( "NeuroKit error: eeg_rereference(): 'robust=True' currently not supported for MNE", " objects.", ) elif reference in ["lap", "csd"]: try: import mne if mne.__version__ < "0.20": raise ImportError except ImportError as e: raise ImportError( "NeuroKit error: eeg_rereference(): the 'mne' module (version > 0.20) is required " "for this function to run. Please install it first (`pip install mne`).", ) from e old_verbosity_level = mne.set_log_level(verbose="WARNING", return_old_level=True) eeg = mne.preprocessing.compute_current_source_density(eeg, **kwargs) # Reconvert CSD type to EEG (https://github.com/mne-tools/mne-python/issues/11426) # channels = np.array(eeg.ch_names)[mne.pick_types(eeg.info, csd=True)] # eeg.set_channel_types(dict(zip(channels, ["eeg"] * len(channels)))) mne.set_log_level(old_verbosity_level) else: eeg = eeg.set_eeg_reference(reference, verbose=False, **kwargs) return eeg
5,622
35.512987
99
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_1f.py
# -*- coding: utf-8 -*- # import numpy as np # import pandas as pd # def mne_channel_extract(raw): # """1/f Neural Noise # Extract parameters related to the 1/f structure of the EEG power spectrum. # Parameters # ---------- # raw : mne.io.Raw # Raw EEG data. # Returns # ---------- # DataFrame # A DataFrame or Series containing the channel(s). # Example # ---------- # >>> import neurokit2 as nk # >>> import mne # >>> # >>> raw = nk.mne_data("raw") # >>> # """ # import mne # import neurokit2 as nk # raw = nk.mne_data("raw") # raw.plot_psd(fmin=0, fmax=40.0, picks=["EEG 050"]) # channel = nk.mne_channel_extract(raw, what=["EEG 050"]).values # psd = nk.signal_psd( # channel, sampling_rate=raw.info["sfreq"], show=True, max_frequency=40, method="multitapers" # ) # plt.loglog(psd["Frequency"], psd["Power"]) # plt.plot(psd["Frequency"], np.log(psd["Power"]))
996
22.738095
101
py
NeuroKit
NeuroKit-master/neurokit2/eeg/utils.py
import numpy as np import pandas as pd from .mne_to_df import mne_to_df def _sanitize_eeg(eeg, sampling_rate=None, time=None): """Convert to DataFrame Input can be an array (channels, time), or an MNE object. Examples --------- >>> import neurokit2 as nk >>> >>> # Raw objects >>> eeg = nk.mne_data("raw") """ # If array (channels, time), tranpose and convert to DataFrame if isinstance(eeg, np.ndarray): eeg = pd.DataFrame(eeg.T) eeg.columns = [f"EEG_{i}" for i in range(eeg.shape[1])] # If dataframe if isinstance(eeg, pd.DataFrame): return eeg, sampling_rate, time # Probably an mne object else: sampling_rate = eeg.info["sfreq"] eeg = mne_to_df(eeg) time = eeg["Time"].values eeg = eeg.drop(columns=["Time"]) return eeg, sampling_rate, time
874
22.648649
66
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_source_extract.py
import pandas as pd def eeg_source_extract(stc, src, segmentation="PALS_B12_Lobes", verbose="WARNING", **kwargs): """**Extract the activity from an anatomical source** Returns a dataframe with the activity from each source in the segmentation. Parcellation models include: * 'aparc' * 'aparc.a2005s' * 'aparc.a2009s' * 'oasis.chubs' * 'PALS_B12_Brodmann' * 'PALS_B12_Lobes' * 'PALS_B12_OrbitoFrontal' * 'PALS_B12_Visuotopic' * 'Yeo2011_17Networks_N1000' * 'Yeo2011_7Networks_N1000' Parameters ---------- stc : mne.SourceEstimate An SourceEstimate object as obtained by ``eeg_source()``. src : mne.SourceSpaces An SourceSpaces object as obtained by ``eeg_source()``. segmentation : str See above. verbose : str Verbosity level for MNE. **kwargs Other arguments to be passed to ``mne.extract_label_time_course()``. Examples --------- .. ipython:: python :verbatim: import neurokit2 as nk raw = nk.mne_data("filt-0-40_raw") src, bem = nk.mne_templateMRI() stc, src = nk.eeg_source(raw, src, bem) data = nk.eeg_source_extract(stc, src, segmentation="PALS_B12_Lobes") data.head() """ # Try loading mne try: import mne except ImportError as e: raise ImportError( "NeuroKit error: mne_templateMRI(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e # Find labels labels = mne.read_labels_from_annot( subject="fsaverage", parc=segmentation, subjects_dir=str(mne.datasets.sample.data_path()) + "/subjects", verbose=verbose, ) # Filter empty ones labels = [lab for lab in labels if len(lab) > 0] # Filter Unknown ones labels = [lab for lab in labels if "?" not in lab.name] tcs = stc.extract_label_time_course( labels, src=src, verbose=verbose, **kwargs, ) return pd.DataFrame(tcs.T, columns=[lab.name for lab in labels])
2,135
25.7
105
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_diss.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd from .eeg_gfp import eeg_gfp def eeg_diss(eeg, gfp=None, **kwargs): """**Global dissimilarity (DISS)** Global dissimilarity (DISS) is an index of configuration differences between two electric fields, independent of their strength. Like GFP, DISS was first introduced by Lehmann and Skrandies (1980). This parameter equals the square root of the mean of the squared differences between the potentials measured at each electrode (versus the average reference), each of which is first scaled to unitary strength by dividing by the instantaneous GFP. Parameters ---------- eeg : np.ndarray An array (channels, times) of M/EEG data or a Raw or Epochs object from MNE. gfp : list The Global Field Power (GFP). If ``None``, will be obtained via ``eeg_gfp()``. **kwargs Optional arguments to be passed into ``nk.eeg_gfp()``. Returns ------- np.ndarray DISS of each sample point in the data. Examples --------- .. ipython:: python import neurokit2 as nk eeg = nk.mne_data("filt-0-40_raw") eeg = eeg.set_eeg_reference('average') gfp = nk.eeg_gfp(eeg) diss = nk.eeg_diss(eeg, gfp=gfp) @savefig p_eeg_diss1.png scale=100% nk.signal_plot([gfp[0:300], diss[0:300]], standardize=True) @suppress plt.close() References ---------- * Lehmann, D., & Skrandies, W. (1980). Reference-free identification of components of checkerboard-evoked multichannel potential fields. Electroencephalography and clinical neurophysiology, 48(6), 609-621. """ if isinstance(eeg, (pd.DataFrame, np.ndarray)) is False: eeg = eeg.get_data() if gfp is None: gfp = eeg_gfp(eeg, **kwargs) normalized = eeg / gfp diff = np.diff(normalized, axis=1) diss = np.mean(np.power(diff, 2), axis=0) # Preserve length diss = np.insert(diss, 0, 0, axis=0) return diss
2,026
27.957143
99
py
NeuroKit
NeuroKit-master/neurokit2/eeg/__init__.py
"""Submodule for NeuroKit.""" from .eeg_badchannels import eeg_badchannels from .eeg_diss import eeg_diss from .eeg_gfp import eeg_gfp from .eeg_power import eeg_power from .eeg_rereference import eeg_rereference from .eeg_simulate import eeg_simulate from .eeg_source import eeg_source from .eeg_source_extract import eeg_source_extract from .mne_channel_add import mne_channel_add from .mne_channel_extract import mne_channel_extract from .mne_crop import mne_crop from .mne_data import mne_data from .mne_templateMRI import mne_templateMRI from .mne_to_df import mne_to_df, mne_to_dict __all__ = [ "mne_data", "mne_channel_add", "mne_channel_extract", "mne_crop", "mne_to_df", "mne_to_dict", "mne_templateMRI", "eeg_simulate", "eeg_source", "eeg_source_extract", "eeg_power", "eeg_rereference", "eeg_gfp", "eeg_diss", "eeg_badchannels", ]
904
24.857143
52
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_templateMRI.py
import os def mne_templateMRI(verbose="WARNING"): """**Return Path of MRI Template** This function is a helper that returns the path of the MRI template for adults (the ``src`` and the ``bem``) that is made available through ``"MNE"``. It downloads the data if need be. These templates can be used for EEG source reconstruction when no individual MRI is available. See https://mne.tools/stable/auto_tutorials/forward/35_eeg_no_mri.html Parameters ---------- verbose : str Verbosity level for MNE. Examples --------- .. ipython:: python import neurokit2 as nk src, bem = nk.mne_templateMRI() """ # Try loading mne (requires also the 'pooch' package) try: import mne except ImportError as e: raise ImportError( "NeuroKit error: mne_templateMRI(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e # Download fsaverage files fs_dir = mne.datasets.fetch_fsaverage(verbose=verbose) # The files live in: src = os.path.join(fs_dir, "bem", "fsaverage-ico-5-src.fif") bem = os.path.join(fs_dir, "bem", "fsaverage-5120-5120-5120-bem-sol.fif") return src, bem
1,277
28.72093
105
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_channel_extract.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd def mne_channel_extract(raw, what, name=None, add_firstsamples=False): """**Channel extraction from MNE objects** Select one or several channels by name and returns them in a dataframe. Parameters ---------- raw : mne.io.Raw Raw EEG data. what : str or list Can be ``"MEG"``, which will extract all MEG channels, ``"EEG"``, which will extract all EEG channels, or ``"EOG"``, which will extract all EOG channels (that is, if channel names are named with prefixes of their type e.g., 'EEG 001' etc. or 'EOG 061'). Provide exact a single or a list of channel's name(s) if not (e.g., ['124', '125']). name : str or list Useful only when extracting one channel. Can also take a list of names for renaming multiple channels, Otherwise, defaults to ``None``. add_firstsamples : bool Defaults to ``False``. MNE's objects store the value of a delay between the start of the system and the start of the recording (see https://mne.tools/stable/generated/mne.io.Raw.html#mne.io.Raw.first_samp). Taking this into account can be useful when extracting channels from the Raw object to detect events indices that are passed back to MNE again. When ``add_firstsamples`` is set to ``True``, the offset will be explicitly added at the beginning of the signal and filled with NaNs. If ``add_firstsamples`` is a float or an integer, the offset will filled with these values instead. If it is set to ``backfill``, will prepend with the first real value. Returns ---------- DataFrame A DataFrame or Series containing the channel(s). Example ---------- .. ipython:: python import neurokit2 as nk import mne raw = nk.mne_data("raw") raw_channel = nk.mne_channel_extract(raw, what=["EEG 060", "EEG 055"], name=['060', '055']) eeg_channels = nk.mne_channel_extract(raw, "EEG") """ channels_all = raw.copy().info["ch_names"] # Select category of channels if what in ["EEG", "EOG", "MEG"]: what = [x for x in channels_all if what in x] # Select a single specified channel elif isinstance(what, str): what = [what] # Select a few specified channels elif isinstance(what, list): if not all(x in channels_all for x in what): raise ValueError( "NeuroKit error: mne_channel_extract(): List of channels not found. Please " "check channel names in raw.info['ch_names']. " ) channels, __ = raw.copy().pick_channels(what, ordered=False)[:] if len(what) > 1: channels = pd.DataFrame(channels.T, columns=what) if name is not None: channels.columns = name else: channels = pd.Series(channels[0]) channels.what = what[0] if name is not None: channels = channels.rename(name) # Add first_samp if isinstance(add_firstsamples, bool) and add_firstsamples is True: # Fill with na add_firstsamples = np.nan if isinstance(add_firstsamples, str): # Back fill add_firstsamples = channels.iloc[0] if isinstance(channels, pd.DataFrame): add_firstsamples = dict(add_firstsamples) if add_firstsamples is not False: if isinstance(channels, pd.Series): fill = pd.Series(add_firstsamples, index=range(-raw.first_samp, 0)) channels = pd.concat([fill, channels], axis=0) elif isinstance(channels, pd.DataFrame): fill = pd.DataFrame( add_firstsamples, index=range(-raw.first_samp, 0), columns=channels.columns ) channels = pd.concat([fill, channels], axis=0) return channels
3,844
38.639175
110
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_simulate.py
import numpy as np from ..misc import check_random_state def eeg_simulate(duration=1, length=None, sampling_rate=1000, noise=0.1, random_state=None): """**EEG Signal Simulation** Simulate an artificial EEG signal. This is a crude implementation based on the MNE-Python raw simulation example. Help is needed to improve this function. Parameters ---------- duration : int Desired recording length in seconds. length : int The desired length of the signal (in samples). sampling_rate : int The desired sampling rate (in Hz, i.e., samples/second). noise : float Noise level. random_state : None, int, numpy.random.RandomState or numpy.random.Generator Seed for the random number generator. See for ``misc.check_random_state`` for further information. Examples ---------- .. ipython:: python import neurokit2 as nk eeg = nk.eeg_simulate(duration=3, sampling_rate=500, noise=0.2) @savefig p_eeg_simulate1.png scale=100% _ = nk.signal_psd(eeg, sampling_rate=500, show=True, max_frequency=100) @suppress plt.close() """ # Try loading mne try: import mne import mne.datasets import mne.simulation except ImportError as e: raise ImportError( "The 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e # Seed the random generator for reproducible results rng = check_random_state(random_state) # Generate number of samples automatically if length is unspecified if length is None: length = duration * sampling_rate if duration is None: duration = length / sampling_rate # Get paths to data path = mne.datasets.sample.data_path() / "MEG" / "sample" raw_file = path / "sample_audvis_raw.fif" fwd_file = path / "sample_audvis-meg-eeg-oct-6-fwd.fif" # Load real data as the template raw = mne.io.read_raw_fif(raw_file, preload=True, verbose=False) raw = raw.set_eeg_reference(projection=True, verbose=False) n_dipoles = 4 # number of dipoles to create def data_fun(times, n_dipoles=4): """Generate time-staggered sinusoids at harmonics of 10Hz""" n = 0 # harmonic number n_samp = len(times) window = np.zeros(n_samp) start, stop = [int(ii * float(n_samp) / (2 * n_dipoles)) for ii in (2 * n, 2 * n + 1)] window[start:stop] = 1.0 n += 1 data = 25e-9 * np.sin(2.0 * np.pi * 10.0 * n * times) data *= window return data times = raw.times[: int(raw.info["sfreq"] * 2)] fwd = mne.read_forward_solution(fwd_file, verbose=False) stc = mne.simulation.simulate_sparse_stc( fwd["src"], n_dipoles=n_dipoles, times=times, data_fun=data_fun, random_state=rng, ) # Repeat the source activation multiple times. raw_sim = mne.simulation.simulate_raw(raw.info, [stc] * int(np.ceil(duration / 2)), forward=fwd, verbose=False) cov = mne.make_ad_hoc_cov(raw_sim.info, std=noise / 1000000) raw_sim = mne.simulation.add_noise(raw_sim, cov, iir_filter=[0.2, -0.2, 0.04], verbose=False, random_state=rng) # Resample raw_sim = raw_sim.resample(sampling_rate, verbose=False) # Add artifacts # mne.simulation.add_ecg(raw_sim, verbose=False) # mne.simulation.add_eog(raw_sim, verbose=False) eeg = raw_sim.pick_types(eeg=True, verbose=False).get_data() return eeg[0, 0 : int(duration * sampling_rate)]
3,600
32.654206
115
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_to_df.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd def mne_to_df(eeg): """**Conversion from MNE to dataframes** Convert MNE objects to dataframe or dict of dataframes. Parameters ---------- eeg : Union[mne.io.Raw, mne.Epochs] Raw or Epochs M/EEG data from MNE. See Also -------- .mne_to_dict Returns ---------- DataFrame A DataFrame containing all epochs identifiable by the ``"Epoch"`` column, which time axis is stored in the ``"Time"`` column. Examples --------- **Raw objects** .. ipython:: python import neurokit2 as nk # Download MNE Raw object eeg = nk.mne_data("filt-0-40_raw") nk.mne_to_df(eeg) **Epoch objects** .. ipython:: python # Download MNE Epochs object eeg = nk.mne_data("epochs") nk.mne_to_df(eeg) **Evoked objects** .. ipython:: python # Download MNE Evoked object eeg = nk.mne_data("evoked") nk.mne_to_df(eeg) """ return _mne_convert(eeg, to_what="df") # Dict def mne_to_dict(eeg): """**Convert MNE Raw or Epochs object to a dictionary** Parameters ---------- eeg : Union[mne.io.Raw, mne.Epochs] Raw or Epochs M/EEG data from MNE. See Also -------- mne_to_df Returns ---------- dict A dict containing all epochs identifiable by the 'Epoch' column, which time axis is stored in the 'Time' column. Examples --------- .. ipython:: python import neurokit2 as nk import mne # Raw objects eeg = nk.mne_data("filt-0-40_raw") nk.mne_to_dict(eeg) # Epochs objects eeg = nk.mne_data("epochs") nk.mne_to_dict(eeg) # Evoked objects eeg = nk.mne_data("evoked") nk.mne_to_dict(eeg) """ return _mne_convert(eeg, to_what="dict") # ============================================================================= # Main function # ============================================================================= def _mne_convert(eeg, to_what="df"): # Try loading mne try: import mne except ImportError as e: raise ImportError( "NeuroKit error: eeg_add_channel(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) from e old_verbosity_level = mne.set_log_level(verbose="WARNING", return_old_level=True) # If raw object if isinstance(eeg, (mne.io.Raw, mne.io.RawArray)): data = eeg.to_data_frame(time_format=None) data = data.rename(columns={"time": "Time"}) if to_what == "dict": data = data.to_dict(orient="list") # If epoch object elif isinstance(eeg, mne.Epochs): data = eeg.to_data_frame(time_format=None) data = data.rename(columns={"time": "Time", "condition": "Condition", "epoch": "Epoch"}) if to_what == "dict": out = {} for epoch in data["Epoch"].unique(): out[epoch] = data[data["Epoch"] == epoch] data = out # If dataframe, skip and return elif isinstance(eeg, pd.DataFrame): data = eeg if to_what == "dict": data = data.to_dict(orient="list") # If dataframe, skip and return elif isinstance(eeg, np.ndarray): data = pd.DataFrame(eeg) if to_what == "dict": data = data.to_dict(orient="list") # it might be an evoked object else: dfs = [] for i, evoked in enumerate(eeg): data = evoked.to_data_frame(time_format=None) data = data.rename(columns={"time": "Time"}) data.insert(1, "Condition", evoked.comment) dfs.append(data) data = pd.concat(dfs, axis=0) if to_what == "dict": out = {} for condition in data["Condition"].unique(): out[condition] = data[data["Condition"] == condition] data = out mne.set_log_level(old_verbosity_level) return data
4,091
24.416149
105
py
NeuroKit
NeuroKit-master/neurokit2/eeg/mne_channel_add.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd def mne_channel_add( raw, channel, channel_type=None, channel_name=None, sync_index_raw=0, sync_index_channel=0 ): """**Add channel as array to MNE** Add a channel to a mne's Raw m/eeg file. It will basically synchronize the channel to the eeg data following a particular index and add it. Parameters ---------- raw : mne.io.Raw Raw EEG data from MNE. channel : list or array The signal to be added. channel_type : str Channel type. Currently supported fields are 'ecg', 'bio', 'stim', 'eog', 'misc', 'seeg', 'ecog', 'mag', 'eeg', 'ref_meg', 'grad', 'emg', 'hbr' or 'hbo'. channel_name : str Desired channel name. sync_index_raw : int or list An index (e.g., the onset of the same event marked in the same signal), in the raw data, by which to align the two inputs. This can be used in case the EEG data and the channel to add do not have the same onsets and must be aligned through some common event. sync_index_channel : int or list An index (e.g., the onset of the same event marked in the same signal), in the channel to add, by which to align the two inputs. This can be used in case the EEG data and the channel to add do not have the same onsets and must be aligned through some common event. Returns ---------- mne.io.Raw Raw data in FIF format. Example ---------- .. ipython:: python import neurokit2 as nk import mne raw = nk.mne_data("filt-0-40_raw") ecg = nk.ecg_simulate(length=50000) # Let the 42nd sample point in the EEG signal correspond to the 333rd point in the ECG event_index_in_eeg = 42 event_index_in_ecg = 333 raw = nk.mne_channel_add(raw, ecg, sync_index_raw=event_index_in_eeg, sync_index_channel=event_index_in_ecg, channel_type="ecg") """ # Try loading mne try: import mne except ImportError: raise ImportError( "NeuroKit error: eeg_channel_add(): the 'mne' module is required for this function to run. ", "Please install it first (`pip install mne`).", ) if channel_name is None: if isinstance(channel, pd.Series): if channel.name is not None: channel_name = channel.name else: channel_name = "Added_Channel" else: channel_name = "Added_Channel" # Compute the distance between the two signals diff = sync_index_channel - sync_index_raw # Pre-empt the channel with nans if shorter or crop if longer if diff > 0: channel = list(channel)[diff : len(channel)] channel = channel + [np.nan] * diff if diff < 0: channel = [np.nan] * abs(diff) + list(channel) # Extend the channel with nans if shorter or crop if longer if len(channel) < len(raw): channel = list(channel) + [np.nan] * (len(raw) - len(channel)) else: # Crop to fit the raw data channel = list(channel)[0 : len(raw)] old_verbosity_level = mne.set_log_level(verbose="WARNING", return_old_level=True) # Create RawArray info = mne.create_info([channel_name], raw.info["sfreq"], ch_types=channel_type) channel = mne.io.RawArray([channel], info) # Add channel raw = raw.copy() raw.add_channels([channel], force_update_info=True) # In-place # Restore old verbosity level mne.set_log_level(old_verbosity_level) return raw
3,693
32.889908
105
py
NeuroKit
NeuroKit-master/neurokit2/eeg/eeg_power.py
# -*- coding: utf-8 -*- import pandas as pd from ..signal import signal_power from .utils import _sanitize_eeg def eeg_power( eeg, sampling_rate=None, frequency_band=["Gamma", "Beta", "Alpha", "Theta", "Delta"], **kwargs ): """**EEG Power in Different Frequency Bands** See our `walkthrough <https://neuropsychology.github.io/NeuroKit/examples/eeg_power/eeg_power.html>`_ for details. * **Gamma** (30-80 Hz) * **Beta** (13-30 Hz) * **Beta 1** (13-16 Hz) * **Beta 2** (16-20 Hz) * **Beta 3** (20-30 Hz) * **SMR** (13-15 Hz) * **Alpha** (8-13 Hz) * **Mu** (9-11 Hz) * **Theta** (4-8 Hz) * **Delta** (1-4 Hz) Parameters ---------- eeg : array An array (channels, times) of M/EEG data or a Raw or Epochs object from MNE. sampling_rate : int The sampling frequency of the signal (in Hz, i.e., samples/second). Only necessary if smoothing is requested. frequency_band : list A list of frequency bands (or tuples of frequencies). **kwargs Other arguments to be passed to ``nk.signal_power()``. Returns ------- pd.DataFrame The power in different frequency bands for each channel. Examples --------- .. ipython:: python import neurokit2 as nk # Raw objects eeg = nk.mne_data("raw") by_channel = nk.eeg_power(eeg) by_channel.head() average = by_channel.mean(numeric_only=True, axis=0) average["Gamma"] References ---------- - Lehmann, D., & Skrandies, W. (1980). Reference-free identification of components of checkerboard-evoked multichannel potential fields. Electroencephalography and clinical neurophysiology, 48(6), 609-621. """ # Sanitize names and values frequency_band, band_names = _eeg_power_sanitize(frequency_band) # Sanitize input eeg, sampling_rate, _ = _sanitize_eeg(eeg, sampling_rate=sampling_rate) # Iterate through channels data = [] for channel in eeg.columns: rez = signal_power( eeg[channel].values, sampling_rate=sampling_rate, frequency_band=frequency_band, **kwargs, ) data.append(rez) data = pd.concat(data, axis=0) data.columns = band_names data.insert(0, "Channel", eeg.columns) data.reset_index(drop=True, inplace=True) return data # ============================================================================= # Utilities # ============================================================================= def _eeg_power_sanitize(frequency_band=["Gamma", "Beta", "Alpha", "Theta", "Delta"]): band_names = frequency_band.copy() # This will used for the names for i, f in enumerate(frequency_band): if isinstance(f, str): f_name = f.lower() if f_name == "gamma": frequency_band[i] = (30, 80) elif f_name == "beta": frequency_band[i] = (13, 30) elif f_name == "beta1": frequency_band[i] = (13, 16) elif f_name == "beta2": frequency_band[i] = (16, 20) elif f_name == "beta3": frequency_band[i] = (20, 30) elif f_name == "smr": frequency_band[i] = (13, 15) elif f_name == "alpha": frequency_band[i] = (8, 13) elif f_name == "mu": frequency_band[i] = (9, 11) elif f_name == "theta": frequency_band[i] = (4, 8) elif f_name == "delta": frequency_band[i] = (1, 4) else: raise ValueError(f"Unknown frequency band: '{f_name}'") elif isinstance(f, tuple): band_names[i] = f"Hz_{f[0]}_{f[1]}" else: raise ValueError("'frequency_band' must be a list of tuples (or strings).") return frequency_band, band_names
3,963
30.460317
109
py
NeuroKit
NeuroKit-master/neurokit2/benchmark/benchmark_ecg.py
# -*- coding: utf-8 -*- import datetime import numpy as np import pandas as pd from ..signal import signal_period def benchmark_ecg_preprocessing(function, ecg, rpeaks=None, sampling_rate=1000): """**Benchmark ECG preprocessing pipelines** Parameters ---------- function : function Must be a Python function which first argument is the ECG signal and which has a ``sampling_rate`` argument. ecg : pd.DataFrame or str The path to a folder where you have an `ECGs.csv` file or directly its loaded DataFrame. Such file can be obtained by running THIS SCRIPT (TO COMPLETE). rpeaks : pd.DataFrame or str The path to a folder where you have an `Rpeaks.csv` fils or directly its loaded DataFrame. Such file can be obtained by running THIS SCRIPT (TO COMPLETE). sampling_rate : int The sampling frequency of ``ecg_signal`` (in Hz, i.e., samples/second). Only used if ``ecgs`` and ``rpeaks`` are single vectors. Returns -------- pd.DataFrame A DataFrame containing the results of the benchmarking Examples -------- .. ipython:: python import neurokit2 as nk # Define a preprocessing routine def function(ecg, sampling_rate): signal, info = nk.ecg_peaks(ecg, method="engzeemod2012", sampling_rate=sampling_rate) return info["ECG_R_Peaks"] # Synthetic example ecg = nk.ecg_simulate(duration=20, sampling_rate=200) true_rpeaks = nk.ecg_peaks(ecg, sampling_rate=200)[1]["ECG_R_Peaks"] nk.benchmark_ecg_preprocessing(function, ecg, true_rpeaks, sampling_rate=200) # Example using database (commented-out) # nk.benchmark_ecg_preprocessing(function, r"path/to/GUDB_database") """ # find data if rpeaks is None: rpeaks = ecg if isinstance(ecg, str): ecg = pd.read_csv(ecg + "/ECGs.csv") if isinstance(rpeaks, str): rpeaks = pd.read_csv(rpeaks + "/Rpeaks.csv") if isinstance(ecg, pd.DataFrame): results = _benchmark_ecg_preprocessing_databases(function, ecg, rpeaks) else: results = _benchmark_ecg_preprocessing(function, ecg, rpeaks, sampling_rate=sampling_rate) return results # ============================================================================= # Utils # ============================================================================= def _benchmark_ecg_preprocessing_databases(function, ecgs, rpeaks): """A wrapper over _benchmark_ecg_preprocessing when the input is a database.""" # Run algorithms results = [] for participant in ecgs["Participant"].unique(): for database in ecgs[ecgs["Participant"] == participant]["Database"].unique(): # Extract the right slice of data ecg_slice = ecgs[(ecgs["Participant"] == participant) & (ecgs["Database"] == database)] rpeaks_slice = rpeaks[(rpeaks["Participant"] == participant) & (rpeaks["Database"] == database)] sampling_rate = ecg_slice["Sampling_Rate"].unique()[0] # Extract values ecg = ecg_slice["ECG"].values rpeak = rpeaks_slice["Rpeaks"].values # Run benchmark result = _benchmark_ecg_preprocessing(function, ecg, rpeak, sampling_rate) # Add info result["Participant"] = participant result["Database"] = database results.append(result) return pd.concat(results) def _benchmark_ecg_preprocessing(function, ecg, rpeak, sampling_rate=1000): # Apply function t0 = datetime.datetime.now() try: found_rpeaks = function(ecg, sampling_rate=sampling_rate) duration = (datetime.datetime.now() - t0).total_seconds() # In case of failure except Exception as error: # pylint: disable=broad-except return pd.DataFrame( { "Sampling_Rate": [sampling_rate], "Duration": [np.nan], "Score": [np.nan], "Recording_Length": [len(ecg) / sampling_rate / 60], "Error": str(error), } ) # Compare R peaks score, error = benchmark_ecg_compareRpeaks(rpeak, found_rpeaks, sampling_rate=sampling_rate) return pd.DataFrame( { "Sampling_Rate": [sampling_rate], "Duration": [duration], "Score": [score], "Recording_Length": [len(ecg) / sampling_rate / 60], "Error": error, } ) # ============================================================================= # Comparison methods # ============================================================================= def benchmark_ecg_compareRpeaks(true_rpeaks, found_rpeaks, sampling_rate=250): # Failure to find sufficient R-peaks if len(found_rpeaks) <= 3: return np.nan, "R-peaks detected <= 3" length = np.max(np.concatenate([true_rpeaks, found_rpeaks])) true_interpolated = signal_period( true_rpeaks, sampling_rate=sampling_rate, desired_length=length, interpolation_method="linear" ) found_interpolated = signal_period( found_rpeaks, sampling_rate=sampling_rate, desired_length=length, interpolation_method="linear" ) return np.mean(np.abs(found_interpolated - true_interpolated)), "None"
5,347
33.727273
108
py
NeuroKit
NeuroKit-master/neurokit2/benchmark/benchmark_utils.py
from timeit import default_timer as timer from wfdb.processing import compare_annotations def benchmark_record(record, sampling_rate, annotation, tolerance, detector): """**Obtain detector performance for an annotated record** Parameters ---------- record : array The raw physiological record. sampling_rate: int The sampling rate of the record in Hertz. annotation : array The manual extrema annotations. tolerance : int Maximum difference in millisecond that is permitted between the manual annotation and the annotation generated by the detector. detector : function A function that takes a physiological record as first positional argument as well as a ``sampling_rate`` keyword argument. Returns ------- precision : float The detectors precision on the record given the tolerance. sensitivity : float The detectors sensitivity on the record given the tolerance. """ detector_annotation = detector(record, sampling_rate=sampling_rate) comparitor = compare_annotations(detector_annotation, annotation, tolerance) tp = comparitor.tp fp = comparitor.fp fn = comparitor.fn sensitivity = tp / (tp + fn) precision = tp / (tp + fp) return precision, sensitivity def time_record(record, sampling_rate, detector, n_runs): """**Obtain the average run time of a detector on a record over N runs** Parameters ---------- record : array The raw physiological record. sampling_rate : int The sampling rate of the record in Hertz detector : function A function that takes a physiological record as first positional argument as well as a ``sampling_rate`` keyword argument. n_runs : int The number of runs. Returns ------- avg_time The run time of the detector on the record averaged over n_runs. In milliseconds. """ start = timer() for _ in range(n_runs): detector(record, sampling_rate=sampling_rate) end = timer() avg_time = (end - start) / n_runs * 1000 return avg_time
2,161
28.216216
80
py
NeuroKit
NeuroKit-master/neurokit2/benchmark/__init__.py
"""Submodule for NeuroKit.""" from .benchmark_ecg import benchmark_ecg_preprocessing __all__ = [ "benchmark_ecg_preprocessing", ]
137
14.333333
54
py
NeuroKit
NeuroKit-master/neurokit2/data/write_csv.py
import numpy as np def write_csv(data, filename, parts=None, **kwargs): """**Write data to multiple csv files** Split the data into multiple CSV files. You can then re-create them as follows: Parameters ---------- data : list List of dictionaries. filename : str Name of the CSV file (without the extension). parts : int Number of parts to split the data into. Returns ------- None Example -------- Save big file in parts .. ipython:: python import pandas as pd import neurokit2 as nk # Split data into multiple files # nk.write_csv(data, 'C:/Users/.../data', parts=6) Read the files back .. ipython:: python # Iterate through 6-parts and concatenate the pieces # data_all = pd.concat( # [pd.read_csv(f"data_part{i}.csv") for i in range(1, 7)], # axis=0, # ) """ if isinstance(parts, int): # Add column to identify parts data["__Part__"] = np.repeat(range(parts), np.ceil(len(data) / parts))[0 : len(data)] for i, part in data.groupby("__Part__"): part.drop(["__Part__"], axis=1).to_csv(filename + f"_part{i + 1}.csv", **kwargs) else: data.to_csv(filename, **kwargs)
1,287
23.301887
93
py
NeuroKit
NeuroKit-master/neurokit2/data/read_bitalino.py
# -*- coding: utf-8 -*- import json import os from warnings import warn import numpy as np import pandas as pd from ..misc import NeuroKitWarning def read_bitalino(filename): """**Read an OpenSignals file (from BITalino)** Reads and loads a BITalino file into a Pandas DataFrame. The function outputs both the dataframe and the information (such as the sampling rate) retrieved from the OpenSignals file). Parameters ---------- filename : str Filename (with or without the extension) of an OpenSignals file (e.g., ``"data.txt"``). Returns ---------- df : DataFrame, dict The BITalino file as a pandas dataframe if one device was read, or a dictionary of pandas dataframes (one dataframe per device) if multiple devices are read. info : dict The metadata information containing the sensors, corresponding channel names, sampling rate, and the events annotation timings if ``events_annotation`` is ``True``. See Also -------- .read_acqknowledge, .signal_resample Examples -------- .. ipython:: python import neurokit2 as nk # data, info = nk.read_bitalino("data.txt") # sampling_rate = info["sampling_rate"] """ # Read metadata # ------------------------------------------------------------------------- with open(filename, "r") as f: lines = f.readlines() if "OpenSignals" not in lines[0]: raise ValueError("Text file is not in OpenSignals format.") metadata = json.loads(lines[1][1:]) # read second line + skip '#' # Remove ":" metadata = {k.replace(":", ""): metadata[k] for k in metadata.keys()} # Try find events annotations # ------------------------------------------------------------------------- annotations = _read_bitalino_annotations(filename) if annotations is not None: for k in annotations.keys(): if k in metadata.keys(): metadata[k]["Annotations"] = annotations[k] else: warn( f"Device {k} not found in metadata ({metadata.keys()})." + " Something might be wrong.", category=NeuroKitWarning, ) # Read data # ------------------------------------------------------------------------- data = {k: None for k in metadata.keys()} raw = pd.read_csv(filename, sep="\t", header=None, comment="#") # Read file for each device for i, k in enumerate(metadata.keys()): # Select right columns ch = np.array(metadata[k]["column"]) data[k] = raw.iloc[:, i * len(ch) : (i + 1) * len(ch)] for j, s in enumerate(metadata[k]["label"]): ch[ch == s] = metadata[k]["sensor"][j] data[k].columns = ch # Add annotations if "Annotations" in metadata[k].keys(): sr = metadata[k]["sampling rate"] data[k]["Events"] = 0 annot = metadata[k]["Annotations"] annot = annot[annot["CHANNEL"].isin(metadata[k]["label"])] annot = annot.drop_duplicates(["START", "END"]) for _, row in annot.iterrows(): data[k]["Events"][int(row["START"] * sr) : int(row["END"] * sr) + 1] = 1 # Format metadata names metadata[k] = {x.replace(" ", "_"): v for x, v in metadata[k].items()} # If only one device is detected, extract from dict if i == 0: data = data[k] metadata = metadata[k] return data, metadata # ============================================================================= # Internals # ============================================================================= def _read_bitalino_annotations(filename): """Read events that are annotated during BITalino signal acquisition. Returns a dictionary containing the start and stop times (in seconds) in each channel detected per unique event (label) within each device.""" file = filename.replace(".txt", "_EventsAnnotation.txt") if os.path.isfile(file) is False: return None with open(file, "r") as f: lines = f.readlines() if "OpenSignals" not in lines[0]: raise ValueError("Text file is not in OpenSignals format.") metadata = json.loads(lines[1][1:]) # read second line + skip '#' data = pd.read_csv(file, sep="\t", header=None, comment="#") data = data.dropna(axis=1, how="all") data.columns = metadata["columns"]["labels"] return {k: data[data["MAC"] == k] for k in data["MAC"].unique()}
4,610
34.469231
98
py
NeuroKit
NeuroKit-master/neurokit2/data/read_video.py
# -*- coding: utf-8 -*- import os import numpy as np def read_video(filename="video.mp4"): """**Reads a video file into an array** Reads a video file (e.g., .mp4) into a numpy array of shape. This function requires OpenCV to be installed via the ``opencv-python`` package. Parameters ---------- filename : str The path of a video file. Returns ------- array numpy array of shape (frame, RGB-channel, height, width). int Sampling rate in frames per second. Examples -------- .. ipython:: python import neurokit2 as nk # video, sampling_rate = nk.read_video("video.mp4") """ # Try loading cv2 try: import cv2 except ImportError: raise ImportError( "The 'cv2' module is required for this function to run. ", "Please install it first (`pip install opencv-python`).", ) # Check if file exists assert os.path.isfile(filename) is True, f"No file found with the specified name ({filename})." capture = cv2.VideoCapture(filename) sampling_rate = int(capture.get(cv2.CAP_PROP_FPS)) frames = [] while capture.isOpened(): success, frame = capture.read() # By default frame is BGR if not success: break frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) # Convert to RGB capture.release() # Swap axes to be (frame, RGB-channel, height, width) return np.array(frames).swapaxes(3, 1).swapaxes(3, 2), sampling_rate
1,541
25.586207
99
py
NeuroKit
NeuroKit-master/neurokit2/data/data.py
# -*- coding: utf-8 -*- import json import os import pickle import urllib import pandas as pd from sklearn import datasets as sklearn_datasets def data(dataset="bio_eventrelated_100hz"): """**NeuroKit Datasets** NeuroKit includes datasets that can be used for testing. These datasets are not downloaded automatically with the package (to avoid increasing its weight), but can be downloaded via the ``nk.data()`` function (note that an internet connection is necessary). See the examples below. **Signals**: The following signals (that will return an array) are available: * **ecg_1000hz**: Returns a vector containing ECG signal (``sampling_rate=1000``). * **ecg_3000hz**: Returns a vector containing ECG signal (``sampling_rate=3000``). * **rsp_1000hz**: Returns a vector containing RSP signal (``sampling_rate=1000``). * **eeg_150hz**: Returns a vector containing EEG signal (``sampling_rate=150``). * **eog_100hz**: Returns a vector containing vEOG signal (``sampling_rate=100``). **DataFrames**: The following datasets (that will return a ``pd.DataFrame``) are available: * **iris**: Convenient access to the Iris dataset in a DataFrame, exactly how it is in R. * **eogs_200hz**: Returns a DataFrame with ``hEOG``, ``vEOG``. * Single subject * Visual and horizontal electrooculagraphy * ``sampling_rate=200`` * **bio_resting_5min_100hz**: Returns a DataFrame with ``ECG``, ``PPG``, ``RSP``. * Single subject * Resting-state of 5 min (pre-cropped, with some ECG noise towards the end) * ``sampling_rate=100`` * **bio_resting_8min_100hz**: Returns a DataFrame with ``ECG``, ``RSP``, ``EDA``, ``PhotoSensor``. * Single subject * Resting-state of 8 min when the photosensor is low (need to crop the data) * ``sampling_rate=100`` * **bio_resting_8min_200hz**: Returns a dictionary with four subjects (``S01``, ``S02``, ``S03``, ``S04``). * Resting-state recordings * 8 min (``sampling_rate=200``) * Each subject is DataFrame with ``ECG``, ``RSP`, ``PhotoSensor``, ``Participant`` * **bio_eventrelated_100hz**: Returns a DataFrame with ``ECG``, ``EDA``, ``Photosensor``, ``RSP``. * Single subject * Event-related recording of a participant watching 4 images for 3 seconds (the condition order was: ``["Negative", "Neutral", "Neutral", "Negative"]``) * ``sampling_rate=100`` * **eeg_1min_200hz**: Returns an MNE raw object containing 1 min of EEG data (from the MNE-sample dataset). Parameters ---------- dataset : str The name of the dataset. Returns ------- DataFrame The data. Examples --------- **Single signals and vectors** .. ipython:: python import neurokit2 as nk ecg = nk.data(dataset="ecg_1000hz") @savefig p_datasets1.png scale=100% nk.signal_plot(ecg[0:10000], sampling_rate=1000) @suppress plt.close() .. ipython:: python rsp = nk.data(dataset="rsp_1000hz") @savefig p_datasets2.png scale=100% nk.signal_plot(rsp[0:20000], sampling_rate=1000) @suppress plt.close() .. ipython:: python eeg = nk.data("eeg_150hz") @savefig p_data3.png scale=100% nk.signal_plot(eeg, sampling_rate=150) @suppress plt.close() .. ipython:: python eog = nk.data("eog_100hz") @savefig p_data4.png scale=100% nk.signal_plot(eog[0:2000], sampling_rate=100) @suppress plt.close() **DataFrames** .. ipython:: python data = nk.data("iris") data.head() .. ipython:: python data = nk.data(dataset="eogs_200hz") @savefig p_datasets5.png scale=100% nk.signal_plot(data[0:4000], standardize=True, sampling_rate=200) @suppress plt.close() .. ipython:: python data = nk.data(dataset="bio_resting_5min_100hz") @savefig p_datasets6.png scale=100% nk.standardize(data).plot() @suppress plt.close() .. ipython:: python data = nk.data(dataset="bio_resting_8min_100hz") @savefig p_datasets7.png scale=100% nk.standardize(data).plot() @suppress plt.close() .. ipython:: python data = nk.data("bio_resting_8min_200hz") data.keys() data["S01"].head() .. ipython:: python data = nk.data("bio_eventrelated_100hz") @savefig p_data8.png scale=100% nk.standardize(data).plot() @suppress plt.close() .. ipython:: python raw = nk.data("eeg_1min_200hz") @savefig p_data9.png scale=100% nk.signal_plot(raw.get_data()[0:3, 0:2000], sampling_rate=200) @suppress plt.close() """ # TODO: one could further improve this function with like # selectors 'ecg=True, eda=True, restingstate=True' that would # find the most appropriate dataset dataset = dataset.lower() # TODO: change this path back to "master" path = "https://raw.githubusercontent.com/neuropsychology/NeuroKit/dev/data/" # Signals as vectors ======================= if dataset in ["eeg", "eeg_150hz", "eeg.txt"]: return pd.read_csv(path + "eeg.txt").values[:, 0] if dataset in ["rsp", "rsp_1000hz", "rsp_1000hz.txt"]: return pd.read_csv(path + "rsp_1000hz.txt", header=None).values[:, 0] if dataset in ["ecg", "ecg_1000hz", "ecg_1000hz.csv"]: return pd.read_csv(path + "ecg_1000hz.csv")["ECG"].values if dataset in ["ecg_3000hz", "ecg_3000hz.csv"]: return pd.read_csv(path + "ecg_1000hz.csv")["ECG"].values if dataset in ["eog", "veog", "eog_100hz", "eog_100hz.csv"]: return pd.read_csv(path + "eog_100hz.csv")["vEOG"].values # Dataframes =============================== if dataset == "iris": info = sklearn_datasets.load_iris() data = pd.DataFrame( info.data, columns=["Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width"] ) data["Species"] = info.target_names[info.target] return data if dataset in ["eogs", "eogs_200hz", "eog_200hz", "eog_200hz.csv"]: return pd.read_csv(path + "eog_200hz.csv") # Add extension if dataset in ["bio_resting_8min_200hz"]: dataset += ".json" # Specific case for json file if dataset.endswith(".json"): if "https" not in dataset: data = pd.read_json(path + dataset, orient="index") else: data = pd.read_json(dataset, orient="index") df = {} for participant, row in data.iterrows(): for _, data_string in row.items(): data_list = json.loads(data_string) data_pd = pd.DataFrame(data_list) df[participant] = data_pd return df # TODO: Add more EEG (fif and edf datasets) if dataset in ["eeg_1min_200hz"]: return pickle.load( urllib.request.urlopen( "https://github.com/neuropsychology/NeuroKit/blob/dev/data/eeg_1min_200hz.pickle?raw=true" ) ) # General case file, ext = os.path.splitext(dataset) # pylint: disable=unused-variable if ext == "": df = pd.read_csv(path + dataset + ".csv") else: if "https" not in dataset: df = pd.read_csv(path + dataset) else: df = pd.read_csv(dataset) return df
7,455
29.432653
106
py
NeuroKit
NeuroKit-master/neurokit2/data/read_acqknowledge.py
# -*- coding: utf-8 -*- import os import numpy as np import pandas as pd from ..signal import signal_resample def read_acqknowledge( filename, sampling_rate="max", resample_method="interpolation", impute_missing=True ): """**Read and format a BIOPAC's AcqKnowledge file into a pandas' dataframe** The function outputs both the dataframe and the sampling rate (retrieved from the AcqKnowledge file). Parameters ---------- filename : str Filename (with or without the extension) of a BIOPAC's AcqKnowledge file (e.g., ``"data.acq"``). sampling_rate : int Sampling rate (in Hz, i.e., samples/second). Since an AcqKnowledge file can contain signals recorded at different rates, harmonization is necessary in order to convert it to a DataFrame. Thus, if `sampling_rate` is set to ``max`` (default), will keep the maximum recorded sampling rate and upsample the channels with lower rate if necessary (using the :func:`.signal_resample()` function). If the sampling rate is set to a given value, will resample the signals to the desired value. Note that the value of the sampling rate is outputted along with the data. resample_method : str Method of resampling (see :func:`.signal_resample()`). impute_missing : bool Sometimes, due to connections issues, there are lapses in the recorded signal (short periods without signal). If ``impute_missing`` is ``True``, will automatically fill the signal interruptions using padding. Returns ---------- df : DataFrame The AcqKnowledge file as a pandas dataframe. sampling rate: int The sampling rate at which the data is sampled. See Also -------- .signal_resample Example ---------- .. ipython:: python import neurokit2 as nk # data, sampling_rate = nk.read_acqknowledge('file.acq') """ # Try loading bioread try: import bioread except ImportError: raise ImportError( "NeuroKit error: read_acqknowledge(): the 'bioread' module is required", " for this function to run. ", "Please install it first (`pip install bioread`).", ) # Check filename if ".acq" not in filename: filename += ".acq" if os.path.exists(filename) is False: raise ValueError( "NeuroKit error: read_acqknowledge(): couldn't" " find the following file: " + filename ) # Read file file = bioread.read(filename) # Get desired frequency if sampling_rate == "max": freq_list = [] for channel in file.named_channels: freq_list.append(file.named_channels[channel].samples_per_second) sampling_rate = np.max(freq_list) # Loop through channels data = {} for channel in file.named_channels: signal = np.array(file.named_channels[channel].data) # Fill signal interruptions if impute_missing is True and np.isnan(np.sum(signal)): signal = pd.Series(signal).fillna(method="pad").values # Resample if necessary if file.named_channels[channel].samples_per_second != sampling_rate: signal = signal_resample( signal, sampling_rate=file.named_channels[channel].samples_per_second, desired_sampling_rate=sampling_rate, method=resample_method, ) data[channel] = signal # Sanitize lengths lengths = [] for channel in data: lengths += [len(data[channel])] if len(set(lengths)) > 1: # If different lengths length = pd.Series(lengths).mode()[0] # Find most common (target length) for channel in data: if len(data[channel]) > length: data[channel] = data[channel][0:length] if len(data[channel]) < length: data[channel] = np.concatenate( [ data[channel], np.full((length - len(data[channel])), data[channel][-1]), ] ) # Final dataframe df = pd.DataFrame(data) return df, sampling_rate
4,253
32.761905
104
py
NeuroKit
NeuroKit-master/neurokit2/data/__init__.py
"""Submodule for NeuroKit.""" from .data import data from .read_acqknowledge import read_acqknowledge from .read_bitalino import read_bitalino from .read_video import read_video from .write_csv import write_csv __all__ = ["read_acqknowledge", "read_bitalino", "read_video", "data", "write_csv"]
297
28.8
83
py
NeuroKit
NeuroKit-master/neurokit2/stats/cluster_quality.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd import scipy.spatial import sklearn.cluster import sklearn.metrics import sklearn.mixture import sklearn.model_selection from ..misc import check_random_state def cluster_quality(data, clustering, clusters=None, info=None, n_random=10, random_state=None, **kwargs): """**Assess Clustering Quality** Compute quality of the clustering using several metrics. Parameters ---------- data : np.ndarray A matrix array of data (e.g., channels, sample points of M/EEG data) clustering : DataFrame Information about the distance of samples from their respective clusters, generated from :func:`.cluster`. clusters : np.ndarray Coordinates of cluster centers, which has a shape of n_clusters x n_features, generated from :func:`.cluster`. info : dict Information about the number of clusters, the function and model used for clustering, generated from :func:`.cluster`. n_random : int The number of random initializations to cluster random data for calculating the GAP statistic. random_state : None, int, numpy.random.RandomState or numpy.random.Generator Seed for the random number generator. See for ``misc.check_random_state`` for further information. **kwargs Other argument to be passed on, for instance ``GFP`` as ``'sd'`` in microstates. Returns ------- individual : DataFrame Indices of cluster quality scores for each sample point. general : DataFrame Indices of cluster quality scores for all clusters. Examples ---------- .. ipython:: python import neurokit2 as nk # Load the iris dataset data = nk.data("iris").drop("Species", axis=1) # Cluster clustering, clusters, info = nk.cluster(data, method="kmeans", n_clusters=3) # Compute indices of clustering quality individual, general = nk.cluster_quality(data, clustering, clusters, info) general References ---------- * Tibshirani, R., Walther, G., & Hastie, T. (2001). Estimating the number of clusters in a data set via the gap statistic. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 63(2), 411-423. * Mohajer, M., Englmeier, K. H., & Schmid, V. J. (2011). A comparison of Gap statistic definitions with and without logarithm function. arXiv preprint arXiv:1103.4767. """ # Seed the random generator for reproducible results rng = check_random_state(random_state) # Sanity checks if isinstance(clustering, tuple): clustering, clusters, info = clustering if isinstance(data, pd.DataFrame): data = data.values if isinstance(clustering, pd.DataFrame): clustering = clustering["Cluster"].values individual, general = _cluster_quality_sklearn(data, clustering, clusters) # Individual distance from centroid distance = _cluster_quality_distance(data, clusters) distance = {"Clustering_Distance_" + str(i): distance[:, i] for i in range(distance.shape[1])} individual.update(distance) individual = pd.DataFrame(individual) # Variance explained general["Score_VarianceExplained"] = _cluster_quality_variance(data, clusters, clustering) general["Score_GEV"], _ = _cluster_quality_gev(data, clusters, clustering, **kwargs) general["Score_CrossValidation"] = _cluster_quality_crossvalidation(data, clusters, clustering) # Dispersion general["Dispersion"] = _cluster_quality_dispersion(data, clustering, **kwargs) # Gap statistic general.update(_cluster_quality_gap(data, clusters, clustering, info, n_random=n_random, rng=rng)) # Mixture models if "sklearn_model" in info: if isinstance(info["sklearn_model"], sklearn.mixture.GaussianMixture): general["Score_AIC"] = info["sklearn_model"].aic(data) general["Score_BIC"] = info["sklearn_model"].bic(data) general["Score_LogLikelihood"] = info["sklearn_model"].score(data) sklearn.model_selection.cross_val_score(info["sklearn_model"], data, cv=10) general = pd.DataFrame.from_dict(general, orient="index").T return individual, general # ============================================================================= # Utils # ============================================================================= def _cluster_quality_sklearn(data, clustering, clusters): """Metrics from sklearn""" n_clusters = len(clusters) # Individual scores individual = {} if n_clusters == 1: individual["Clustering_Silhouette"] = np.full(len(clustering), np.nan) else: individual["Clustering_Silhouette"] = sklearn.metrics.silhouette_samples(data, clustering) # General clustering quality scores general = {"n_Clusters": n_clusters} if n_clusters == 1: general["Score_Silhouette"] = np.nan general["Score_Calinski"] = np.nan general["Score_Bouldin"] = np.nan else: general["Score_Silhouette"] = sklearn.metrics.silhouette_score(data, clustering) general["Score_Calinski"] = sklearn.metrics.calinski_harabasz_score(data, clustering) general["Score_Bouldin"] = sklearn.metrics.davies_bouldin_score(data, clustering) return individual, general def _cluster_quality_distance(data, clusters, to_dataframe=False): """Distance between samples and clusters""" distance = scipy.spatial.distance.cdist(data, clusters) if to_dataframe is True: distance = pd.DataFrame(distance).add_prefix("Distance_") return distance def _cluster_quality_sumsquares(data, clusters, clustering): """Sumsquares of the distance of each data point to its respective cluster""" distance = _cluster_quality_distance(data, clusters) min_distance = [] # distance from each sample to the centroid of its cluster for idx in range(len(data)): cluster_identity = clustering[idx] min_distance.append(distance[idx, cluster_identity]) min_distance_squared = [i**2 for i in min_distance] return np.sum(min_distance_squared) def _cluster_quality_dispersion(data, clustering, n_clusters=4): """Sumsquares of the distances between samples within each clusters. An error measure for a n_clusters cluster where the lower the better. Can be used to compare and find the optimal number of clusters. """ dispersion_state = np.full(n_clusters, np.nan) for i in range(n_clusters): idx = clustering == i data_state = data[idx, :] state_size = len(data_state) # number of samples in this cluster if state_size > 0: # pair-wise distance between members of the same cluster distance = scipy.spatial.distance.cdist(data_state, data_state) # sumsquares of distances dispersion_state[i] = 0.5 * np.nansum(distance**2) / state_size else: dispersion_state[i] = np.nan dispersion = np.nansum(dispersion_state) return dispersion def _cluster_quality_variance(data, clusters, clustering): """Variance explained by clustering""" sum_squares_within = _cluster_quality_sumsquares(data, clusters, clustering) sum_squares_total = np.sum(scipy.spatial.distance.pdist(data) ** 2) / data.shape[0] return (sum_squares_total - sum_squares_within) / sum_squares_total def _cluster_quality_gap(data, clusters, clustering, info, n_random=10, rng=None): """GAP statistic and modified GAP statistic by Mohajer (2011). The GAP statistic compares the total within intra-cluster variation for different values of k with their expected values under null reference distribution of the data. """ dispersion = _cluster_quality_sumsquares(data, clusters, clustering) mins, maxs = np.min(data, axis=0), np.max(data, axis=0) dispersion_random = np.full(n_random, np.nan) for i in range(n_random): # Random data random_data = rng.uniform(size=data.shape) # Rescale random m = (maxs - mins) / (np.max(random_data, axis=0) - np.min(random_data, axis=0)) b = mins - (m * np.min(random_data, axis=0)) random_data = np.array(m) * random_data + np.array(b) # Cluster random _, random_clusters, info = info["clustering_function"](random_data) random_activation = random_clusters.dot(random_data.T) random_clustering = np.argmax(np.abs(random_activation), axis=0) dispersion_random[i] = _cluster_quality_sumsquares( random_data, random_clusters, random_clustering ) # Compute GAP gap = np.mean(np.log(dispersion_random)) - np.log(dispersion) # Compute standard deviation sd_k = np.sqrt(np.mean((np.log(dispersion_random) - np.mean(np.log(dispersion_random))) ** 2.0)) s_k = np.sqrt(1.0 + 1.0 / n_random) * sd_k # Calculate Gap* statistic by Mohajer (2011) gap_star = np.mean(dispersion_random) - dispersion sd_k_star = np.sqrt(np.mean((dispersion_random - dispersion) ** 2.0)) s_k_star = np.sqrt(1.0 + 1.0 / n_random) * sd_k_star out = { "Score_GAP": gap, "Score_GAPmod": gap_star, "Score_GAP_sk": s_k, "Score_GAPmod_sk": s_k_star, } return out def _cluster_quality_crossvalidation(data, clusters, clustering): """Cross-validation index The original code by https://github.com/Frederic-vW/eeg_microstates/blob/master/eeg_microstates.py#L600 leads to an error when the denominator is 0. """ n_rows, n_cols = data.shape # n_sample, n_channel var = np.nansum(data**2) - np.nansum(np.nansum(clusters[clustering, :] * data, axis=1) ** 2) var /= n_rows * (n_cols - 1) denominator = (n_cols - len(clusters) - 1) ** 2 if np.abs(denominator) > 0: cv = var * (n_cols - 1) ** 2 / denominator else: # warnings.warn( # "Number of columns in data (" + str(n_cols) + ") is smaller " # "than the number of cluster (" + str(len(clusters)) + ") plus 1. " # "Returning the residual noise instead." # ) cv = np.nan # cv = var * (n_cols - 1)**2 / len(clusters) return cv # def _cluster_quality_gev(data, clusters, clustering, sd=None): # if sd is None: # sd = np.std(data, axis=1) # # # Normalize row-wise (across columns) # clusters /= np.sqrt(np.sum(clusters**2, axis=1, keepdims=True)) # activation = np.dot(data, clusters.T) # activation /= (data.shape[1] * np.outer(sd, np.std(clusters, axis=1))) # # gev = np.zeros(len(clusters)) # for k in range(len(clusters)): # idx = (clustering == k) # gev[k] = np.sum(sd[idx]**2 * activation[idx, k]**2) # gev_total = np.sum(gev) / np.sum(sd ** 2) # return gev_total def _cluster_quality_gev(data, clusters, clustering, sd=None, n_clusters=4): """Global Variance Explained (GEV)""" if sd is None: sd = np.nanstd(data, axis=1) map_corr = _correlate_vectors(data.T, clusters[clustering].T) gev_all = np.zeros(n_clusters) for state in range(n_clusters): idx = clustering == state gev_all[state] = np.nansum((sd[idx] * map_corr[idx]) ** 2) / np.nansum(sd**2) gev = np.nansum(gev_all) # gev = np.sum((sd * map_corr) ** 2) / np.sum(sd**2) return gev, gev_all def _correlate_vectors(A, B, axis=0): """Compute pairwise correlation of multiple pairs of vectors. Fast way to compute correlation of multiple pairs of vectors without computing all pairs as would with corr(A,B). Borrowed from Oli at Stack overflow. Note the resulting coefficients vary slightly from the ones obtained from corr due differences in the order of the calculations. (Differences are of a magnitude of 1e-9 to 1e-17 depending of the tested data). Parameters ---------- A : array The first collection of vectors of shape (n, m) B : array The second collection of vectors of shape (n, m) axis : int The axis that contains the elements of each vector. Defaults to 0. Returns ------- corr : array For each pair of vectors, the correlation between them with shape (m, ) """ An = A - np.nanmean(A, axis=axis) Bn = B - np.nanmean(B, axis=axis) An /= np.linalg.norm(An, axis=axis) Bn /= np.linalg.norm(Bn, axis=axis) return np.nansum(An * Bn, axis=axis)
12,509
36.794562
107
py
NeuroKit
NeuroKit-master/neurokit2/stats/density_bandwidth.py
# -*- coding: utf-8 -*- import warnings import numpy as np import scipy.stats def density_bandwidth(x, method="KernSmooth", resolution=401): """**Bandwidth Selection for Density Estimation** Bandwidth selector for :func:`.density` estimation. See ``bw_method`` argument in :func:`.scipy.stats.gaussian_kde`. The ``"KernSmooth"`` method is adapted from the ``dpik()`` function from the *KernSmooth* R package. In this case, it estimates the optimal AMISE bandwidth using the direct plug-in method with 2 levels for the Parzen-Rosenblatt estimator with Gaussian kernel. Parameters ----------- x : Union[list, np.array, pd.Series] A vector of values. method : float The bandwidth of the kernel. The larger the values, the smoother the estimation. Can be a number, or ``"scott"`` or ``"silverman"`` (see ``bw_method`` argument in :func:`.scipy.stats.gaussian_kde`), or ``"KernSmooth"``. resolution : int Only when ``method="KernSmooth"``. The number of equally-spaced points over which binning is performed to obtain kernel functional approximation (see ``gridsize`` argument in ``KernSmooth::dpik()``). Returns ------- float Bandwidth value. See Also -------- density Examples -------- .. ipython:: python import neurokit2 as nk x = np.random.normal(0, 1, size=100) bw = nk.density_bandwidth(x) bw nk.density_bandwidth(x, method="scott") nk.density_bandwidth(x, method=1) @savefig p_density_bandwidth.png scale=100% x, y = nk.density(signal, bandwidth=bw, show=True) @suppress plt.close() References ---------- * Jones, W. M. (1995). Kernel Smoothing, Chapman & Hall. """ if isinstance(method, str): method = method.lower() if isinstance(method, (float, int)) or method != "kernsmooth": return scipy.stats.gaussian_kde(x, bw_method=method).factor n = len(x) stdev = np.nanstd(x, ddof=1) iqr = np.diff(np.percentile(x, [25, 75]))[0] / 1.349 scalest = min(stdev, iqr) data_scaled = (x - np.nanmean(x)) / scalest min_scaled = np.nanmin(data_scaled) max_scaled = np.nanmax(data_scaled) gcounts = _density_linearbinning( x=data_scaled, gpoints=np.linspace(min_scaled, max_scaled, resolution), truncate=True, ) alpha = (2 * np.sqrt(2) ** 9 / (7 * n)) ** (1 / 9) psi6hat = _density_bkfe(gcounts, 6, alpha, min_scaled, max_scaled) alpha = (-3 * np.sqrt(2 / np.pi) / (psi6hat * n)) ** (1 / 7) psi4hat = _density_bkfe(gcounts, 4, alpha, min_scaled, max_scaled) delta_0 = 1 / ((4 * np.pi) ** (1 / 10)) output = scalest * delta_0 * (1 / (psi4hat * n)) ** (1 / 5) return output def _density_linearbinning(x, gpoints, truncate=True): """ Linear binning. Adapted from KernSmooth R package. """ n = len(x) M = gpoints.shape[0] a = gpoints[0] b = gpoints[-1] # initialization of gcounts: gcounts = np.zeros(M) Delta = (b - a) / (M - 1) for i in range(n): lxi = ((x[i] - a) / Delta) + 1 li = int(lxi) rem = lxi - li if (li >= 1) and (li < M): gcounts[li - 1] = gcounts[li - 1] + 1 - rem gcounts[li] = gcounts[li] + rem elif (li < 1) and (truncate is False): gcounts[0] = gcounts[0] + 1 elif (li >= M) and (truncate is False): gcounts[M - 1] = gcounts[M - 1] + 1 return gcounts def _density_bkfe(gcounts, drv, h, a, b): """ 'bkfe' function adapted from KernSmooth R package. """ resol = len(gcounts) # Set the sample size and bin width n = np.nansum(gcounts) delta = (b - a) / (resol - 1) # Obtain kernel weights tau = drv + 4 L = min(int(tau * h / delta), resol) if L == 0: warnings.warn( "WARNING : Binning grid too coarse for current (small) bandwidth: consider increasing 'resolution'" ) lvec = np.arange(L + 1) arg = lvec * delta / h dnorm = np.exp(-np.square(arg) / 2) / np.sqrt(2 * np.pi) kappam = dnorm / h ** (drv + 1) hmold0 = 1 hmold1 = arg hmnew = 1 if drv >= 2: for i in np.arange(2, drv + 1): hmnew = arg * hmold1 - (i - 1) * hmold0 hmold0 = hmold1 # Compute mth degree Hermite polynomial hmold1 = hmnew # by recurrence. kappam = hmnew * kappam # Now combine weights and counts to obtain estimate P = 2 ** (int(np.log(resol + L + 1) / np.log(2)) + 1) kappam = np.concatenate((kappam, np.zeros(P - 2 * L - 1), kappam[1:][::-1]), axis=0) Gcounts = np.concatenate((gcounts, np.zeros(P - resol)), axis=0) kappam = np.fft.fft(kappam) Gcounts = np.fft.fft(Gcounts) gcounter = gcounts * (np.real(np.fft.ifft(kappam * Gcounts)))[0:resol] return np.nansum(gcounter) / n**2
4,949
28.464286
117
py
NeuroKit
NeuroKit-master/neurokit2/stats/hdi.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np from ..misc import find_closest from .density import density def hdi(x, ci=0.95, show=False, **kwargs): """**Highest Density Interval (HDI)** Compute the Highest Density Interval (HDI) of a distribution. All points within this interval have a higher probability density than points outside the interval. The HDI can be used in the context of uncertainty characterisation of posterior distributions (in the Bayesian farmework) as Credible Interval (CI). Unlike equal-tailed intervals that typically exclude 2.5% from each tail of the distribution and always include the median, the HDI is not equal-tailed and therefore always includes the mode(s) of posterior distributions. Parameters ---------- x : Union[list, np.array, pd.Series] A vector of values. ci : float Value of probability of the (credible) interval - CI (between 0 and 1) to be estimated. Default to .95 (95%). show : bool If ``True``, the function will produce a figure. **kwargs : Line2D properties Other arguments to be passed to :func:`nk.density`. See Also -------- density Returns ---------- float(s) The HDI low and high limits. fig Distribution plot. Examples ---------- .. ipython:: python import numpy as np import neurokit2 as nk x = np.random.normal(loc=0, scale=1, size=100000) @savefig p_hdi1.png scale=100% ci_min, ci_high = nk.hdi(x, ci=0.95, show=True) @suppress plt.close() """ x_sorted = np.sort(x) window_size = np.ceil(ci * len(x_sorted)).astype("int") if window_size < 2: raise ValueError( "NeuroKit error: hdi(): `ci` is too small or x does not contain enough data points." ) nCIs = len(x_sorted) - window_size ciWidth = [0] * nCIs for i in np.arange(0, nCIs): ciWidth[i] = x_sorted[i + window_size] - x_sorted[i] hdi_low = x_sorted[ciWidth.index(np.min(ciWidth))] hdi_high = x_sorted[ciWidth.index(np.min(ciWidth)) + window_size] if show is True: _hdi_plot(x, hdi_low, hdi_high, **kwargs) return hdi_low, hdi_high def _hdi_plot(vals, hdi_low, hdi_high, ci=0.95, **kwargs): x, y = density(vals, show=False, **kwargs) where = np.full(len(x), False) where[0 : find_closest(hdi_low, x, return_index=True)] = True where[find_closest(hdi_high, x, return_index=True) : :] = True fig, ax = plt.subplots() # pylint: disable=unused-variable ax.plot(x, y, color="white") ax.fill_between( x, y, where=where, color="#E91E63", label="CI {:.0%} [{:.2f}, {:.2f}]".format(ci, hdi_low, hdi_high), ) ax.fill_between(x, y, where=~where, color="#2196F3") ax.legend(loc="upper right")
2,905
29.270833
98
py
NeuroKit
NeuroKit-master/neurokit2/stats/correlation.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import scipy.stats def cor(x, y, method="pearson", show=False): """**Density estimation** Computes kernel density estimates. Parameters ----------- x : Union[list, np.array, pd.Series] Vectors of values. y : Union[list, np.array, pd.Series] Vectors of values. method : str Correlation method. Can be one of ``"pearson"``, ``"spearman"``, ``"kendall"``. show : bool Draw a scatterplot with a regression line. Returns ------- r The correlation coefficient. Examples -------- .. ipython:: python import neurokit2 as nk x = [1, 2, 3, 4, 5] y = [3, 1, 5, 6, 6] @savefig p_cor1.png scale=100% corr = nk.cor(x, y, method="pearson", show=True) @suppress plt.close() corr """ r, p = _cor_methods(x, y, method) # pylint: disable=unused-variable if show is True: _cor_plot(x, y) return r # ============================================================================= # Internals # ============================================================================= def _cor_methods(x, y, method="pearson"): method = method.lower() if method in ["pearson", "pears", "p", "r"]: r, p = scipy.stats.pearsonr(x, y) elif method in ["spearman", "spear", "s", "rho"]: r, p = scipy.stats.spearmanr(x, y, nan_policy="omit") elif method in ["kendall", "kend", "k", "tau"]: r, p = scipy.stats.kendalltau(x, y, nan_policy="omit") else: raise ValueError("NeuroKit error: cor(): 'method' not recognized.") return r, p def _cor_plot(x, y): # Create scatter plt.plot(x, y, "o") # Add regresion line m, b = np.polyfit(x, y, 1) plt.plot(np.array(x), m * np.array(x) + b)
1,874
23.671053
87
py
NeuroKit
NeuroKit-master/neurokit2/stats/density.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd import scipy.stats from .density_bandwidth import density_bandwidth def density(x, desired_length=100, bandwidth="Scott", show=False, **kwargs): """Density estimation. Computes kernel density estimates. Parameters ----------- x : Union[list, np.array, pd.Series] A vector of values. desired_length : int The amount of values in the returned density estimation. bandwidth : float Passed to the ``method`` argument from the :func:`.density_bandwidth` function. show : bool Display the density plot. **kwargs Additional arguments to be passed to :func:`.density_bandwidth`. Returns ------- x The x axis of the density estimation. y The y axis of the density estimation. See Also -------- density_bandwidth Examples -------- .. ipython:: python import neurokit2 as nk signal = nk.ecg_simulate(duration=20) @savefig p_density1.png scale=100% x, y = nk.density(signal, bandwidth=0.5, show=True) @suppress plt.close() .. ipython:: python # Bandwidth comparison _, y2 = nk.density(signal, bandwidth=1) _, y3 = nk.density(signal, bandwidth=2) _, y4 = nk.density(signal, bandwidth="scott") _, y5 = nk.density(signal, bandwidth="silverman") _, y6 = nk.density(signal, bandwidth="kernsmooth") @savefig p_density2.png scale=100% nk.signal_plot([y, y2, y3, y4, y5, y6], labels=["0.5", "1", "2", "Scott", "Silverman", "KernSmooth"]) @suppress plt.close() """ if "method" in kwargs: kwargs.pop("method") bw = density_bandwidth(x, method=bandwidth, **kwargs) density_function = scipy.stats.gaussian_kde(x, bw_method=bw) x = np.linspace(np.nanmin(x), np.nanmax(x), num=desired_length) y = density_function(x) if show is True: pd.DataFrame({"x": x, "y": y}).plot(x="x") return x, y
2,045
24.898734
87
py
NeuroKit
NeuroKit-master/neurokit2/stats/cluster.py
# -*- coding: utf-8 -*- import functools import warnings import numpy as np import pandas as pd import scipy.linalg import scipy.spatial import sklearn.cluster import sklearn.decomposition import sklearn.mixture from ..misc import check_random_state from .cluster_quality import _cluster_quality_distance def cluster(data, method="kmeans", n_clusters=2, random_state=None, optimize=False, **kwargs): """**Data Clustering** Performs clustering of data using different algorithms. * **kmod**: Modified k-means algorithm. * **kmeans**: Normal k-means. * **kmedoids**: k-medoids clustering, a more stable version of k-means. * **pca**: Principal Component Analysis. * **ica**: Independent Component Analysis. * **aahc**: Atomize and Agglomerate Hierarchical Clustering. Computationally heavy. * **hierarchical** * **spectral** * **mixture** * **mixturebayesian** See ``sklearn`` for methods details. Parameters ---------- data : np.ndarray Matrix array of data (E.g., an array (channels, times) of M/EEG data). method : str The algorithm for clustering. Can be one of ``"kmeans"`` (default), ``"kmod"``, ``"kmedoids"``, ``"pca"``, ``"ica"``, ``"aahc"``, ``"hierarchical"``, ``"spectral"``, ``"mixture"``, ``"mixturebayesian"``. n_clusters : int The desired number of clusters. random_state : Union[int, numpy.random.RandomState] The ``RandomState`` for the random number generator. Defaults to ``None``, in which case a different random state is chosen each time this function is called. optimize : bool Optimized method in Poulsen et al. (2018) for the *k*-means modified method. **kwargs Other arguments to be passed into ``sklearn`` functions. Returns ------- clustering : DataFrame Information about the distance of samples from their respective clusters. clusters : np.ndarray Coordinates of cluster centers, which has a shape of n_clusters x n_features. info : dict Information about the number of clusters, the function and model used for clustering. Examples ---------- .. ipython:: python import neurokit2 as nk import matplotlib.pyplot as plt # Load the iris dataset data = nk.data("iris").drop("Species", axis=1) # Cluster using different methods clustering_kmeans, clusters_kmeans, info = nk.cluster(data, method="kmeans", n_clusters=3) clustering_spectral, clusters_spectral, info = nk.cluster(data, method="spectral", n_clusters=3) clustering_hierarchical, clusters_hierarchical, info = nk.cluster(data, method="hierarchical", n_clusters=3) clustering_agglomerative, clusters_agglomerative, info= nk.cluster(data, method="agglomerative", n_clusters=3) clustering_mixture, clusters_mixture, info = nk.cluster(data, method="mixture", n_clusters=3) clustering_bayes, clusters_bayes, info = nk.cluster(data, method="mixturebayesian", n_clusters=3) clustering_pca, clusters_pca, info = nk.cluster(data, method="pca", n_clusters=3) clustering_ica, clusters_ica, info = nk.cluster(data, method="ica", n_clusters=3) clustering_kmod, clusters_kmod, info = nk.cluster(data, method="kmod", n_clusters=3) clustering_kmedoids, clusters_kmedoids, info = nk.cluster(data, method="kmedoids", n_clusters=3) clustering_aahc, clusters_aahc, info = nk.cluster(data, method='aahc_frederic', n_clusters=3) # Visualize classification and 'average cluster' @savefig p_cluster1.png scale=100% fig, axes = plt.subplots(ncols=2, nrows=5) axes[0, 0].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_kmeans['Cluster']) axes[0, 0].scatter(clusters_kmeans[:, 2], clusters_kmeans[:, 3], c='red') axes[0, 0].set_title("k-means") axes[0, 1].scatter(data.iloc[:,[2]], data.iloc[:, [3]], c=clustering_spectral['Cluster']) axes[0, 1].scatter(clusters_spectral[:, 2], clusters_spectral[:, 3], c='red') axes[0, 1].set_title("Spectral") axes[1, 0].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_hierarchical['Cluster']) axes[1, 0].scatter(clusters_hierarchical[:, 2], clusters_hierarchical[:, 3], c='red') axes[1, 0].set_title("Hierarchical") axes[1, 1].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_agglomerative['Cluster']) axes[1, 1].scatter(clusters_agglomerative[:, 2], clusters_agglomerative[:, 3], c='red') axes[1, 1].set_title("Agglomerative") axes[2, 0].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_mixture['Cluster']) axes[2, 0].scatter(clusters_mixture[:, 2], clusters_mixture[:, 3], c='red') axes[2, 0].set_title("Mixture") axes[2, 1].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_bayes['Cluster']) axes[2, 1].scatter(clusters_bayes[:, 2], clusters_bayes[:, 3], c='red') axes[2, 1].set_title("Bayesian Mixture") axes[3, 0].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_pca['Cluster']) axes[3, 0].scatter(clusters_pca[:, 2], clusters_pca[:, 3], c='red') axes[3, 0].set_title("PCA") axes[3, 1].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_ica['Cluster']) axes[3, 1].scatter(clusters_ica[:, 2], clusters_ica[:, 3], c='red') axes[3, 1].set_title("ICA") axes[4, 0].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_kmod['Cluster']) axes[4, 0].scatter(clusters_kmod[:, 2], clusters_kmod[:, 3], c='red') axes[4, 0].set_title("modified K-means") axes[4, 1].scatter(data.iloc[:,[2]], data.iloc[:,[3]], c=clustering_aahc['Cluster']) axes[4, 1].scatter(clusters_aahc[:, 2], clusters_aahc[:, 3], c='red') axes[4, 1].set_title("AAHC (Frederic's method)") @suppress plt.close() References ----------- * Park, H. S., & Jun, C. H. (2009). A simple and fast algorithm for K-medoids clustering. Expert systems with applications, 36(2), 3336-3341. """ # Sanity checks if isinstance(data, pd.DataFrame): data = data.values method = method.lower() # K-means if method in ["kmeans", "k", "k-means", "kmean"]: out = _cluster_kmeans(data, n_clusters=n_clusters, random_state=random_state, **kwargs) # Modified k-means elif method in ["kmods", "kmod", "kmeans modified", "modified kmeans"]: out = _cluster_kmod( data, n_clusters=n_clusters, random_state=random_state, optimize=optimize, **kwargs ) # K-medoids elif method in ["kmedoids", "k-medoids", "k-centers"]: out = _cluster_kmedoids(data, n_clusters=n_clusters, random_state=random_state, **kwargs) # PCA elif method in ["pca", "principal", "principal component analysis"]: out = _cluster_pca(data, n_clusters=n_clusters, random_state=random_state, **kwargs) # ICA elif method in ["ica", "independent", "independent component analysis"]: out = _cluster_pca(data, n_clusters=n_clusters, random_state=random_state, **kwargs) # Mixture elif method in ["mixture", "mixt"]: out = _cluster_mixture( data, n_clusters=n_clusters, bayesian=False, random_state=random_state, **kwargs ) # Frederic's AAHC elif method in ["aahc_frederic", "aahc_eegmicrostates"]: out = _cluster_aahc(data, n_clusters=n_clusters, random_state=random_state, **kwargs) # Bayesian elif method in ["bayesianmixture", "bayesmixt", "mixturebayesian", "mixturebayes"]: out = _cluster_mixture( data, n_clusters=n_clusters, bayesian=True, random_state=random_state, **kwargs ) # Others else: out = _cluster_sklearn(data, n_clusters=n_clusters, **kwargs) return out # ============================================================================= # ============================================================================= # # Methods # ============================================================================= # ============================================================================= # ============================================================================= # Kmeans # ============================================================================= def _cluster_kmeans(data, n_clusters=2, random_state=None, n_init="auto", **kwargs): """K-means clustering algorithm""" # Initialize clustering function clustering_model = sklearn.cluster.KMeans( n_clusters=n_clusters, random_state=random_state, n_init=n_init, **kwargs ) # Fit clustering = clustering_model.fit_predict(data) # Get representatives (identical to _cluster_getclusters(), # but why recompute when already included) clusters = clustering_model.cluster_centers_ # Get distance prediction = _cluster_quality_distance(data, clusters, to_dataframe=True) prediction["Cluster"] = clustering # Copy function with given parameters clustering_function = functools.partial( _cluster_kmeans, n_clusters=n_clusters, random_state=random_state, n_init=n_init, **kwargs ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "sklearn_model": clustering_model, "random_state": random_state, } return prediction, clusters, info # ============================================================================= # K-medoids # ============================================================================= def _cluster_kmedoids(data, n_clusters=2, max_iterations=1000, random_state=None, **kwargs): """Peforms k-medoids clustering which is based on the most centrally located object in a cluster. Less sensitive to outliers than K-means clustering. Adapted from https://github.com/rakeshvar/kmedoids/. Original proposed algorithm from Park & Jun (2009). """ # Sanitize if isinstance(data, pd.DataFrame): data = np.array(data) n_samples = data.shape[0] # Step 1: Initialize random medoids rng = check_random_state(random_state) ids_of_medoids = rng.choice(n_samples, n_clusters, replace=False) # Find distance between objects to their medoids, can be euclidean or manhatten def find_distance(x, y, dist_method="euclidean"): if dist_method == "euclidean": return np.sqrt(np.sum(np.square(x - y), axis=-1)) elif dist_method == "manhatten": return np.sum(np.abs(x - y), axis=-1) individual_points = data[:, None, :] medoid_points = data[None, ids_of_medoids, :] distance = find_distance(individual_points, medoid_points) # Assign each point to the nearest medoid segmentation = np.argmin(distance, axis=1) # Step 2: Update medoids for i in range(max_iterations): # Find new medoids ids_of_medoids = np.full(n_clusters, -1, dtype=int) for i in range(n_clusters): indices = np.where(segmentation == i)[0] distances = find_distance(data[indices, None, :], data[None, indices, :]).sum(axis=0) ids_of_medoids[i] = indices[np.argmin(distances)] # Step 3: Reassign objects to medoids new_distances = find_distance(data[:, None, :], data[None, ids_of_medoids, :]) new_assignments = np.argmin(new_distances, axis=1) diffs = np.mean(new_assignments != segmentation) segmentation = new_assignments # If more than 10% of assignments are equal to previous segmentation, # go back to step 2 (threshold arbitrary?) if diffs <= 0.01: break # Data points as centroids clusters = data[ids_of_medoids] # Get prediction prediction = _cluster_quality_distance(data, clusters, to_dataframe=True) prediction["Cluster"] = segmentation # Copy function with given parameters clustering_function = functools.partial( _cluster_kmedoids, n_clusters=n_clusters, max_iterations=max_iterations, random_state=random_state, ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "random_state": random_state, "clusters": clusters, } return prediction, clusters, info # ============================================================================= # Modified K-means # ============================================================================= def _cluster_kmod( data, n_clusters=4, max_iterations=1000, threshold=1e-6, random_state=None, optimize=False, **kwargs ): """The modified K-means clustering algorithm, adapted from Marijn van Vliet and Frederic von Wegner. https://github.com/wmvanvliet/mne_microstates https://github.com/Frederic-vW/eeg_microstates Parameters ----------- n_clusters : int The number of unique microstates to find. Defaults to 4. max_iterations : int The maximum number of iterations to perform in the k-means algorithm. Defaults to 1000. threshold : float The threshold of convergence for the k-means algorithm, based on relative change in noise variance. Defaults to 1e-6. random_state : Union[int, numpy.random.RandomState, None] The seed or ``RandomState`` for the random number generator. Defaults to ``None``, in which case a different seed is chosen each time this function is called. optimized : bool To use a new optimized method in https://www.biorxiv.org/content/10.1101/289850v1.full.pdf. For the Kmeans modified method. Default to False. **kwargs Other arguments to be passed into ``sklearn`` functions. Returns ------- clustering : DataFrame Information about the distance of samples from their respective clusters. clusters : np.ndarray Coordinates of cluster centers, which has a shape of n_clusters x n_features. info : dict Information about the number of clusters, the function and model used for clustering. """ n_samples, n_channels = data.shape # Cache this value for later to compute residual data_sum_sq = np.sum(data**2) # Select random timepoints for our initial topographic maps rng = check_random_state(random_state) init_times = rng.choice(n_samples, size=n_clusters, replace=False) # Initialize random cluster centroids clusters = data[init_times, :] # Normalize row-wise (across EEG channels) clusters /= np.linalg.norm(clusters, axis=1, keepdims=True) # Normalize the maps # Initialize iteration prev_residual = 0 for i in range(max_iterations): # Step 3: Assign each sample to the best matching microstate activation = clusters.dot(data.T) segmentation = np.argmax(np.abs(activation), axis=0) # Step 4: Recompute the topographic maps of the microstates, based on the # samples that were assigned to each state. for state in np.arange(n_clusters): # Get data fro specific state idx = segmentation == state data_state = data[idx, :] # Sanity check if np.sum(idx) == 0: clusters[state] = 0 continue # Retrieve map values if optimize: # Method 2 - optimized segmentation state_vals = data_state.T.dot(activation[state, idx]) else: # Method 1 - eighen value # step 4a Sk = np.dot(data_state.T, data_state) # step 4b eigen_vals, eigen_vectors = scipy.linalg.eigh(Sk) state_vals = eigen_vectors[:, np.argmax(np.abs(eigen_vals))] state_vals /= np.linalg.norm(state_vals) # Normalize Map clusters[state, :] = state_vals # Store map # Estimate residual noise (step 5) act_sum_sq = np.sum(np.sum(clusters[segmentation, :] * data, axis=1) ** 2) residual = np.abs(data_sum_sq - act_sum_sq) residual = residual / float(n_samples * (n_channels - 1)) # Have we converged? Convergence criterion: variance estimate (step 6) if np.abs(prev_residual - residual) < (threshold * residual): break # Next iteration prev_residual = residual.copy() if i == max_iterations: warnings.warn( "Modified K-means algorithm failed to converge after " + str(i) + "", "iterations. Consider increasing 'max_iterations'.", ) # De-normalize clusters_unnormalized = _cluster_getclusters(data, segmentation) prediction = _cluster_quality_distance(data, clusters_unnormalized, to_dataframe=True) prediction["Cluster"] = segmentation # Copy function with given parameters clustering_function = functools.partial( _cluster_kmod, n_clusters=n_clusters, max_iterations=max_iterations, threshold=threshold, random_state=random_state, **kwargs ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "random_state": random_state, "clusters_normalized": clusters, "residual": residual, } return prediction, clusters_unnormalized, info # ============================================================================= # PCA # ============================================================================= def _cluster_pca(data, n_clusters=2, random_state=None, **kwargs): """Principal Component Analysis (PCA) for clustering.""" # Fit PCA pca = sklearn.decomposition.PCA( n_components=n_clusters, copy=True, whiten=True, svd_solver="auto", random_state=random_state, **kwargs ) pca = pca.fit(data) # clusters = np.array([pca.components_[state, :] for state in range(n_clusters)]) # Compute variance explained # explained_var = pca.explained_variance_ratio_ # total_explained_var = np.sum(pca.explained_variance_ratio_) # Get distance prediction = pca.transform(data) prediction = pd.DataFrame(prediction).add_prefix("Loading_") prediction["Cluster"] = prediction.abs().idxmax(axis=1).values prediction["Cluster"] = [ np.where(prediction.columns == state)[0][0] for state in prediction["Cluster"] ] # Recover states from clustering clusters = _cluster_getclusters(data, prediction["Cluster"]) # Copy function with given parameters clustering_function = functools.partial( _cluster_pca, n_clusters=n_clusters, random_state=random_state, **kwargs ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "random_state": random_state, } return prediction, clusters, info # ============================================================================= # ICA # ============================================================================= def _cluster_ica(data, n_clusters=2, random_state=None, **kwargs): """Independent Component Analysis (ICA) for clustering.""" # Fit ICA ica = sklearn.decomposition.FastICA( n_components=n_clusters, algorithm="parallel", whiten=True, fun="exp", random_state=random_state, **kwargs ) ica = ica.fit(data) # clusters = np.array([ica.components_[state, :] for state in range(n_clusters)]) # Get distance prediction = ica.transform(data) prediction = pd.DataFrame(prediction).add_prefix("Loading_") prediction["Cluster"] = prediction.abs().idxmax(axis=1).values prediction["Cluster"] = [ np.where(prediction.columns == state)[0][0] for state in prediction["Cluster"] ] # Copy function with given parameters clustering_function = functools.partial( _cluster_ica, n_clusters=n_clusters, random_state=random_state, **kwargs ) # Recover states from clustering clusters = _cluster_getclusters(data, prediction["Cluster"]) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "random_state": random_state, } return prediction, clusters, info # ============================================================================= # SKLEARN # ============================================================================= def _cluster_sklearn(data, method="spectral", n_clusters=2, **kwargs): """Spectral clustering""" # Initialize clustering function if method in ["spectral"]: clustering_model = sklearn.cluster.SpectralClustering(n_clusters=n_clusters, **kwargs) elif method in ["hierarchical", "ward"]: clustering_model = sklearn.cluster.AgglomerativeClustering( n_clusters=n_clusters, linkage="ward", **kwargs ) elif method in ["agglomerative", "single"]: clustering_model = sklearn.cluster.AgglomerativeClustering( n_clusters=n_clusters, linkage="single", **kwargs ) # Fit clustering = clustering_model.fit_predict(data) # Get representatives clusters = _cluster_getclusters(data, clustering) # Get distance prediction = _cluster_quality_distance(data, clusters, to_dataframe=True) prediction["Cluster"] = clustering # Else, copy function clustering_function = functools.partial(_cluster_sklearn, n_clusters=n_clusters, **kwargs) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "sklearn_model": clustering_model, } return prediction, clusters, info def _cluster_mixture(data, n_clusters=2, bayesian=False, random_state=None, **kwargs): """Mixture model""" # Initialize clustering function if bayesian is False: clustering_model = sklearn.mixture.GaussianMixture( n_components=n_clusters, random_state=random_state, **kwargs ) else: clustering_model = sklearn.mixture.BayesianGaussianMixture( n_components=n_clusters, random_state=random_state, **kwargs ) # Fit clustering = clustering_model.fit_predict(data) # Get representatives clusters = clustering_model.means_ # Get probability prediction = clustering_model.predict_proba(data) prediction = pd.DataFrame(prediction).add_prefix("Probability_") prediction["Cluster"] = clustering # Else, copy function clustering_function = functools.partial( _cluster_mixture, n_clusters=n_clusters, random_state=random_state, **kwargs ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "sklearn_model": clustering_model, "random_state": random_state, } return prediction, clusters, info # ============================================================================= # AAHC # ============================================================================= def _cluster_aahc( data, n_clusters=2, gfp=None, gfp_peaks=None, gfp_sum_sq=None, random_state=None, use_peaks=False, **kwargs ): """Atomize and Agglomerative Hierarchical Clustering Algorithm, AAHC (Murray et al., Brain Topography, 2008), implemented by https://github.com/Frederic-vW/eeg_microstates/blob/master/eeg_microstates.py#L518 Preprocessing steps of GFP computation are necessary for the algorithm to run. If gfp arguments are specified, data is assumed to have been filtered out based on gfp peaks (e.g., data[:, indices]), if not specified, gfp indices will be calculated in the algorithm and data is assumed to be the full un-preprocessed input. """ # Internal functions for aahc def extract_row(A, k): v = A[k, :] A_ = np.vstack((A[:k, :], A[k + 1 :, :])) return A_, v def extract_item(A, k): a = A[k] A_ = A[:k] + A[k + 1 :] return A_, a def locmax(x): """Get local maxima of 1D-array Args: x: numeric sequence Returns: m: list, 1D-indices of local maxima """ dx = np.diff(x) # discrete 1st derivative zc = np.diff(np.sign(dx)) # zero-crossings of dx m = 1 + np.where(zc == -2)[0] # indices of local max. return m # Sanitize if isinstance(data, pd.DataFrame): data = np.array(data) _, nch = data.shape # If preprocessing is not Done already if gfp is None and gfp_peaks is None and gfp_sum_sq is None: gfp = data.std(axis=1) gfp_peaks = locmax(gfp) gfp_sum_sq = np.sum(gfp**2) # normalizing constant in GEV if use_peaks: maps = data[gfp_peaks, :] # initialize clusters cluster_data = data[gfp_peaks, :] # store original gfp peak indices else: maps = data.copy() cluster_data = data.copy() else: maps = data.copy() cluster_data = data.copy() n_maps = maps.shape[0] # cluster indices w.r.t. original size, normalized GFP peak data Ci = [[k] for k in range(n_maps)] # Main loop: atomize + agglomerate while n_maps > n_clusters: # correlations of the data sequence with each cluster m_x, s_x = data.mean(axis=1, keepdims=True), data.std(axis=1) m_y, s_y = maps.mean(axis=1, keepdims=True), maps.std(axis=1) s_xy = 1.0 * nch * np.outer(s_x, s_y) C = np.dot(data - m_x, np.transpose(maps - m_y)) / s_xy # microstate sequence, ignore polarity L = np.argmax(C**2, axis=1) # GEV (global explained variance) of cluster k gev = np.zeros(n_maps) for k in range(n_maps): r = L == k gev[k] = np.sum(gfp[r] ** 2 * C[r, k] ** 2) / gfp_sum_sq # merge cluster with the minimum GEV imin = np.argmin(gev) # N => N-1 maps, _ = extract_row(maps, imin) Ci, reC = extract_item(Ci, imin) re_cluster = [] # indices of updated clusters for k in reC: # map index to re-assign c = cluster_data[k, :] m_x, s_x = maps.mean(axis=1, keepdims=True), maps.std(axis=1) m_y, s_y = c.mean(), c.std() s_xy = 1.0 * nch * s_x * s_y C = np.dot(maps - m_x, c - m_y) / s_xy inew = np.argmax(C**2) # ignore polarity re_cluster.append(inew) Ci[inew].append(k) n_maps = len(Ci) # Update clusters re_cluster = list(set(re_cluster)) # unique list of updated clusters # re-clustering by eigenvector method for i in re_cluster: idx = Ci[i] Vt = cluster_data[idx, :] Sk = np.dot(Vt.T, Vt) evals, evecs = np.linalg.eig(Sk) c = evecs[:, np.argmax(np.abs(evals))] c = np.real(c) maps[i] = c / np.sqrt(np.sum(c**2)) # Get distance prediction = _cluster_quality_distance(cluster_data, maps, to_dataframe=True) prediction["Cluster"] = prediction.abs().idxmax(axis=1).values prediction["Cluster"] = [ np.where(prediction.columns == state)[0][0] for state in prediction["Cluster"] ] # Function clustering_function = functools.partial( _cluster_aahc, n_clusters=n_clusters, random_state=random_state, **kwargs ) # Info dump info = { "n_clusters": n_clusters, "clustering_function": clustering_function, "random_state": random_state, } return prediction, maps, info # ============================================================================= # ============================================================================= # # Utils # ============================================================================= # ============================================================================= def _cluster_getclusters(data, clustering): """Get average representatives of clusters""" n_clusters = len(np.unique(clustering)) return np.asarray([np.mean(data[np.where(clustering == i)], axis=0) for i in range(n_clusters)])
28,345
35.812987
116
py
NeuroKit
NeuroKit-master/neurokit2/stats/cluster_findnumber.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd from .cluster import cluster from .cluster_quality import cluster_quality def cluster_findnumber(data, method="kmeans", n_max=10, show=False, **kwargs): """**Optimal Number of Clusters** Find the optimal number of clusters based on different indices of quality of fit. Parameters ---------- data : np.ndarray An array (channels, times) of M/EEG data. method : str The clustering algorithm to be passed into :func:`.nk.cluster`. n_max : int Runs the clustering alogrithm from 1 to n_max desired clusters in :func:`.nk.cluster` with quality metrices produced for each cluster number. show : bool Plot indices normalized on the same scale. **kwargs Other arguments to be passed into :func:`.nk.cluster` and :func:`.nk.cluster_quality`. Returns ------- DataFrame The different quality scores for each number of clusters: * Score_Silhouette * Score_Calinski * Score_Bouldin * Score_VarianceExplained * Score_GAP * Score_GAPmod * Score_GAP_diff * Score_GAPmod_diff See Also -------- cluster, cluster_quality Examples ---------- .. ipython:: python import neurokit2 as nk # Load the iris dataset data = nk.data("iris").drop("Species", axis=1) # How many clusters @savefig p_cluster_findnumber1.png scale=100% results = nk.cluster_findnumber(data, method="kmeans", show=True) @suppress plt.close() """ results = [] for i in range(1, n_max): # Cluster clustering, clusters, info = cluster(data, method=method, n_clusters=i, **kwargs) # Compute indices of clustering quality _, quality = cluster_quality(data, clustering, clusters, info, **kwargs) results.append(quality) results = pd.concat(results, axis=0).reset_index(drop=True) # Gap Diff results["Score_GAP_diff"] = ( results["Score_GAP"] - results["Score_GAP"].shift(-1) + results["Score_GAP_sk"].shift(-1) ) results["Score_GAPmod_diff"] = ( results["Score_GAPmod"] - results["Score_GAPmod"].shift(-1) + results["Score_GAPmod_sk"].shift(-1) ) results = results.drop(["Score_GAP_sk", "Score_GAPmod_sk"], axis=1) if show is True: normalized = (results - results.min()) / (results.max() - results.min()) normalized["n_Clusters"] = np.rint(np.arange(1, n_max)) normalized.columns = normalized.columns.str.replace("Score", "Normalized") normalized.plot(x="n_Clusters") return results
2,697
28.977778
98
py
NeuroKit
NeuroKit-master/neurokit2/stats/fit_loess.py
# -*- coding: utf-8 -*- import numpy as np import scipy.linalg def fit_loess(y, X=None, alpha=0.75, order=2): """**Local Polynomial Regression (LOESS)** Performs a LOWESS (LOcally WEighted Scatter-plot Smoother) regression. Parameters ---------- y : Union[list, np.array, pd.Series] The response variable (the y axis). X : Union[list, np.array, pd.Series] Explanatory variable (the x axis). If ``None``, will treat y as a continuous signal (useful for smoothing). alpha : float The parameter which controls the degree of smoothing, which corresponds to the proportion of the samples to include in local regression. order : int Degree of the polynomial to fit. Can be 1 or 2 (default). Returns ------- array Prediction of the LOESS algorithm. dict Dictionary containing additional information such as the parameters (``order`` and ``alpha``). See Also ---------- signal_smooth, signal_detrend, fit_error Examples --------- .. ipython:: python import pandas as pd import neurokit2 as nk # Simulate Signal signal = np.cos(np.linspace(start=0, stop=10, num=1000)) # Add noise to signal distorted = nk.signal_distort(signal, noise_amplitude=[0.3, 0.2, 0.1], noise_frequency=[5, 10, 50]) # Smooth signal using local regression @savefig p_fit_loess1.png scale=100% pd.DataFrame({ "Raw": distorted, "Loess_1": nk.fit_loess(distorted, order=1)[0], "Loess_2": nk.fit_loess(distorted, order=2)[0]}).plot() @suppress plt.close() References ---------- * https://simplyor.netlify.com/loess-from-scratch-in-python-animation.en-us/ """ if X is None: X = np.linspace(0, 100, len(y)) assert order in [1, 2], "Deg has to be 1 or 2" assert 0 < alpha <= 1, "Alpha has to be between 0 and 1" assert len(X) == len(y), "Length of X and y are different" X_domain = X n = len(X) span = int(np.ceil(alpha * n)) y_predicted = np.zeros(len(X_domain)) x_space = np.zeros_like(X_domain) for i, val in enumerate(X_domain): distance = abs(X - val) sorted_dist = np.sort(distance) ind = np.argsort(distance) Nx = X[ind[:span]] Ny = y[ind[:span]] delx0 = sorted_dist[span - 1] u = distance[ind[:span]] / delx0 w = (1 - u**3) ** 3 W = np.diag(w) A = np.vander(Nx, N=1 + order) V = np.matmul(np.matmul(A.T, W), A) Y = np.matmul(np.matmul(A.T, W), Ny) Q, R = scipy.linalg.qr(V) p = scipy.linalg.solve_triangular(R, np.matmul(Q.T, Y)) y_predicted[i] = np.polyval(p, val) x_space[i] = val return y_predicted, {"alpha": alpha, "order": order}
2,919
27.349515
102
py
NeuroKit
NeuroKit-master/neurokit2/stats/standardize.py
# -*- coding: utf-8 -*- from warnings import warn import numpy as np import pandas as pd from ..misc import NeuroKitWarning from ..misc.check_type import is_string from .mad import mad def standardize(data, robust=False, window=None, **kwargs): """**Standardization of data** Performs a standardization of data (Z-scoring), i.e., centering and scaling, so that the data is expressed in terms of standard deviation (i.e., mean = 0, SD = 1) or Median Absolute Deviance (median = 0, MAD = 1). Parameters ---------- data : Union[list, np.array, pd.Series] Raw data. robust : bool If ``True``, centering is done by substracting the median from the variables and dividing it by the median absolute deviation (MAD). If ``False``, variables are standardized by substracting the mean and dividing it by the standard deviation (SD). window : int Perform a rolling window standardization, i.e., apply a standardization on a window of the specified number of samples that rolls along the main axis of the signal. Can be used for complex detrending. **kwargs : optional Other arguments to be passed to :func:`.pandas.rolling`. Returns ---------- list The standardized values. Examples ---------- .. ipython:: python import neurokit2 as nk import pandas as pd # Simple example nk.standardize([3, 1, 2, 4, 6, np.nan]) nk.standardize([3, 1, 2, 4, 6, np.nan], robust=True) nk.standardize(np.array([[1, 2, 3, 4], [5, 6, 7, 8]]).T) nk.standardize(pd.DataFrame({"A": [3, 1, 2, 4, 6, np.nan], "B": [3, 1, 2, 4, 6, 5]})) # Rolling standardization of a signal signal = nk.signal_simulate(frequency=[0.1, 2], sampling_rate=200) z = nk.standardize(signal, window=200) @savefig p_standardize1.png scale=100% nk.signal_plot([signal, z], standardize=True) @suppress plt.close() """ # Return appropriate type if isinstance(data, list): if any(is_string(data)): out = data warn( "The data is not standardized." "Some elements in the list is of string type.", category=NeuroKitWarning ) else: out = list(_standardize(np.array(data), robust=robust, window=window, **kwargs)) elif isinstance(data, pd.DataFrame): # only standardize columns that are not string and are not nan _data = data.loc[:, ~is_string(data) & ~np.array(data.isnull().all())] to_append = data.loc[:, is_string(data) | np.array(data.isnull().all())] out = pd.DataFrame(_standardize(_data, robust=robust, window=window, **kwargs)) out = pd.concat([to_append, out], axis=1) elif isinstance(data, pd.Series): if is_string(data): out = data warn( "The data is not standardized as it is of string type.", category=NeuroKitWarning) else: out = pd.Series(_standardize(data, robust=robust, window=window, **kwargs)) else: if is_string(data): out = data warn( "The data is not standardized as it is of string type.", category=NeuroKitWarning) else: out = _standardize(data, robust=robust, window=window, **kwargs) return out # ============================================================================= # Internals # ============================================================================= def _standardize(data, robust=False, window=None, **kwargs): # Compute standardized on whole data if window is None: if robust is False: z = (data - np.nanmean(data, axis=0)) / np.nanstd(data, axis=0, ddof=1) else: z = (data - np.nanmedian(data, axis=0)) / mad(data) # Rolling standardization on windows else: df = pd.DataFrame(data) # Force dataframe if robust is False: z = (df - df.rolling(window, min_periods=0, **kwargs).mean()) / df.rolling( window, min_periods=0, **kwargs ).std(ddof=1) else: z = (df - df.rolling(window, min_periods=0, **kwargs).median()) / df.rolling( window, min_periods=0, **kwargs ).apply(mad) # Fill the created nans z = z.fillna(method="bfill") # Restore to vector or array if z.shape[1] == 1: z = z[0].values else: z = z.values return z
4,654
32.014184
100
py
NeuroKit
NeuroKit-master/neurokit2/stats/rescale.py
# -*- coding: utf-8 -*- import numpy as np def rescale(data, to=[0, 1], scale=None): """**Rescale data** Rescale a numeric variable to a new range. Parameters ---------- data : Union[list, np.array, pd.Series] Raw data. to : list New range of values of the data after rescaling. scale : list A list or tuple of two values specifying the actual range of the data. If ``None``, the minimum and the maximum of the provided data will be used. Returns ---------- list The rescaled values. Examples ---------- .. ipython:: python import neurokit2 as nk nk.rescale([3, 1, 2, 4, 6], to=[0, 1]) """ # Return appropriate type if isinstance(data, list): data = list(_rescale(np.array(data), to=to, scale=scale)) else: data = _rescale(data, to=to, scale=scale) return data # ============================================================================= # Internals # ============================================================================= def _rescale(data, to=[0, 1], scale=None): if scale is None: scale = [np.nanmin(data), np.nanmax(data)] return (to[1] - to[0]) / (scale[1] - scale[0]) * (data - scale[0]) + to[0]
1,291
22.925926
79
py
NeuroKit
NeuroKit-master/neurokit2/stats/fit_polynomial.py
# -*- coding: utf-8 -*- import numpy as np import sklearn.linear_model import sklearn.metrics from .fit_error import fit_rmse def fit_polynomial(y, X=None, order=2, method="raw"): """**Polynomial Regression** Performs a polynomial regression of given order. Parameters ---------- y : Union[list, np.array, pd.Series] The response variable (the y axis). X : Union[list, np.array, pd.Series] Explanatory variable (the x axis). If ``None``, will treat y as a continuous signal. order : int The order of the polynomial. 0, 1 or > 1 for a baseline, linear or polynomial fit, respectively. Can also be ``"auto"``, in which case it will attempt to find the optimal order to minimize the RMSE. method : str If ``"raw"`` (default), compute standard polynomial coefficients. If ``"orthogonal"``, compute orthogonal polynomials (and is equivalent to R's ``poly`` default behavior). Returns ------- array Prediction of the regression. dict Dictionary containing additional information such as the parameters (``order``) used and the coefficients (``coefs``). See Also ---------- signal_detrend, fit_error, fit_polynomial_findorder Examples --------- .. ipython:: python import pandas as pd import neurokit2 as nk y = np.cos(np.linspace(start=0, stop=10, num=100)) @savefig p_fit_polynomial1.png scale=100% pd.DataFrame({"y": y, "Poly_0": nk.fit_polynomial(y, order=0)[0], "Poly_1": nk.fit_polynomial(y, order=1)[0], "Poly_2": nk.fit_polynomial(y, order=2)[0], "Poly_3": nk.fit_polynomial(y, order=3)[0], "Poly_5": nk.fit_polynomial(y, order=5)[0], "Poly_auto": nk.fit_polynomial(y, order='auto')[0]}).plot() @suppress plt.close() """ if X is None: X = np.linspace(0, 100, len(y)) # Optimal order if isinstance(order, str): order = fit_polynomial_findorder(y, X, max_order=6) # Make prediction if method == "raw": y_predicted, coefs = _fit_polynomial(y, X, order=order) else: y_predicted, coefs = _fit_polynomial_orthogonal(y, X, order=order) return y_predicted, { "order": order, "coefs": coefs, "R2": sklearn.metrics.r2_score(y, y_predicted), } # ============================================================================= # Find order # ============================================================================= def fit_polynomial_findorder(y, X=None, max_order=6): """Polynomial Regression. Find the optimal order for polynomial fitting. Currently, the only method implemented is RMSE minimization. Parameters ---------- y : Union[list, np.array, pd.Series] The response variable (the y axis). X : Union[list, np.array, pd.Series] Explanatory variable (the x axis). If 'None', will treat y as a continuous signal. max_order : int The maximum order to test. Returns ------- int Optimal order. See Also ---------- fit_polynomial Examples --------- import neurokit2 as nk y = np.cos(np.linspace(start=0, stop=10, num=100)) nk.fit_polynomial_findorder(y, max_order=10) 9 """ # TODO: add cross-validation or some kind of penalty to prevent over-fitting? if X is None: X = np.linspace(0, 100, len(y)) best_rmse = 0 for order in range(max_order): y_predicted, _ = _fit_polynomial(y, X, order=order) rmse = fit_rmse(y, y_predicted) if rmse < best_rmse or best_rmse == 0: best_order = order return best_order # ============================================================================= # Internals # ============================================================================= def _fit_polynomial(y, X, order=2): coefs = np.polyfit(X, y, order) # Generating weights and model for polynomial function with a given degree y_predicted = np.polyval(coefs, X) return y_predicted, coefs def _fit_polynomial_orthogonal(y, X, order=2): """Fit an orthogonal polynomial regression in Python (equivalent to R's poly()) from sklearn.datasets import load_iris import pandas as pd df = load_iris() df = pd.DataFrame(data=df.data, columns=df.feature_names) y = df.iloc[:, 0].values # Sepal.Length X = df.iloc[:, 1].values # Sepal.Width _fit_polynomial_orthogonal(y, X, order=2) # doctest: +SKIP # Equivalent to R's: # coef(lm(Sepal.Length ~ poly(Sepal.Width, 2), data=iris)) """ X = np.transpose([X**k for k in range(order + 1)]) X = np.linalg.qr(X)[0][:, 1:] model = sklearn.linear_model.LinearRegression().fit(X, y) return model.predict(X), np.insert(model.coef_, 0, model.intercept_)
4,961
29.819876
126
py
NeuroKit
NeuroKit-master/neurokit2/stats/mad.py
# -*- coding: utf-8 -*- import numpy as np def mad(x, constant=1.4826, **kwargs): """**Median Absolute Deviation: a "robust" version of standard deviation** Parameters ---------- x : Union[list, np.array, pd.Series] A vector of values. constant : float Scale factor. Use 1.4826 for results similar to default R. Returns ---------- float The MAD. Examples ---------- .. ipython:: python import neurokit2 as nk nk.mad([2, 8, 7, 5, 4, 12, 5, 1]) References ----------- * https://en.wikipedia.org/wiki/Median_absolute_deviation """ median = np.nanmedian(np.ma.array(x).compressed(), **kwargs) mad_value = np.nanmedian(np.abs(x - median), **kwargs) mad_value = mad_value * constant return mad_value
815
21.054054
78
py
NeuroKit
NeuroKit-master/neurokit2/stats/fit_error.py
# -*- coding: utf-8 -*- import numpy as np def fit_error(y, y_predicted, n_parameters=2): """**Calculate the fit error for a model** Also specific and direct access functions can be used, such as :func:`.fit_mse`, :func:`.fit_rmse` and :func:`.fit_r2`. Parameters ---------- y : Union[list, np.array, pd.Series] The response variable (the y axis). y_predicted : Union[list, np.array, pd.Series] The fitted data generated by a model. n_parameters : int Number of model parameters (for the degrees of freedom used in R2). Returns ------- dict A dictionary containing different indices of fit error. See Also -------- fit_mse, fit_rmse, fit_r2 Examples -------- .. ipython:: python import neurokit2 as nk y = np.array([-1.0, -0.5, 0, 0.5, 1]) y_predicted = np.array([0.0, 0, 0, 0, 0]) # Master function x = nk.fit_error(y, y_predicted) x # Direct access nk.fit_mse(y, y_predicted) nk.fit_rmse(y, y_predicted) nk.fit_r2(y, y_predicted, adjusted=False) nk.fit_r2(y, y_predicted, adjusted=True, n_parameters=2) """ # Get information SSE, n, df = _fit_error_prepare(y, y_predicted, n_parameters) # Mean squared error MSE = SSE / n # Root mean squared error RMSE = np.sqrt(SSE / n) # Adjusted r-squared # For optimization use 1 - adjR2 since we want to minimize the function SST = np.std(y) * n # Get R2 if SST == 0: R2 = 1 else: R2 = SSE / SST # R2 adjusted R2_adjusted = 1 - (1 - (1 - R2)) * (n - 1) / df return {"SSE": SSE, "MSE": MSE, "RMSE": RMSE, "R2": R2, "R2_adjusted": R2_adjusted} # ============================================================================= # Direct accessors # ============================================================================= def fit_mse(y, y_predicted): """Compute Mean Square Error (MSE).""" return fit_error(y, y_predicted)["MSE"] def fit_rmse(y, y_predicted): """Compute Root Mean Square Error (RMSE).""" return fit_error(y, y_predicted)["RMSE"] def fit_r2(y, y_predicted, adjusted=True, n_parameters=2): """Compute R2.""" if adjusted is True: return fit_error(y, y_predicted, n_parameters=n_parameters)["R2_adjusted"] return fit_error(y, y_predicted, n_parameters=n_parameters)["R2"] # ============================================================================= # Internals # ============================================================================= def _fit_error_prepare(y, y_predicted, n_parameters=2): # n, i.e., how many observations (signal length) n = len(y) # Sanitize if n != len(y_predicted): raise TypeError("NeuroKit error: fit_error(): 'y' and 'y_predicted' are not of the same length.") # Residual, i.e. the difference between data and model residual = y - y_predicted # Degrees of freedom, i.e., number of observations (length of signal) minus number of parameters df = n - n_parameters # Calculate sum of squared errors SSE = np.sum(residual ** 2) return SSE, n, df
3,199
25.666667
105
py
NeuroKit
NeuroKit-master/neurokit2/stats/__init__.py
"""Submodule for NeuroKit.""" from .cluster import cluster from .cluster_findnumber import cluster_findnumber from .cluster_quality import cluster_quality from .correlation import cor from .density import density from .density_bandwidth import density_bandwidth from .distance import distance from .fit_error import fit_error, fit_mse, fit_r2, fit_rmse from .fit_loess import fit_loess from .fit_mixture import fit_mixture from .fit_polynomial import fit_polynomial, fit_polynomial_findorder from .hdi import hdi from .mad import mad from .rescale import rescale from .standardize import standardize from .summary import summary_plot __all__ = [ "standardize", "hdi", "mad", "cor", "density", "density_bandwidth", "distance", "rescale", "fit_loess", "fit_polynomial", "fit_polynomial_findorder", "fit_mixture", "fit_error", "fit_mse", "fit_rmse", "fit_r2", "summary_plot", "cluster", "cluster_quality", "cluster_findnumber", ]
1,009
23.047619
68
py
NeuroKit
NeuroKit-master/neurokit2/stats/distance.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd import scipy import scipy.spatial from .standardize import standardize def distance(X=None, method="mahalanobis"): """**Distance** Compute distance using different metrics. Parameters ---------- X : array or DataFrame A dataframe of values. method : str The method to use. One of ``"mahalanobis"`` or ``"mean"`` for the average distance from the mean. Returns ------- array Vector containing the distance values. Examples --------- .. ipython:: python import neurokit2 as nk # Load the iris dataset data = nk.data("iris").drop("Species", axis=1) data["Distance"] = nk.distance(data, method="mahalanobis") @savefig p_distance1.png scale=100% fig = data.plot(x="Petal.Length", y="Petal.Width", s="Distance", c="Distance", kind="scatter") @suppress plt.close() .. ipython:: python data["DistanceZ"] = np.abs(nk.distance(data.drop("Distance", axis=1), method="mean")) @savefig p_distance2.png scale=100% fig = data.plot(x="Petal.Length", y="Sepal.Length", s="DistanceZ", c="DistanceZ", kind="scatter") @suppress plt.close() """ if not isinstance(X, pd.DataFrame): X = pd.DataFrame(X) method = method.lower() # remove capitalised letters if method in ["mahalanobis"]: dist = _distance_mahalanobis(X) elif method in ["mean", "center", "average"]: dist = _distance_mean(X) else: raise ValueError("NeuroKit error: distance(): 'method' should be one of 'mahalanobis'.") return dist # ============================================================================= # Methods # ============================================================================= def _distance_mahalanobis(X=None): cov = X.cov().values cov = scipy.linalg.inv(cov) col_means = X.mean().values dist = np.full(len(X), np.nan) for i in range(len(X)): dist[i] = scipy.spatial.distance.mahalanobis(X.iloc[i, :].values, col_means, cov) ** 2 return dist def _distance_mean(X=None): Z = standardize(X) dist = Z.mean(axis=1).values return dist
2,251
24.590909
103
py
NeuroKit
NeuroKit-master/neurokit2/stats/summary.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import scipy.stats as st from .density import density from .rescale import rescale def summary_plot(x, errorbars=0, **kwargs): """**Descriptive plot** Visualize a distribution with density, histogram, boxplot and rugs plots all at once. Examples -------- .. ipython:: python import neurokit2 as nk import numpy as np x = np.random.normal(size=100) @savefig p_summary1.png scale=100% fig = nk.summary_plot(x) @suppress plt.close() """ if "ax" in kwargs: fig = None ax = kwargs.get("ax") kwargs.pop("ax") else: fig, ax = plt.subplots() # Histogram counts, bins = np.histogram(x, **kwargs) bin_centers = 0.5*(bins[1:] + bins[:-1]) if errorbars > 0: samperr = np.std(counts) / np.sqrt(counts) * (st.norm.ppf(1-(1-errorbars)/2)) ax.errorbar(bin_centers, counts, yerr=samperr, ecolor="#FF8C00", fmt='.', capsize=5, capthick=2) ax.hist(bins[:-1], bins, weights=counts, color="#2196F3", edgecolor="white", zorder=1, **kwargs) # Density x_axis, y_axis = density(x, **kwargs) y_axis = rescale(y_axis, to=[0, np.max(counts)]) ax.plot(x_axis, y_axis, color="#E91E63", linewidth=1.5, zorder=2, **kwargs) # Points y_axis = np.full(len(x), 0.1) ax.scatter(x, y_axis, c="black", alpha=0.5, marker="|", zorder=3, **kwargs) # Boxplot ax.boxplot( x, vert=False, positions=[np.max(counts) / 10], widths=np.max(counts) / 10, manage_ticks=False, boxprops=dict(linewidth=1.5), medianprops=dict(linewidth=1.5), whiskerprops=dict(linewidth=1.5), capprops=dict(linewidth=1.5), zorder=4, **kwargs ) return fig
1,845
25.753623
104
py
NeuroKit
NeuroKit-master/neurokit2/stats/fit_mixture.py
# -*- coding: utf-8 -*- import pandas as pd import sklearn.mixture def fit_mixture(X=None, n_clusters=2): """**Gaussian Mixture Model** Performs a polynomial regression of given order. Parameters ---------- X : Union[list, np.array, pd.Series] The values to classify. n_clusters : int Number of components to look for. Returns ------- pd.DataFrame DataFrame containing the probability of belonging to each cluster. dict Dictionary containing additional information such as the parameters (:func:`.n_clusters`). See Also ---------- signal_detrend, fit_error Examples --------- .. ipython:: python import pandas as pd import neurokit2 as nk x = nk.signal_simulate() probs, info = nk.fit_mixture(x, n_clusters=2) # Rmb to merge with main to return ``info`` @savefig p_fit_mixture.png scale=100% fig = nk.signal_plot([x, probs["Cluster_0"], probs["Cluster_1"]], standardize=True) @suppress plt.close() """ if X.ndim == 1: X = X.reshape(-1, 1) # fit a Gaussian Mixture Model with two components clf = sklearn.mixture.GaussianMixture(n_components=n_clusters, random_state=333) clf = clf.fit(X) # Get predicted probabilities predicted = clf.predict_proba(X) probabilities = pd.DataFrame(predicted).add_prefix("Cluster_") return probabilities, {"n_clusters": n_clusters}
1,465
25.178571
98
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_eventrelated.py
# -*- coding: utf-8 -*- from ..epochs.eventrelated_utils import ( _eventrelated_addinfo, _eventrelated_rate, _eventrelated_sanitizeinput, _eventrelated_sanitizeoutput, ) def ppg_eventrelated(epochs, silent=False): """**Performs event-related PPG analysis on epochs** Parameters ---------- epochs : Union[dict, pd.DataFrame] A dict containing one DataFrame per event/trial, usually obtained via :func:`.epochs_create`, or a DataFrame containing all epochs, usually obtained via :func:`.epochs_to_df`. silent : bool If ``True``, silence possible warnings. Returns ------- DataFrame A dataframe containing the analyzed PPG features for each epoch, with each epoch indicated by the `Label` column (if not present, by the `Index` column). The analyzed features consist of the following: * ``"PPG_Rate_Baseline"``: the baseline heart rate (at stimulus onset). * ``"PPG_Rate_Max"``: the maximum heart rate after stimulus onset. * ``"PPG_Rate_Min"``: the minimum heart rate after stimulus onset. * ``"PPG_Rate_Mean"``: the mean heart rate after stimulus onset. * ``"PPG_Rate_SD"``: the standard deviation of the heart rate after stimulus onset. * ``"PPG_Rate_Max_Time"``: the time at which maximum heart rate occurs. * ``"PPG_Rate_Min_Time"``: the time at which minimum heart rate occurs. We also include the following *experimental* features related to the parameters of a quadratic model: * ``"PPG_Rate_Trend_Linear"``: The parameter corresponding to the linear trend. * ``"PPG_Rate_Trend_Quadratic"``: The parameter corresponding to the curvature. * ``"PPG_Rate_Trend_R2"``: the quality of the quadratic model. If too low, the parameters might not be reliable or meaningful. See Also -------- events_find, epochs_create, ppg_process Examples ---------- .. ipython:: python import neurokit2 as nk # Example with simulated data ppg, info = nk.ppg_process(nk.ppg_simulate(duration=20)) # Process the data epochs = nk.epochs_create(ppg, events=[5000, 10000, 15000], epochs_start=-0.1, epochs_end=1.9) nk.ppg_eventrelated(epochs) """ # Sanity checks epochs = _eventrelated_sanitizeinput(epochs, what="ppg", silent=silent) # Extract features and build dataframe data = {} # Initialize an empty dict for i in epochs.keys(): data[i] = {} # Initialize empty container # Rate data[i] = _eventrelated_rate(epochs[i], data[i], var="PPG_Rate") # Fill with more info data[i] = _eventrelated_addinfo(epochs[i], data[i]) # Return dataframe return _eventrelated_sanitizeoutput(data)
2,862
31.168539
98
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_process.py
# -*- coding: utf-8 -*- import pandas as pd from ..misc import as_vector from ..misc.report import create_report from ..signal import signal_rate from ..signal.signal_formatpeaks import _signal_from_indices from .ppg_clean import ppg_clean from .ppg_findpeaks import ppg_findpeaks from .ppg_methods import ppg_methods from .ppg_plot import ppg_plot def ppg_process( ppg_signal, sampling_rate=1000, method="elgendi", report=None, **kwargs ): """**Process a photoplethysmogram (PPG) signal** Convenience function that automatically processes a photoplethysmogram signal. Parameters ---------- ppg_signal : Union[list, np.array, pd.Series] The raw PPG channel. sampling_rate : int The sampling frequency of :func:`.ppg_signal` (in Hz, i.e., samples/second). method : str The processing pipeline to apply. Can be one of ``"elgendi"``. Defaults to ``"elgendi"``. report : str The filename of a report containing description and figures of processing (e.g. ``"myreport.html"``). Needs to be supplied if a report file should be generated. Defaults to ``None``. Can also be ``"text"`` to just print the text in the console without saving anything. **kwargs Other arguments to be passed to specific methods. For more information, see :func:`.ppg_methods`. Returns ------- signals : DataFrame A DataFrame of same length as :func:`.ppg_signal` containing the following columns: * ``"PPG_Raw"``: the raw signal. * ``"PPG_Clean"``: the cleaned signal. * ``"PPG_Rate"``: the heart rate as measured based on PPG peaks. * ``"PPG_Peaks"``: the PPG peaks marked as "1" in a list of zeros. info : dict A dictionary containing the information of peaks and the signals' sampling rate. See Also -------- ppg_clean, ppg_findpeaks Examples -------- .. ipython:: python import neurokit2 as nk ppg = nk.ppg_simulate(duration=10, sampling_rate=1000, heart_rate=70) signals, info = nk.ppg_process(ppg, sampling_rate=1000) @savefig p_ppg_process1.png scale=100% nk.ppg_plot(signals) @suppress plt.close() """ # Sanitize input ppg_signal = as_vector(ppg_signal) methods = ppg_methods(sampling_rate=sampling_rate, method=method, **kwargs) # Clean signal ppg_cleaned = ppg_clean( ppg_signal, sampling_rate=sampling_rate, method=methods["method_cleaning"], **methods["kwargs_cleaning"] ) # Find peaks info = ppg_findpeaks( ppg_cleaned, sampling_rate=sampling_rate, method=methods["method_peaks"], **methods["kwargs_peaks"] ) info["sampling_rate"] = sampling_rate # Add sampling rate in dict info # Mark peaks peaks_signal = _signal_from_indices( info["PPG_Peaks"], desired_length=len(ppg_cleaned) ) # Rate computation rate = signal_rate( info["PPG_Peaks"], sampling_rate=sampling_rate, desired_length=len(ppg_cleaned) ) # Prepare output signals = pd.DataFrame( { "PPG_Raw": ppg_signal, "PPG_Clean": ppg_cleaned, "PPG_Rate": rate, "PPG_Peaks": peaks_signal, } ) if report is not None: # Generate report containing description and figures of processing if ".html" in str(report): fig = ppg_plot(signals, sampling_rate=sampling_rate) else: fig = None create_report(file=report, signals=signals, info=methods, fig=fig) return signals, info
3,664
29.289256
91
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_intervalrelated.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd from ..hrv import hrv def ppg_intervalrelated(data, sampling_rate=1000): """**Performs PPG analysis on longer periods of data (typically > 10 seconds), such as resting-state data** Parameters ---------- data : Union[dict, pd.DataFrame] A DataFrame containing the different processed signal(s) as different columns, typically generated by :func:`.ppg_process`. Can also take a dict containing sets of separately processed DataFrames. sampling_rate : int The sampling frequency of the signal (in Hz, i.e., samples/second). Returns ------- DataFrame A dataframe containing the analyzed PPG features. The analyzed features consist of the following: * ``"PPG_Rate_Mean"``: the mean heart rate. * ``"HRV"``: the different heart rate variability metrices. See :func:`.hrv` docstrings for details. See Also -------- ppg_process Examples ---------- .. ipython:: python import neurokit2 as nk # Download data data = nk.data("bio_resting_5min_100hz") # Process the data df, info = nk.ppg_process(data["PPG"], sampling_rate=100) # Single dataframe is passed nk.ppg_intervalrelated(df, sampling_rate=100) epochs = nk.epochs_create(df, events=[0, 15000], sampling_rate=100, epochs_end=150) nk.ppg_intervalrelated(epochs) """ intervals = {} # Format input if isinstance(data, pd.DataFrame): rate_cols = [col for col in data.columns if "PPG_Rate" in col] if len(rate_cols) == 1: intervals.update(_ppg_intervalrelated_formatinput(data)) intervals.update(*_ppg_intervalrelated_hrv(data, sampling_rate)) else: raise ValueError( "NeuroKit error: ppg_intervalrelated(): Wrong input," "we couldn't extract heart rate. Please make sure" "your DataFrame contains a `PPG_Rate` column." ) ppg_intervals = pd.DataFrame.from_dict(intervals, orient="index").T elif isinstance(data, dict): for epoch in data: intervals[epoch] = {} # Initialize empty container # Add label info intervals[epoch]["Label"] = data[epoch]["Label"].iloc[0] # Rate intervals[epoch].update(_ppg_intervalrelated_formatinput(data[epoch])) # HRV intervals[epoch].update( *_ppg_intervalrelated_hrv(data[epoch], sampling_rate) ) ppg_intervals = pd.DataFrame.from_dict(intervals, orient="index") return ppg_intervals # ============================================================================= # Internals # ============================================================================= def _ppg_intervalrelated_formatinput(data): # Sanitize input colnames = data.columns.values if "PPG_Rate" not in colnames: raise ValueError( "NeuroKit error: ppg_intervalrelated(): Wrong input," "we couldn't extract heart rate. Please make sure" "your DataFrame contains a `PPG_Rate` column." ) signal = data["PPG_Rate"].values PPG_Rate_Mean = np.mean(signal) return {"PPG_Rate_Mean": PPG_Rate_Mean} def _ppg_intervalrelated_hrv(data, sampling_rate): # Sanitize input colnames = data.columns.values if "PPG_Peaks" not in colnames: raise ValueError( "NeuroKit error: ppg_intervalrelated(): Wrong input," "we couldn't extract peaks. Please make sure" "your DataFrame contains a `PPG_Peaks` column." ) # Transform rpeaks from "signal" format to "info" format. peaks = np.where(data["PPG_Peaks"].values)[0] peaks = {"PPG_Peaks": peaks} results = hrv(peaks, sampling_rate=sampling_rate) return results.astype("float").to_dict("records")
4,011
29.165414
105
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_plot.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import pandas as pd def ppg_plot(ppg_signals, sampling_rate=None, static=True): """**Visualize photoplethysmogram (PPG) data** Visualize the PPG signal processing. Parameters ---------- ppg_signals : DataFrame DataFrame obtained from :func:`.ppg_process`. sampling_rate : int The sampling frequency of the PPG (in Hz, i.e., samples/second). Needs to be supplied if the data should be plotted over time in seconds. Otherwise the data is plotted over samples. Defaults to ``None``. static : bool If True, a static plot will be generated with matplotlib. If False, an interactive plot will be generated with plotly. Defaults to True. Returns ------- fig Figure representing a plot of the processed PPG signals. Examples -------- .. ipython:: python import neurokit2 as nk # Simulate data ppg = nk.ppg_simulate(duration=10, sampling_rate=1000, heart_rate=70) # Process signal signals, info = nk.ppg_process(ppg, sampling_rate=1000) # Plot @savefig p_ppg_plot1.png scale=100% nk.ppg_plot(signals) @suppress plt.close() See Also -------- ppg_process """ # Sanity-check input. if not isinstance(ppg_signals, pd.DataFrame): raise ValueError( "NeuroKit error: The `ppg_signals` argument must" " be the DataFrame returned by `ppg_process()`." ) # X-axis if sampling_rate is not None: x_axis = np.linspace(0, ppg_signals.shape[0] / sampling_rate, ppg_signals.shape[0]) else: x_axis = np.arange(0, ppg_signals.shape[0]) # Get peak indices peaks = np.where(ppg_signals["PPG_Peaks"] == 1)[0] if static: # Prepare figure fig, (ax0, ax1) = plt.subplots(nrows=2, ncols=1, sharex=True) if sampling_rate is not None: ax0.set_xlabel("Time (seconds)") ax1.set_xlabel("Time (seconds)") elif sampling_rate is None: ax0.set_xlabel("Samples") ax1.set_xlabel("Samples") fig.suptitle("Photoplethysmogram (PPG)", fontweight="bold") plt.tight_layout(h_pad=0.4) # Plot cleaned and raw PPG ax0.set_title("Raw and Cleaned Signal") ax0.plot(x_axis, ppg_signals["PPG_Raw"], color="#B0BEC5", label="Raw", zorder=1) ax0.plot( x_axis, ppg_signals["PPG_Clean"], color="#FB1CF0", label="Cleaned", zorder=1, linewidth=1.5, ) # Plot peaks ax0.scatter( x_axis[peaks], ppg_signals["PPG_Clean"][peaks], color="#D60574", label="Peaks", zorder=2, ) ax0.legend(loc="upper right") # Rate ax1.set_title("Heart Rate") ppg_rate_mean = ppg_signals["PPG_Rate"].mean() ax1.plot( x_axis, ppg_signals["PPG_Rate"], color="#FB661C", label="Rate", linewidth=1.5, ) ax1.axhline(y=ppg_rate_mean, label="Mean", linestyle="--", color="#FBB41C") ax1.legend(loc="upper right") return fig else: try: import plotly.graph_objects as go from plotly.subplots import make_subplots except ImportError as e: raise ImportError( "NeuroKit error: ppg_plot(): the 'plotly'", " module is required when 'static' is False.", " Please install it first (`pip install plotly`).", ) from e fig = make_subplots( rows=2, cols=1, shared_xaxes=True, subplot_titles=("Raw and Cleaned Signal", "Rate"), ) # Plot cleaned and raw PPG fig.add_trace(go.Scatter(x=x_axis, y=ppg_signals["PPG_Raw"], name="Raw"), row=1, col=1) fig.add_trace( go.Scatter( x=x_axis, y=ppg_signals["PPG_Clean"], name="Cleaned", marker_color="#FB1CF0", ), row=1, col=1, ) # Plot peaks fig.add_trace( go.Scatter( x=x_axis[peaks], y=ppg_signals["PPG_Clean"][peaks], name="Peaks", mode="markers", marker_color="#D60574", ), row=1, col=1, ) # Rate ppg_rate_mean = ppg_signals["PPG_Rate"].mean() fig.add_trace( go.Scatter( x=x_axis, y=ppg_signals["PPG_Rate"], name="Rate", mode="lines", marker_color="#FB661C", ), row=2, col=1, ) fig.add_hline( y=ppg_rate_mean, line_dash="dash", line_color="#FBB41C", name="Mean", row=2, col=1, ) fig.update_layout(title_text="Photoplethysmogram (PPG)", height=500, width=750) if sampling_rate is not None: fig.update_xaxes(title_text="Time (seconds)", row=1, col=1) fig.update_xaxes(title_text="Time (seconds)", row=2, col=1) elif sampling_rate is None: fig.update_xaxes(title_text="Samples", row=1, col=1) fig.update_xaxes(title_text="Samples", row=2, col=1) return fig
5,587
28.723404
100
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_clean.py
# -*- coding: utf-8 -*- from warnings import warn import numpy as np from ..misc import NeuroKitWarning, as_vector from ..signal import signal_fillmissing, signal_filter def ppg_clean(ppg_signal, sampling_rate=1000, heart_rate=None, method="elgendi"): """**Clean a photoplethysmogram (PPG) signal** Prepare a raw PPG signal for systolic peak detection. Parameters ---------- ppg_signal : Union[list, np.array, pd.Series] The raw PPG channel. heart_rate : Union[int, float] The heart rate of the PPG signal. Applicable only if method is ``"nabian2018"`` to check that filter frequency is appropriate. sampling_rate : int The sampling frequency of the PPG (in Hz, i.e., samples/second). The default is 1000. method : str The processing pipeline to apply. Can be one of ``"elgendi"``, ``"nabian2018"``, or ``"none"``. The default is ``"elgendi"``. If ``"none"`` is passed, the raw signal will be returned without any cleaning. Returns ------- clean : array A vector containing the cleaned PPG. See Also -------- ppg_simulate, ppg_findpeaks Examples -------- .. ipython:: python import neurokit2 as nk import pandas as pd import matplotlib.pyplot as plt # Simulate and clean signal ppg = nk.ppg_simulate(heart_rate=75, duration=30) ppg_elgendi = nk.ppg_clean(ppg, method='elgendi') ppg_nabian = nk.ppg_clean(ppg, method='nabian2018', heart_rate=75) # Plot and compare methods signals = pd.DataFrame({'PPG_Raw' : ppg, 'PPG_Elgendi' : ppg_elgendi, 'PPG_Nabian' : ppg_nabian}) @savefig p_ppg_clean1.png scale=100% signals.plot() @suppress plt.close() References ---------- * Nabian, M., Yin, Y., Wormwood, J., Quigley, K. S., Barrett, L. F., & Ostadabbas, S. (2018). An open-source feature extraction tool for the analysis of peripheral physiological data. IEEE Journal of Translational Engineering in Health and Medicine, 6, 1-11. """ ppg_signal = as_vector(ppg_signal) # Missing data n_missing = np.sum(np.isnan(ppg_signal)) if n_missing > 0: warn( "There are " + str(n_missing) + " missing data points in your signal." " Filling missing values using `signal_fillmissing`.", category=NeuroKitWarning, ) ppg_signal = signal_fillmissing(ppg_signal, method="both") method = str(method).lower() if method in ["elgendi"]: clean = _ppg_clean_elgendi(ppg_signal, sampling_rate) elif method in ["nabian2018"]: clean = _ppg_clean_nabian2018(ppg_signal, sampling_rate, heart_rate=heart_rate) elif method in ["none"]: clean = ppg_signal else: raise ValueError( "`method` not found. Must be one of 'elgendi', 'nabian2018', or 'none'." ) return clean # ============================================================================= # Methods # ============================================================================= def _ppg_clean_elgendi(ppg_signal, sampling_rate): filtered = signal_filter( ppg_signal, sampling_rate=sampling_rate, lowcut=0.5, highcut=8, order=3, method="butterworth", ) return filtered def _ppg_clean_nabian2018(ppg_signal, sampling_rate, heart_rate=None): """Low-pass filter for continuous BP signal preprocessing, adapted from Nabian et al. (2018).""" # Determine low-pass filter value highcut = 40 # Convert heart rate to seconds, check if low-pass filter within appropriate range if heart_rate is not None: heart_rate = heart_rate / 60 if not highcut >= 10 * heart_rate and not highcut < 0.5 * sampling_rate: raise ValueError( "Highcut value should be at least 10 times heart rate and" " less than 0.5 times sampling rate." ) filtered = signal_filter( ppg_signal, sampling_rate=sampling_rate, lowcut=None, highcut=highcut, order=2, method="butterworth", ) return filtered
4,274
29.978261
103
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_findpeaks.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import scipy.signal from ..signal import signal_smooth def ppg_findpeaks(ppg_cleaned, sampling_rate=1000, method="elgendi", show=False, **kwargs): """**Find systolic peaks in a photoplethysmogram (PPG) signal** Parameters ---------- ppg_cleaned : Union[list, np.array, pd.Series] The cleaned PPG channel as returned by :func:`.ppg_clean`. sampling_rate : int The sampling frequency of the PPG (in Hz, i.e., samples/second). The default is 1000. method : str The processing pipeline to apply. Can be one of ``"elgendi"``, ``"bishop"``. The default is ``"elgendi"``. show : bool If ``True``, returns a plot of the thresholds used during peak detection. Useful for debugging. The default is ``False``. Returns ------- info : dict A dictionary containing additional information, in this case the samples at which systolic peaks occur, accessible with the key ``"PPG_Peaks"``. See Also -------- ppg_simulate, ppg_clean Examples -------- .. ipython:: python import neurokit2 as nk import matplotlib.pyplot as plt ppg = nk.ppg_simulate(heart_rate=75, duration=20, sampling_rate=50) ppg_clean = nk.ppg_clean(ppg, sampling_rate=50) @savefig p_ppg_findpeaks1.png scale=100% peaks = nk.ppg_findpeaks(ppg_clean, sampling_rate=100, show=True) @suppress plt.close() # Method by Bishop et al., (2018) @savefig p_ppg_findpeaks2.png scale=100% peaks = nk.ppg_findpeaks(ppg, method="bishop", show=True) @suppress plt.close() References ---------- * Elgendi, M., Norton, I., Brearley, M., Abbott, D., & Schuurmans, D. (2013). Systolic peak detection in acceleration photoplethysmograms measured from emergency responders in tropical conditions. PloS one, 8(10), e76585. * Bishop, S. M., & Ercole, A. (2018). Multi-scale peak and trough detection optimised for periodic and quasi-periodic neuroscience data. In Intracranial Pressure & Neuromonitoring XVI (pp. 189-195). Springer International Publishing. """ method = method.lower() if method in ["elgendi"]: peaks = _ppg_findpeaks_elgendi(ppg_cleaned, sampling_rate, show=show, **kwargs) elif method in ["msptd", "bishop2018", "bishop"]: peaks, _ = _ppg_findpeaks_bishop(ppg_cleaned, show=show, **kwargs) else: raise ValueError("`method` not found. Must be one of the following: 'elgendi', 'bishop'.") # Prepare output. info = {"PPG_Peaks": peaks} return info def _ppg_findpeaks_elgendi( signal, sampling_rate=1000, peakwindow=0.111, beatwindow=0.667, beatoffset=0.02, mindelay=0.3, show=False, ): """Implementation of Elgendi M, Norton I, Brearley M, Abbott D, Schuurmans D (2013) Systolic Peak Detection in Acceleration Photoplethysmograms Measured from Emergency Responders in Tropical Conditions. PLoS ONE 8(10): e76585. doi:10.1371/journal.pone.0076585. All tune-able parameters are specified as keyword arguments. `signal` must be the bandpass-filtered raw PPG with a lowcut of .5 Hz, a highcut of 8 Hz. """ if show: _, (ax0, ax1) = plt.subplots(nrows=2, ncols=1, sharex=True) ax0.plot(signal, label="filtered") # Ignore the samples with negative amplitudes and square the samples with # values larger than zero. signal_abs = signal.copy() signal_abs[signal_abs < 0] = 0 sqrd = signal_abs**2 # Compute the thresholds for peak detection. Call with show=True in order # to visualize thresholds. ma_peak_kernel = int(np.rint(peakwindow * sampling_rate)) ma_peak = signal_smooth(sqrd, kernel="boxcar", size=ma_peak_kernel) ma_beat_kernel = int(np.rint(beatwindow * sampling_rate)) ma_beat = signal_smooth(sqrd, kernel="boxcar", size=ma_beat_kernel) thr1 = ma_beat + beatoffset * np.mean(sqrd) # threshold 1 if show: ax1.plot(sqrd, label="squared") ax1.plot(thr1, label="threshold") ax1.legend(loc="upper right") # Identify start and end of PPG waves. waves = ma_peak > thr1 beg_waves = np.where(np.logical_and(np.logical_not(waves[0:-1]), waves[1:]))[0] end_waves = np.where(np.logical_and(waves[0:-1], np.logical_not(waves[1:])))[0] # Throw out wave-ends that precede first wave-start. end_waves = end_waves[end_waves > beg_waves[0]] # Identify systolic peaks within waves (ignore waves that are too short). num_waves = min(beg_waves.size, end_waves.size) min_len = int(np.rint(peakwindow * sampling_rate)) # this is threshold 2 in the paper min_delay = int(np.rint(mindelay * sampling_rate)) peaks = [0] for i in range(num_waves): beg = beg_waves[i] end = end_waves[i] len_wave = end - beg if len_wave < min_len: continue # Visualize wave span. if show: ax1.axvspan(beg, end, facecolor="m", alpha=0.5) # Find local maxima and their prominence within wave span. data = signal[beg:end] locmax, props = scipy.signal.find_peaks(data, prominence=(None, None)) if locmax.size > 0: # Identify most prominent local maximum. peak = beg + locmax[np.argmax(props["prominences"])] # Enforce minimum delay between peaks. if peak - peaks[-1] > min_delay: peaks.append(peak) peaks.pop(0) if show: ax0.scatter(peaks, signal_abs[peaks], c="r") ax0.legend(loc="upper right") ax0.set_title("PPG Peaks (Method by Elgendi et al., 2013)") peaks = np.asarray(peaks).astype(int) return peaks def _ppg_findpeaks_bishop( signal, show=False, ): """Implementation of Bishop SM, Ercole A (2018) Multi-scale peak and trough detection optimised for periodic and quasi-periodic neuroscience data. doi:10.1007/978-3-319-65798-1_39. Currently designed for short signals of relatively low sampling frequencies (e.g. 6 seconds at 100 Hz). Also, the function currently only returns peaks, but it does identify pulse onsets too. """ # TODO: create ppg_peaks() that also returns onsets and stuff # Setup N = len(signal) L = int(np.ceil(N / 2) - 1) # Step 1: calculate local maxima and local minima scalograms # - detrend: this removes the best-fit straight line x = scipy.signal.detrend(signal, type="linear") # - initialise LMS matrices m_max = np.full((L, N), False) m_min = np.full((L, N), False) # - populate LMS matrices for k in range(1, L): # scalogram scales for i in range(k + 2, N - k + 1): if x[i - 1] > x[i - k - 1] and x[i - 1] > x[i + k - 1]: m_max[k - 1, i - 1] = True if x[i - 1] < x[i - k - 1] and x[i - 1] < x[i + k - 1]: m_min[k - 1, i - 1] = True # Step 2: find the scale with the most local maxima (or local minima) # - row-wise summation (i.e. sum each row) gamma_max = np.sum(m_max, axis=1) # the "axis=1" option makes it row-wise gamma_min = np.sum(m_min, axis=1) # - find scale with the most local maxima (or local minima) lambda_max = np.argmax(gamma_max) lambda_min = np.argmax(gamma_min) # Step 3: Use lambda to remove all elements of m for which k>lambda m_max = m_max[: (lambda_max + 1), :] m_min = m_min[: (lambda_min + 1), :] # Step 4: Find peaks (and onsets) # - column-wise summation m_max_sum = np.sum(m_max == False, axis=0) m_min_sum = np.sum(m_min == False, axis=0) peaks = np.where(m_max_sum == 0)[0].astype(int) onsets = np.where(m_min_sum == 0)[0].astype(int) if show: _, ax0 = plt.subplots(nrows=1, ncols=1, sharex=True) ax0.plot(signal, label="signal") ax0.scatter(peaks, signal[peaks], c="r") ax0.scatter(onsets, signal[onsets], c="b") ax0.set_title("PPG Peaks (Method by Bishop et al., 2018)") return peaks, onsets
8,147
33.820513
119
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_methods.py
# -*- coding: utf-8 -*- import numpy as np from ..misc.report import get_kwargs from .ppg_clean import ppg_clean from .ppg_findpeaks import ppg_findpeaks def ppg_methods( sampling_rate=1000, method="elgendi", method_cleaning="default", method_peaks="default", **kwargs, ): """**PPG Preprocessing Methods** This function analyzes and specifies the methods used in the preprocessing, and create a textual description of the methods used. It is used by :func:`ppg_process()` to dispatch the correct methods to each subroutine of the pipeline and :func:`ppg_report()` to create a preprocessing report. Parameters ---------- sampling_rate : int The sampling frequency of the raw PPG signal (in Hz, i.e., samples/second). method : str The method used for cleaning and peak finding if ``"method_cleaning"`` and ``"method_peaks"`` are set to ``"default"``. Can be one of ``"elgendi"``. Defaults to ``"elgendi"``. method_cleaning: str The method used to clean the raw PPG signal. If ``"default"``, will be set to the value of ``"method"``. Defaults to ``"default"``. For more information, see the ``"method"`` argument of :func:`.ppg_clean`. method_peaks: str The method used to find peaks. If ``"default"``, will be set to the value of ``"method"``. Defaults to ``"default"``. For more information, see the ``"method"`` argument of :func:`.ppg_findpeaks`. **kwargs Other arguments to be passed to :func:`.ppg_clean` and :func:`.ppg_findpeaks`. Returns ------- report_info : dict A dictionary containing the keyword arguments passed to the cleaning and peak finding functions, text describing the methods, and the corresponding references. See Also -------- ppg_process, ppg_clean, ppg_findpeaks Examples -------- .. ipython:: python import neurokit2 as nk methods = nk.ppg_methods(sampling_rate=100, method="elgendi", method_cleaning="nabian2018") print(methods["text_cleaning"]) print(methods["references"][0]) """ # Sanitize inputs method_cleaning = ( str(method).lower() if method_cleaning == "default" else str(method_cleaning).lower() ) method_peaks = ( str(method).lower() if method_peaks == "default" else str(method_peaks).lower() ) # Create dictionary with all inputs report_info = { "sampling_rate": sampling_rate, "method": method, "method_cleaning": method_cleaning, "method_peaks": method_peaks, **kwargs, } # Get arguments to be passed to cleaning and peak finding functions kwargs_cleaning, report_info = get_kwargs(report_info, ppg_clean) kwargs_peaks, report_info = get_kwargs(report_info, ppg_findpeaks) # Save keyword arguments in dictionary report_info["kwargs_cleaning"] = kwargs_cleaning report_info["kwargs_peaks"] = kwargs_peaks # Initialize refs list with NeuroKit2 reference refs = ["""Makowski, D., Pham, T., Lau, Z. J., Brammer, J. C., Lespinasse, F., Pham, H., Schölzel, C., & Chen, S. A. (2021). NeuroKit2: A Python toolbox for neurophysiological signal processing. Behavior Research Methods, 53(4), 1689–1696. https://doi.org/10.3758/s13428-020-01516-y """] # 1. Cleaning # ------------ report_info["text_cleaning"] = f"The raw signal, sampled at {sampling_rate} Hz," if method_cleaning in [ "elgendi", "elgendi2013", ]: report_info["text_cleaning"] += ( " was preprocessed using a bandpass filter ([0.5 - 8 Hz], Butterworth 3rd order;" + " following Elgendi et al., 2013)." ) refs.append( """Elgendi M, Norton I, Brearley M, Abbott D, Schuurmans D (2013) Systolic Peak Detection in Acceleration Photoplethysmograms Measured from Emergency Responders in Tropical Conditions PLoS ONE 8(10): e76585. doi:10.1371/journal.pone.0076585.""" ) elif method_cleaning in ["nabian", "nabian2018"]: if report_info["heart_rate"] is None: cutoff = "of 40 Hz" else: cutoff = f' based on the heart rate of {report_info["heart_rate"]} bpm' report_info["text_cleaning"] = ( f" was preprocessed using a lowpass filter (with a cutoff frequency {cutoff}," + " butterworth 2nd order; following Nabian et al., 2018)." ) refs.append( """Nabian, M., Yin, Y., Wormwood, J., Quigley, K. S., Barrett, L. F., & Ostadabbas, S.(2018). An open-source feature extraction tool for the analysis of peripheral physiological data. IEEE Journal of Translational Engineering in Health and Medicine, 6, 1-11.""" ) elif method_cleaning in ["none"]: report_info[ "text_cleaning" ] += " was directly used for peak detection without preprocessing." else: # just in case more methods are added report_info["text_cleaning"] = ( "was cleaned following the " + method + " method." ) # 2. Peaks # ---------- if method_peaks in ["elgendi", "elgendi13"]: report_info[ "text_peaks" ] = "The peak detection was carried out using the method described in Elgendi et al. (2013)." refs.append( """Elgendi M, Norton I, Brearley M, Abbott D, Schuurmans D (2013) Systolic Peak Detection in Acceleration Photoplethysmograms Measured from Emergency Responders in Tropical Conditions PLoS ONE 8(10): e76585. doi:10.1371/journal.pone.0076585.""" ) elif method_peaks in ["none"]: report_info["text_peaks"] = "There was no peak detection carried out." else: report_info[ "text_peaks" ] = f"The peak detection was carried out using the method {method_peaks}." report_info["references"] = list(np.unique(refs)) return report_info
6,124
36.576687
109
py
NeuroKit
NeuroKit-master/neurokit2/ppg/__init__.py
"""Submodule for NeuroKit.""" # Aliases from ..signal import signal_rate as ppg_rate from .ppg_analyze import ppg_analyze from .ppg_clean import ppg_clean from .ppg_eventrelated import ppg_eventrelated from .ppg_findpeaks import ppg_findpeaks from .ppg_intervalrelated import ppg_intervalrelated from .ppg_methods import ppg_methods from .ppg_plot import ppg_plot from .ppg_process import ppg_process from .ppg_simulate import ppg_simulate __all__ = [ "ppg_simulate", "ppg_clean", "ppg_findpeaks", "ppg_rate", "ppg_process", "ppg_plot", "ppg_methods", "ppg_intervalrelated", "ppg_eventrelated", "ppg_analyze", ]
654
23.259259
52
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_simulate.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np from ..misc import check_random_state, check_random_state_children from ..signal import signal_distort, signal_interpolate def ppg_simulate( duration=120, sampling_rate=1000, heart_rate=70, frequency_modulation=0.2, ibi_randomness=0.1, drift=0, motion_amplitude=0.1, powerline_amplitude=0.01, burst_number=0, burst_amplitude=1, random_state=None, random_state_distort="spawn", show=False, ): """**Simulate a photoplethysmogram (PPG) signal** Phenomenological approximation of PPG. The PPG wave is described with four landmarks: wave onset, location of the systolic peak, location of the dicrotic notch and location of the diastolic peaks. These landmarks are defined as x and y coordinates (in a time series). These coordinates are then interpolated at the desired sampling rate to obtain the PPG signal. Parameters ---------- duration : int Desired recording length in seconds. The default is 120. sampling_rate : int The desired sampling rate (in Hz, i.e., samples/second). The default is 1000. heart_rate : int Desired simulated heart rate (in beats per minute). The default is 70. Note that for the ECGSYN method, random fluctuations are to be expected to mimic a real heart rate. These fluctuations can cause some slight discrepancies between the requested heart rate and the empirical heart rate, especially for shorter signals. frequency_modulation : float Float between 0 and 1. Determines how pronounced respiratory sinus arrythmia (RSA) is (0 corresponds to absence of RSA). The default is 0.3. ibi_randomness : float Float between 0 and 1. Determines how much random noise there is in the duration of each PPG wave (0 corresponds to absence of variation). The default is 0.1. drift : float Float between 0 and 1. Determines how pronounced the baseline drift (.05 Hz) is (0 corresponds to absence of baseline drift). The default is 1. motion_amplitude : float Float between 0 and 1. Determines how pronounced the motion artifact (0.5 Hz) is (0 corresponds to absence of motion artifact). The default is 0.1. powerline_amplitude : float Float between 0 and 1. Determines how pronounced the powerline artifact (50 Hz) is (0 corresponds to absence of powerline artifact). Note that powerline_amplitude > 0 is only possible if ``sampling_rate`` is >= 500. The default is 0.1. burst_amplitude : float Float between 0 and 1. Determines how pronounced high frequency burst artifacts are (0 corresponds to absence of bursts). The default is 1. burst_number : int Determines how many high frequency burst artifacts occur. The default is 0. show : bool If ``True``, returns a plot of the landmarks and interpolated PPG. Useful for debugging. random_state : None, int, numpy.random.RandomState or numpy.random.Generator Seed for the random number generator. See for ``misc.check_random_state`` for further information. random_state_distort : {'legacy', 'spawn'}, None, int, numpy.random.RandomState or numpy.random.Generator Random state to be used to distort the signal. If ``"legacy"``, use the same random state used to generate the signal (discouraged as it creates dependent random streams). If ``"spawn"``, spawn independent children random number generators from the random_state argument. If any of the other types, generate independent children random number generators from the random_state_distort provided (this allows generating multiple version of the same signal distorted by different random noise realizations). Returns ------- ppg : array A vector containing the PPG. See Also -------- ecg_simulate, rsp_simulate, eda_simulate, emg_simulate Examples -------- .. ipython:: python import neurokit2 as nk ppg = nk.ppg_simulate(duration=40, sampling_rate=500, heart_rate=75, random_state=42) """ # Seed the random generator for reproducible results rng = check_random_state(random_state) random_state_distort = check_random_state_children(random_state, random_state_distort, n_children=4) # At the requested sampling rate, how long is a period at the requested # heart-rate and how often does that period fit into the requested # duration? period = 60 / heart_rate # in seconds n_period = int(np.floor(duration / period)) periods = np.ones(n_period) * period # Seconds at which waves begin. x_onset = np.cumsum(periods) x_onset -= x_onset[0] # make sure seconds start at zero # Add respiratory sinus arrythmia (frequency modulation). periods, x_onset = _frequency_modulation( periods, x_onset, modulation_frequency=0.05, modulation_strength=frequency_modulation, ) # Randomly modulate duration of waves by subracting a random value between # 0 and ibi_randomness% of the wave duration (see function definition). x_onset = _random_x_offset(x_onset, ibi_randomness, rng) # Corresponding signal amplitudes. y_onset = rng.normal(0, 0.1, n_period) # Seconds at which the systolic peaks occur within the waves. x_sys = x_onset + rng.normal(0.175, 0.01, n_period) * periods # Corresponding signal amplitudes. y_sys = y_onset + rng.normal(1.5, 0.15, n_period) # Seconds at which the dicrotic notches occur within the waves. x_notch = x_onset + rng.normal(0.4, 0.001, n_period) * periods # Corresponding signal amplitudes (percentage of systolic peak height). y_notch = y_sys * rng.normal(0.49, 0.01, n_period) # Seconds at which the diastolic peaks occur within the waves. x_dia = x_onset + rng.normal(0.45, 0.001, n_period) * periods # Corresponding signal amplitudes (percentage of systolic peak height). y_dia = y_sys * rng.normal(0.51, 0.01, n_period) x_all = np.concatenate((x_onset, x_sys, x_notch, x_dia)) x_all.sort(kind="mergesort") x_all = np.ceil(x_all * sampling_rate).astype(int) # convert seconds to samples y_all = np.zeros(n_period * 4) y_all[0::4] = y_onset y_all[1::4] = y_sys y_all[2::4] = y_notch y_all[3::4] = y_dia if show: __, (ax0, ax1) = plt.subplots(nrows=2, ncols=1, sharex=True) ax0.scatter(x_all, y_all, c="r") # Interpolate a continuous signal between the landmarks (i.e., Cartesian # coordinates). samples = np.arange(int(np.ceil(duration * sampling_rate))) ppg = signal_interpolate(x_values=x_all, y_values=y_all, x_new=samples, method="akima") # Remove NAN (values outside interpolation range, i.e., after last sample). ppg[np.isnan(ppg)] = np.nanmean(ppg) if show: ax0.plot(ppg) # Add baseline drift. if drift > 0: drift_freq = 0.05 if drift_freq < (1 / duration) * 2: drift_freq = (1 / duration) * 2 ppg = signal_distort( ppg, sampling_rate=sampling_rate, noise_amplitude=drift, noise_frequency=drift_freq, random_state=random_state_distort[0], silent=True, ) # Add motion artifacts. if motion_amplitude > 0: motion_freq = 0.5 ppg = signal_distort( ppg, sampling_rate=sampling_rate, noise_amplitude=motion_amplitude, noise_frequency=motion_freq, random_state=random_state_distort[1], silent=True, ) # Add high frequency bursts. if burst_amplitude > 0: ppg = signal_distort( ppg, sampling_rate=sampling_rate, artifacts_amplitude=burst_amplitude, artifacts_frequency=100, artifacts_number=burst_number, random_state=random_state_distort[2], silent=True, ) # Add powerline noise. if powerline_amplitude > 0: ppg = signal_distort( ppg, sampling_rate=sampling_rate, powerline_amplitude=powerline_amplitude, powerline_frequency=50, random_state=random_state_distort[3], silent=True, ) if show: ax1.plot(ppg) return ppg def _frequency_modulation(periods, seconds, modulation_frequency, modulation_strength): """modulator_frequency determines the frequency at which respiratory sinus arrhythmia occurs (in Hz). modulator_strength must be between 0 and 1. """ modulation_mean = 1 # Enforce minimum inter-beat-interval of 300 milliseconds. if (modulation_mean - modulation_strength) * periods[ 0 ] < 0.3: # elements in periods all have the same value at this point print( "Skipping frequency modulation, since the modulation_strength" f" {modulation_strength} leads to physiologically implausible" f" wave durations of {((modulation_mean - modulation_strength) * periods[0]) * 1000}" f" milliseconds." ) return periods, seconds # Apply a very conservative Nyquist criterion. nyquist = (1 / periods[0]) * 0.1 if modulation_frequency > nyquist: print(f"Please choose a modulation frequency lower than {nyquist}.") # Generate a sine with mean 1 and amplitude 0.5 * modulation_strength, that is, # ranging from 1 - 0.5 * modulation_strength to 1 + 0.5 * modulation_strength. # For example, at a heart rate of 100 and modulation_strenght=1, the heart rate will # fluctuate between 150 and 50. At the default modulatiom_strenght=.2, it will # fluctuate between 110 and 90. modulator = ( 0.5 * modulation_strength * np.sin(2 * np.pi * modulation_frequency * seconds) + modulation_mean ) periods_modulated = periods * modulator seconds_modulated = np.cumsum(periods_modulated) seconds_modulated -= seconds_modulated[0] # make sure seconds start at zero return periods_modulated, seconds_modulated def _random_x_offset(x, offset_weight, rng): """From each wave onset xi subtract offset_weight * (xi - xi-1) where xi-1 is the wave onset preceding xi. offset_weight must be between 0 and 1. """ # Sanitize offset to min 0 and max .99 offset_weight = min(offset_weight, 0.99) offset_weight = max(offset_weight, 0) x_diff = np.diff(x) # Enforce minimum inter-beat-interval of 300 milliseconds. min_x_diff = min(x_diff) if (min_x_diff - (min_x_diff * offset_weight)) < 0.3: print( "Skipping random IBI modulation, since the offset_weight" f" {offset_weight} leads to physiologically implausible wave" f" durations of {(min_x_diff - (min_x_diff * offset_weight)) * 1000}" f" milliseconds." ) return x max_offsets = offset_weight * x_diff offsets = [rng.uniform(0, i) for i in max_offsets] x_offset = x.copy() x_offset[1:] -= offsets return x_offset def _amplitude_modulation(): # TODO pass
11,257
38.780919
112
py
NeuroKit
NeuroKit-master/neurokit2/ppg/ppg_analyze.py
# -*- coding: utf-8 -*- import pandas as pd from .ppg_eventrelated import ppg_eventrelated from .ppg_intervalrelated import ppg_intervalrelated def ppg_analyze(data, sampling_rate=1000, method="auto"): """**Photoplethysmography (PPG) Analysis**. Performs PPG analysis on either epochs (event-related analysis) or on longer periods of data such as resting-state data. Parameters ---------- data : Union[dict, pd.DataFrame] A dictionary of epochs, containing one DataFrame per epoch, usually obtained via :func:`.epochs_create`, or a DataFrame containing all epochs, usually obtained via :func:`.epochs_to_df`. Can also take a DataFrame of processed signals from a longer period of data, typically generated by :func:`.ppg_process` or :func:`.bio_process`. Can also take a dict containing sets of separate periods of data. sampling_rate : int The sampling frequency of the signal (in Hz, i.e., samples/second). Defaults to 1000Hz. method : str Can be one of ``"event-related"`` for event-related analysis on epochs, or ``"interval-related"`` for analysis on longer periods of data. Defaults to ``"auto"`` where the right method will be chosen based on the mean duration of the data (``"event-related"`` for duration under 10s). Returns ------- DataFrame A dataframe containing the analyzed PPG features. If event-related analysis is conducted, each epoch is indicated by the ``Label`` column. See :func:`.ppg_eventrelated` and :func:`.ppg_intervalrelated` docstrings for details. See Also -------- bio_process, ppg_process, epochs_create, ppg_eventrelated, ppg_intervalrelated Examples ---------- .. ipython:: python import neurokit2 as nk # Example 1: Simulate data for event-related analysis ppg = nk.ppg_simulate(duration=20, sampling_rate=1000) # Process data ppg_signals, info = nk.ppg_process(ppg, sampling_rate=1000) epochs = nk.epochs_create(ppg_signals, events=[5000, 10000, 15000], epochs_start=-0.1, epochs_end=1.9) # Analyze analyze_epochs = nk.ppg_analyze(epochs, sampling_rate=1000) analyze_epochs # Example 2: Download the resting-state data data = nk.data("bio_resting_5min_100hz") # Process the data df, info = nk.ppg_process(data["PPG"], sampling_rate=100) # Analyze analyze_df = nk.ppg_analyze(df, sampling_rate=100) analyze_df """ method = method.lower() # Event-related analysis if method in ["event-related", "event", "epoch"]: # Sanity checks if isinstance(data, dict): for i in data: colnames = data[i].columns.values elif isinstance(data, pd.DataFrame): colnames = data.columns.values if len([i for i in colnames if "Label" in i]) == 0: raise ValueError( "NeuroKit error: ppg_analyze(): Wrong input or method," "we couldn't extract epochs features." ) else: features = ppg_eventrelated(data) # Interval-related analysis elif method in ["interval-related", "interval", "resting-state"]: features = ppg_intervalrelated(data, sampling_rate=sampling_rate) # Auto elif method in ["auto"]: if isinstance(data, dict): for i in data: duration = len(data[i]) / sampling_rate if duration >= 10: features = ppg_intervalrelated(data, sampling_rate=sampling_rate) else: features = ppg_eventrelated(data) if isinstance(data, pd.DataFrame): if "Label" in data.columns: epoch_len = data["Label"].value_counts()[0] duration = epoch_len / sampling_rate else: duration = len(data) / sampling_rate if duration >= 10: features = ppg_intervalrelated(data, sampling_rate=sampling_rate) else: features = ppg_eventrelated(data) return features
4,193
34.542373
99
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_plot.py
# -*- coding: utf-8 -*- import matplotlib.pyplot as plt import numpy as np import pandas as pd def emg_plot(emg_signals, sampling_rate=None, static=True): """**EMG Graph** Visualize electromyography (EMG) data. Parameters ---------- emg_signals : DataFrame DataFrame obtained from ``emg_process()``. sampling_rate : int The sampling frequency of the EMG (in Hz, i.e., samples/second). Needs to be supplied if the data should be plotted over time in seconds. Otherwise the data is plotted over samples. Defaults to ``None``. static : bool If True, a static plot will be generated with matplotlib. If False, an interactive plot will be generated with plotly. Defaults to True. See Also -------- emg_process Returns ------- Though the function returns nothing, the figure can be retrieved and saved as follows: .. code-block:: console # To be run after emg_plot() fig = plt.gcf() fig.savefig("myfig.png") Examples -------- .. ipython:: python import neurokit2 as nk # Simulate data emg = nk.emg_simulate(duration=10, sampling_rate=1000, burst_number=3) # Process signal emg_signals, _ = nk.emg_process(emg, sampling_rate=1000) # Plot @savefig p_emg_plot.png scale=100% nk.emg_plot(emg_signals) @suppress plt.close() """ # Mark onsets, offsets, activity onsets = np.where(emg_signals["EMG_Onsets"] == 1)[0] offsets = np.where(emg_signals["EMG_Offsets"] == 1)[0] # Sanity-check input. if not isinstance(emg_signals, pd.DataFrame): raise ValueError( "NeuroKit error: The `emg_signals` argument must" " be the DataFrame returned by `emg_process()`." ) # Determine what to display on the x-axis, mark activity. if sampling_rate is not None: x_axis = np.linspace(0, emg_signals.shape[0] / sampling_rate, emg_signals.shape[0]) else: x_axis = np.arange(0, emg_signals.shape[0]) if static is True: return _emg_plot_static(emg_signals, x_axis, onsets, offsets, sampling_rate) else: return _emg_plot_interactive(emg_signals, x_axis, onsets, offsets, sampling_rate) # ============================================================================= # Internals # ============================================================================= def _emg_plot_activity(emg_signals, onsets, offsets): activity_signal = pd.Series(np.full(len(emg_signals), np.nan)) activity_signal[onsets] = emg_signals["EMG_Amplitude"][onsets].values activity_signal[offsets] = emg_signals["EMG_Amplitude"][offsets].values activity_signal = activity_signal.fillna(method="backfill") if np.any(activity_signal.isna()): index = np.min(np.where(activity_signal.isna())) - 1 value_to_fill = activity_signal[index] activity_signal = activity_signal.fillna(value_to_fill) return activity_signal def _emg_plot_static(emg_signals, x_axis, onsets, offsets, sampling_rate): # Prepare figure. fig, (ax0, ax1) = plt.subplots(nrows=2, ncols=1, sharex=True) if sampling_rate is not None: ax1.set_xlabel("Time (seconds)") elif sampling_rate is None: ax1.set_xlabel("Samples") fig.suptitle("Electromyography (EMG)", fontweight="bold") plt.tight_layout(h_pad=0.2) # Plot cleaned and raw EMG. ax0.set_title("Raw and Cleaned Signal") ax0.plot(x_axis, emg_signals["EMG_Raw"], color="#B0BEC5", label="Raw", zorder=1) ax0.plot( x_axis, emg_signals["EMG_Clean"], color="#FFC107", label="Cleaned", zorder=1, linewidth=1.5 ) ax0.legend(loc="upper right") # Plot Amplitude. ax1.set_title("Muscle Activation") ax1.plot( x_axis, emg_signals["EMG_Amplitude"], color="#FF9800", label="Amplitude", linewidth=1.5 ) # Shade activity regions. activity_signal = _emg_plot_activity(emg_signals, onsets, offsets) ax1.fill_between( x_axis, emg_signals["EMG_Amplitude"], activity_signal, where=emg_signals["EMG_Amplitude"] > activity_signal, color="#f7c568", alpha=0.5, label=None, ) # Mark onsets and offsets. ax1.scatter( x_axis[onsets], emg_signals["EMG_Amplitude"][onsets], color="#f03e65", label=None, zorder=3 ) ax1.scatter( x_axis[offsets], emg_signals["EMG_Amplitude"][offsets], color="#f03e65", label=None, zorder=3, ) if sampling_rate is not None: onsets = onsets / sampling_rate offsets = offsets / sampling_rate for i, j in zip(list(onsets), list(offsets)): ax1.axvline(i, color="#4a4a4a", linestyle="--", label=None, zorder=2) ax1.axvline(j, color="#4a4a4a", linestyle="--", label=None, zorder=2) ax1.legend(loc="upper right") plt.close() return fig def _emg_plot_interactive(emg_signals, x_axis, onsets, offsets, sampling_rate): try: import plotly.graph_objects as go from plotly.subplots import make_subplots except ImportError: raise ImportError( "NeuroKit error: emg_plot(): the 'plotly' " "module is required for this feature." "Please install it first (`pip install plotly`)." ) # Prepare figure. fig = make_subplots(rows=2, cols=1, shared_xaxes=True) fig.update_layout(title="Electromyography (EMG)", font=dict(size=18), height=600) # Plot cleaned and raw EMG. fig.add_trace( go.Scatter( x=x_axis, y=emg_signals["EMG_Raw"], mode="lines", name="Raw", line=dict(color="#B0BEC5"), ), row=1, col=1, ) fig.add_trace( go.Scatter( x=x_axis, y=emg_signals["EMG_Clean"], mode="lines", name="Cleaned", line=dict(color="#FFC107"), ), row=1, col=1, ) # Plot Amplitude. fig.add_trace( go.Scatter( x=x_axis, y=emg_signals["EMG_Amplitude"], mode="lines", name="Amplitude", line=dict(color="#FF9800"), ), row=2, col=1, ) # Mark onsets and offsets. fig.add_trace( go.Scatter( x=x_axis[onsets], y=emg_signals["EMG_Amplitude"][onsets], mode="markers", name="Onsets", marker=dict(color="#f03e65", size=10), ), row=2, col=1, ) fig.add_trace( go.Scatter( x=x_axis[offsets], y=emg_signals["EMG_Amplitude"][offsets], mode="markers", name="Offsets", marker=dict(color="#f03e65", size=10), ), row=2, col=1, ) if sampling_rate is not None: onsets = onsets / sampling_rate offsets = offsets / sampling_rate fig.update_xaxes(title_text="Time (seconds)", row=2, col=1) elif sampling_rate is None: fig.update_xaxes(title_text="Samples", row=2, col=1) for i, j in zip(list(onsets), list(offsets)): fig.add_shape( type="line", x0=i, y0=0, x1=i, y1=1, line=dict(color="#4a4a4a", width=2, dash="dash"), row=2, col=1, ) fig.add_shape( type="line", x0=j, y0=0, x1=j, y1=1, line=dict(color="#4a4a4a", width=2, dash="dash"), row=2, col=1, ) fig.update_yaxes(title_text="Amplitude", row=2, col=1) return fig
7,786
27.947955
100
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_analyze.py
# -*- coding: utf-8 -*- import pandas as pd from .emg_eventrelated import emg_eventrelated from .emg_intervalrelated import emg_intervalrelated def emg_analyze(data, sampling_rate=1000, method="auto"): """**EMG Analysis** Performs EMG analysis on either epochs (event-related analysis) or on longer periods of data such as resting-state data. Parameters ---------- data : Union[dict, pd.DataFrame] A dictionary of epochs, containing one DataFrame per epoch, usually obtained via ``epochs_create()``, or a DataFrame containing all epochs, usually obtained via ``epochs_to_df()``. Can also take a DataFrame of processed signals from a longer period of data, typically generated by ``emg_process()`` or ``bio_process()``. Can also take a dict containing sets of separate periods of data. sampling_rate : int The sampling frequency of the signal (in Hz, i.e., samples/second). Defaults to 1000Hz. method : str Can be one of ``"event-related"`` for event-related analysis on epochs, or ``"interval-related"`` for analysis on longer periods of data. Defaults to ``auto`` where the right method will be chosen based on the mean duration of the data (``"event-related"`` for duration under 10s). Returns ------- DataFrame A dataframe containing the analyzed EMG features. If event-related analysis is conducted, each epoch is indicated by the `Label` column. See :func:`emg_eventrelated` and :func:`emg_intervalrelated` docstrings for details. See Also -------- .bio_process, emg_process, .epochs_create, emg_eventrelated, emg_intervalrelated Examples ---------- .. ipython:: python import neurokit2 as nk import pandas as pd # Example with simulated data emg = nk.emg_simulate(duration=20, sampling_rate=1000, burst_number=3) emg_signals, info = nk.emg_process(emg, sampling_rate=1000) epochs = nk.epochs_create(emg_signals, events=[3000, 6000, 9000], sampling_rate=1000, epochs_start=-0.1, epochs_end=1.9) # Event-related analysis analyze_epochs = nk.emg_analyze(epochs, method="event-related") analyze_epochs # Interval-related analysis analyze_df = nk.emg_analyze(emg_signals, method="interval-related") analyze_df """ method = method.lower() # Event-related analysis if method in ["event-related", "event", "epoch"]: # Sanity checks if isinstance(data, dict): for i in data: colnames = data[i].columns.values elif isinstance(data, pd.DataFrame): colnames = data.columns.values if len([i for i in colnames if "Label" in i]) == 0: raise ValueError( "NeuroKit error: emg_analyze(): Wrong input or method, we couldn't extract extract epochs features." ) else: features = emg_eventrelated(data) # Interval-related analysis elif method in ["interval-related", "interval", "resting-state"]: features = emg_intervalrelated(data) # Auto elif method in ["auto"]: if isinstance(data, dict): for i in data: duration = len(data[i]) / sampling_rate if duration >= 10: features = emg_intervalrelated(data) else: features = emg_eventrelated(data) if isinstance(data, pd.DataFrame): if "Label" in data.columns: epoch_len = data["Label"].value_counts()[0] duration = epoch_len / sampling_rate else: duration = len(data) / sampling_rate if duration >= 10: features = emg_intervalrelated(data) else: features = emg_eventrelated(data) return features
3,915
35.943396
127
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_eventrelated.py
# -*- coding: utf-8 -*- from warnings import warn import numpy as np from ..epochs.eventrelated_utils import ( _eventrelated_addinfo, _eventrelated_sanitizeinput, _eventrelated_sanitizeoutput, ) from ..misc import NeuroKitWarning def emg_eventrelated(epochs, silent=False): """**Event-related EMG Analysis** Performs event-related EMG analysis on epochs. Parameters ---------- epochs : Union[dict, pd.DataFrame] A dict containing one DataFrame per event/trial, usually obtained via ``epochs_create()``, or a DataFrame containing all epochs, usually obtained via ``epochs_to_df()``. silent : bool If ``True``, silence possible warnings. Returns ------- DataFrame A dataframe containing the analyzed EMG features for each epoch, with each epoch indicated by the `Label` column (if not present, by the `Index` column). The analyzed features consist of the following: * ``"EMG_Activation*``: indication of whether there is muscular activation following the onset of the event (1 if present, 0 if absent) and if so, its corresponding amplitude features and the number of activations in each epoch. If there is no activation, nans are displayed for the below features. * ``"EMG_Amplitude_Mean*``: the mean amplitude of the activity. * ``"EMG_Amplitude_Max*``: the maximum amplitude of the activity. * ``"EMG_Amplitude_SD*``: the standard deviation of the activity amplitude. * ``"EMG_Amplitude_Max_Time*``: the time of maximum amplitude. * ``"EMG_Bursts*``: the number of activations, or bursts of activity, within each epoch. See Also -------- emg_simulate, emg_process, .events_find, .epochs_create Examples ---------- .. ipython:: python import neurokit2 as nk # Example with simulated data emg = nk.emg_simulate(duration=20, sampling_rate=1000, burst_number=3) emg_signals, info = nk.emg_process(emg, sampling_rate=1000) epochs = nk.epochs_create(emg_signals, events=[3000, 6000, 9000], sampling_rate=1000, epochs_start=-0.1,epochs_end=1.9) nk.emg_eventrelated(epochs) """ # Sanity checks epochs = _eventrelated_sanitizeinput(epochs, what="emg", silent=silent) # Extract features and build dataframe data = {} # Initialize an empty dict for i in epochs.keys(): data[i] = {} # Initialize an empty dict for the current epoch # Activation following event if "EMG_Onsets" not in epochs[i]: warn( "Input does not have an `EMG_Onsets` column." " Unable to process EMG features.", category=NeuroKitWarning, ) data[i]["EMG_Activation"] = 0 elif np.any(epochs[i]["EMG_Onsets"][epochs[i].index > 0] != 0): data[i]["EMG_Activation"] = 1 else: data[i]["EMG_Activation"] = 0 # Analyze features based on activation if data[i]["EMG_Activation"] == 1: data[i] = _emg_eventrelated_features(epochs[i], data[i]) else: data[i]["EMG_Amplitude_Mean"] = np.nan data[i]["EMG_Amplitude_Max"] = np.nan data[i]["EMG_Amplitude_SD"] = np.nan data[i]["EMG_Amplitude_Max_Time"] = np.nan data[i]["EMG_Bursts"] = np.nan # Fill with more info data[i] = _eventrelated_addinfo(epochs[i], data[i]) df = _eventrelated_sanitizeoutput(data) return df # ============================================================================= # Internals # ============================================================================= def _emg_eventrelated_features(epoch, output={}): # Sanitize input if "EMG_Activity" not in epoch or "EMG_Amplitude" not in epoch: warn( "Input does not have an `EMG_Activity` column or `EMG_Amplitude` column." " Will skip computation of EMG amplitudes.", category=NeuroKitWarning, ) return output # Peak amplitude and Time of peak activations = len(np.where(epoch["EMG_Onsets"][epoch.index > 0] == 1)[0]) activated_signal = np.where(epoch["EMG_Activity"][epoch.index > 0] == 1) mean = np.array(epoch["EMG_Amplitude"][epoch.index > 0].iloc[activated_signal]).mean() maximum = np.array(epoch["EMG_Amplitude"][epoch.index > 0].iloc[activated_signal]).max() index_time = np.where(epoch["EMG_Amplitude"][epoch.index > 0] == maximum)[0] time = np.array(epoch["EMG_Amplitude"][epoch.index > 0].index[index_time])[0] output["EMG_Amplitude_Mean"] = mean output["EMG_Amplitude_Max"] = maximum output["EMG_Amplitude_SD"] = np.std( epoch["EMG_Amplitude"][epoch.index > 0].iloc[activated_signal] ) output["EMG_Amplitude_Max_Time"] = time output["EMG_Bursts"] = activations return output
4,995
36.56391
100
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_amplitude.py
# -*- coding: utf-8 -*- import numpy as np from ..signal import signal_filter def emg_amplitude(emg_cleaned): """**Compute electromyography (EMG) amplitude** Compute electromyography amplitude given the cleaned respiration signal, done by calculating the linear envelope of the signal. Parameters ---------- emg_cleaned : Union[list, np.array, pd.Series] The cleaned electromyography channel as returned by ``emg_clean()``. Returns ------- array A vector containing the electromyography amplitude. See Also -------- emg_clean, emg_rate, emg_process, emg_plot Examples -------- .. ipython:: python import neurokit2 as nk import pandas as pd emg = nk.emg_simulate(duration=10, sampling_rate=1000, burst_number=3) cleaned = nk.emg_clean(emg, sampling_rate=1000) amplitude = nk.emg_amplitude(cleaned) @savefig p_emg_amplitude1.png scale=100% fig = pd.DataFrame({"EMG": emg, "Amplitude": amplitude}).plot(subplots=True) @suppress plt.close() """ tkeo = _emg_amplitude_tkeo(emg_cleaned) amplitude = _emg_amplitude_envelope(tkeo) return amplitude # ============================================================================= # Taeger-Kaiser Energy Operator # ============================================================================= def _emg_amplitude_tkeo(emg_cleaned): """Calculates the Teager–Kaiser Energy operator to improve onset detection, described by Marcos Duarte at https://github.com/demotu/BMC/blob/master/notebooks/Electromyography.ipynb. Parameters ---------- emg_cleaned : Union[list, np.array, pd.Series] The cleaned electromyography channel as returned by `emg_clean()`. Returns ------- tkeo : array The emg signal processed by the Teager–Kaiser Energy operator. References ---------- - BMCLab: https://github.com/demotu/BMC/blob/master/notebooks/Electromyography.ipynb - Li, X., Zhou, P., & Aruin, A. S. (2007). Teager–Kaiser energy operation of surface EMG improves muscle activity onset detection. Annals of biomedical engineering, 35(9), 1532-1538. """ tkeo = emg_cleaned.copy() # Teager–Kaiser Energy operator tkeo[1:-1] = emg_cleaned[1:-1] * emg_cleaned[1:-1] - emg_cleaned[:-2] * emg_cleaned[2:] # Correct the data in the extremities tkeo[0], tkeo[-1] = tkeo[1], tkeo[-2] return tkeo # ============================================================================= # Linear Envelope # ============================================================================= def _emg_amplitude_envelope( emg_cleaned, sampling_rate=1000, lowcut=10, highcut=400, envelope_filter=8 ): """Calculate the linear envelope of a signal. This function implements a 2nd-order Butterworth filter with zero lag, described by Marcos Duarte at <https://github.com/demotu/BMC/blob/master/notebooks/Electromyography.ipynb>. Parameters ---------- emg_cleaned : Union[list, np.array, pd.Series] The cleaned electromyography channel as returned by `emg_clean()`. sampling_rate : int The sampling frequency of `emg_signal` (in Hz, i.e., samples/second). lowcut : float Low-cut frequency for the band-pass filter (in Hz). Defaults to 10Hz. highcut : float High-cut frequency for the band-pass filter (in Hz). Defaults to 400Hz. envelope_filter : float Cuttoff frequency for the high-pass filter (in Hz). Defauts to 8Hz. Returns ------- envelope : array The linear envelope of the emg signal. References ---------- - BMCLab: https://github.com/demotu/BMC/blob/master/notebooks/Electromyography.ipynb """ filtered = signal_filter( emg_cleaned, sampling_rate=sampling_rate, lowcut=lowcut, highcut=highcut, method="butterworth", order=2, ) envelope = np.abs(filtered) envelope = signal_filter( envelope, sampling_rate=sampling_rate, lowcut=None, highcut=envelope_filter, method="butterworth", order=2, ) return envelope
4,219
29.142857
109
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_methods.py
# -*- coding: utf-8 -*- import numpy as np from ..misc.report import get_kwargs from .emg_activation import emg_activation def emg_methods( sampling_rate=1000, method_cleaning="biosppy", method_activation="threshold", **kwargs, ): """**EMG Preprocessing Methods** This function analyzes and specifies the methods used in the preprocessing, and create a textual description of the methods used. It is used by :func:`eda_process()` to dispatch the correct methods to each subroutine of the pipeline and to create a preprocessing report. Parameters ---------- sampling_rate : int The sampling frequency of the raw EMG signal (in Hz, i.e., samples/second). method_cleaning : str The method used for cleaning the raw EMG signal. Can be one of ``"biosppy"`` or ``"none"``. Defaults to ``"biosppy"``. If ``"none"`` is passed, the raw signal will be used without any cleaning. method_activation: str The method used for locating EMG activity. Defaults to ``"threshold"``. For more information, see the ``"method"`` argument of :func:`.emg_activation`. **kwargs Other arguments to be passed to :func:`.emg_activation`, Returns ------- report_info : dict A dictionary containing the keyword arguments passed to the cleaning and activation functions, text describing the methods, and the corresponding references. See Also -------- """ # Sanitize inputs method_cleaning = str(method_cleaning).lower() method_activation = str(method_activation).lower() # Create dictionary with all inputs report_info = { "sampling_rate": sampling_rate, "method_cleaning": method_cleaning, "method_activation": method_activation, **kwargs, } # Get arguments to be passed to activation function kwargs_activation, report_info = get_kwargs(report_info, emg_activation) # Save keyword arguments in dictionary report_info["kwargs_activation"] = kwargs_activation # Initialize refs list with NeuroKit2 reference refs = [ """Makowski, D., Pham, T., Lau, Z. J., Brammer, J. C., Lespinasse, F., Pham, H., Schölzel, C., & Chen, S. A. (2021). NeuroKit2: A Python toolbox for neurophysiological signal processing. Behavior Research Methods, 53(4), 1689–1696. https://doi.org/10.3758/s13428-020-01516-y """ ] # 1. Cleaning # ------------ # If no cleaning report_info["text_cleaning"] = f"The raw signal, sampled at {sampling_rate} Hz," if method_cleaning in ["none"]: report_info["text_cleaning"] += " was directly used without any cleaning." else: report_info["text_cleaning"] += ( " was cleaned using the " + method_cleaning + " method." ) # 2. Activation # ------------- report_info["text_activation"] = ( "EMG activity was detected using the " + method_activation + " method. " ) if method_activation in ["silva"]: if str(report_info["threshold"]) == "default": threshold_str = "0.05" else: threshold_str = str(report_info["threshold"]) report_info["text_activation"] += f"""The threshold was {threshold_str}. """ refs.append( """Silva H, Scherer R, Sousa J, Londral A , "Towards improving the ssability of electromyographic interfacess", Journal of Oral Rehabilitation, pp. 1-2, 2012.""" ) if method_activation in ["mixture"]: report_info[ "text_activation" ] += """A Gaussian mixture model was used to discriminate between activity and baseline. """ if str(report_info["threshold"]) == "default": threshold_str = "0.33" else: threshold_str = str(report_info["threshold"]) report_info[ "text_activation" ] += f"""The minimum probability required to be considered as activated was {threshold_str}. """ elif method_activation in ["threshold"]: report_info[ "text_activation" ] += """The signal was considered as activated when the amplitude exceeded a threshold. """ if str(report_info["threshold"]) == "default": threshold_str = "one tenth of the standard deviation of emg_amplitude" else: threshold_str = str(report_info["threshold"]) report_info[ "text_activation" ] += f"""The minimum amplitude to detect as onset was set to {threshold_str}.""" elif method_activation in ["biosppy"]: if str(report_info["threshold"]) == "default": threshold_str = "1.2 times of the mean of the absolute of the smoothed, full-wave-rectified signal" else: threshold_str = str(report_info["threshold"]) report_info[ "text_activation" ] += f"""The threshold was set to {threshold_str}.""" # 3. References # ------------- report_info["references"] = list(np.unique(refs)) return report_info
5,092
36.448529
111
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_activation.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd from ..events import events_find from ..misc import as_vector from ..signal import (signal_binarize, signal_changepoints, signal_formatpeaks, signal_smooth) def emg_activation( emg_amplitude=None, emg_cleaned=None, sampling_rate=1000, method="threshold", threshold="default", duration_min="default", size=None, threshold_size=None, **kwargs, ): """**Locate EMG Activity** Detects onset in EMG signal based on the amplitude threshold. Parameters ---------- emg_amplitude : array At least one EMG-related signal. Either the amplitude of the EMG signal, obtained from ``emg_amplitude()`` for methods like ``"threshold"`` or ``"mixture"``), and / or the cleaned EMG signal (for methods like ``"pelt"``, ``"biosppy"`` or ``"silva"``). emg_cleaned : array At least one EMG-related signal. Either the amplitude of the EMG signal, obtained from ``emg_amplitude()`` for methods like ``"threshold"`` or ``"mixture"``), and / or the cleaned EMG signal (for methods like ``"pelt"``, ``"biosppy"`` or ``"silva"``). sampling_rate : int The sampling frequency of ``emg_signal`` (in Hz, i.e., samples/second). method : str The algorithm used to discriminate between activity and baseline. Can be one of ``"mixture"`` (default) or ``"threshold"``. If ``"mixture"``, will use a Gaussian Mixture Model to categorize between the two states. If ``"threshold"``, will consider as activated all points which amplitude is superior to the threshold. Can also be ``"pelt"`` or ``"biosppy"`` or ``"silva"``. threshold : str If ``method`` is ``"mixture"``, then it corresponds to the minimum probability required to be considered as activated (default to 0.33). If ``method`` is ``"threshold"``, then it corresponds to the minimum amplitude to detect as onset i.e., defaults to one tenth of the standard deviation of ``emg_amplitude``. If ``method`` is ``"silva"``, defaults to 0.05. If ``method`` is ``"biosppy"``, defaults to 1.2 times of the mean of the absolute of the smoothed, full-wave-rectified signal. If ``method`` is ``"pelt"``, threshold defaults to ``None`` as changepoints are used as a basis for detection. duration_min : float The minimum duration of a period of activity or non-activity in seconds. If ``default``, will be set to 0.05 (50 ms). size: float or int Detection window size (seconds). Applicable only if ``method`` is ``"biosppy"`` or ``"silva"``. If ``None``, defaults to 0.05 for ``"biosppy"`` and 20 for ``"silva"``. threshold_size : int Window size for calculation of the adaptive threshold. Must be bigger than the detection window size. Applicable only if ``method`` is ``"silva``". If ``None``, defaults to 22. kwargs : optional Other arguments. Returns ------- info : dict A dictionary containing additional information, in this case the samples at which the onsets, offsets, and periods of activations of the EMG signal occur, accessible with the key ``"EMG_Onsets"``, ``"EMG_Offsets"``, and ``"EMG_Activity"`` respectively. activity_signal : DataFrame A DataFrame of same length as the input signal in which occurences of onsets, offsets, and activity (above the threshold) of the EMG signal are marked as "1" in lists of zeros with the same length as ``emg_amplitude``. Accessible with the keys ``"EMG_Onsets"``, ``"EMG_Offsets"``, and ``"EMG_Activity"`` respectively. See Also -------- emg_simulate, emg_clean, emg_amplitude, emg_process, emg_plot Examples -------- .. ipython:: python import neurokit2 as nk # Simulate signal and obtain amplitude emg = nk.emg_simulate(duration=10, burst_number=3) emg_cleaned = nk.emg_clean(emg) emg_amplitude = nk.emg_amplitude(emg_cleaned) * **Example 1:** Threshold method .. ipython:: python activity, info = nk.emg_activation(emg_amplitude=emg_amplitude, method="threshold") @savefig p_emg_activation1.png scale=100% nk.events_plot([info["EMG_Offsets"], info["EMG_Onsets"]], emg_cleaned) @suppress plt.close() * **Example 2:** Pelt method .. ipython:: python activity, info = nk.emg_activation(emg_cleaned=emg_cleaned, method="pelt") @savefig p_emg_activation2.png scale=100% nk.events_plot([info["EMG_Offsets"], info["EMG_Onsets"]], emg_cleaned) @suppress plt.close() * **Example 3:** Biosppy method .. ipython:: python activity, info = nk.emg_activation(emg_cleaned=emg_cleaned, method="biosppy") @savefig p_emg_activation3.png scale=100% nk.events_plot([info["EMG_Offsets"], info["EMG_Onsets"]], emg_cleaned) @suppress plt.close() * **Example 4:** Silva method .. ipython:: python activity, info = nk.emg_activation(emg_cleaned=emg_cleaned, method="silva") @savefig p_emg_activation4.png scale=100% nk.events_plot([info["EMG_Offsets"], info["EMG_Onsets"]], emg_cleaned) @suppress plt.close() References ---------- * Silva H, Scherer R, Sousa J, Londral A , "Towards improving the ssability of electromyographic interfacess", Journal of Oral Rehabilitation, pp. 1-2, 2012. """ # Sanity checks. if emg_amplitude is not None: emg_amplitude = as_vector(emg_amplitude) if emg_cleaned is not None: emg_cleaned = as_vector(emg_cleaned) if emg_amplitude is None: emg_amplitude = as_vector(emg_cleaned) if duration_min == "default": duration_min = int(0.05 * sampling_rate) # Find offsets and onsets. method = method.lower() # remove capitalised letters if method == "threshold": if emg_amplitude is None: raise ValueError( "NeuroKit error: emg_activation(): 'threshold' method needs 'emg_amplitude' signal to be passed." ) activity = _emg_activation_threshold(emg_amplitude, threshold=threshold) elif method == "mixture": if emg_amplitude is None: raise ValueError( "NeuroKit error: emg_activation(): 'mixture' method needs 'emg_amplitude' signal to be passed." ) activity = _emg_activation_mixture(emg_amplitude, threshold=threshold) elif method == "pelt": if emg_cleaned is None: raise ValueError( "NeuroKit error: emg_activation(): 'pelt' method needs 'emg_cleaned' (cleaned or raw EMG) signal to " "be passed." ) activity = _emg_activation_pelt(emg_cleaned, duration_min=duration_min, **kwargs) elif method == "biosppy": if emg_cleaned is None: raise ValueError( "NeuroKit error: emg_activation(): 'biosppy' method needs 'emg_cleaned' (cleaned EMG) " "signal to be passed." ) if size is None: size = 0.05 activity = _emg_activation_biosppy( emg_cleaned, sampling_rate=sampling_rate, size=size, threshold=threshold ) elif method == "silva": if emg_cleaned is None: raise ValueError( "NeuroKit error: emg_activation(): 'silva' method needs 'emg_cleaned' (cleaned EMG) " "signal to be passed." ) if size is None: size = 20 if threshold_size is None: threshold_size = 22 activity = _emg_activation_silva( emg_cleaned, size=size, threshold=threshold, threshold_size=threshold_size ) else: raise ValueError( "NeuroKit error: emg_activation(): 'method' should be one of 'mixture', 'threshold', 'pelt' or 'biosppy'." ) # Sanitize activity. info = _emg_activation_activations(activity, duration_min=duration_min) # Prepare Output. df_activity = signal_formatpeaks( {"EMG_Activity": info["EMG_Activity"]}, desired_length=len(emg_amplitude), peak_indices=info["EMG_Activity"], ) df_onsets = signal_formatpeaks( {"EMG_Onsets": info["EMG_Onsets"]}, desired_length=len(emg_amplitude), peak_indices=info["EMG_Onsets"], ) df_offsets = signal_formatpeaks( {"EMG_Offsets": info["EMG_Offsets"]}, desired_length=len(emg_amplitude), peak_indices=info["EMG_Offsets"], ) # Modify output produced by signal_formatpeaks. for x in range(len(emg_amplitude)): if df_activity["EMG_Activity"][x] != 0: if df_activity.index[x] == df_activity.index.get_loc(x): df_activity["EMG_Activity"][x] = 1 else: df_activity["EMG_Activity"][x] = 0 if df_offsets["EMG_Offsets"][x] != 0: if df_offsets.index[x] == df_offsets.index.get_loc(x): df_offsets["EMG_Offsets"][x] = 1 else: df_offsets["EMG_Offsets"][x] = 0 activity_signal = pd.concat([df_activity, df_onsets, df_offsets], axis=1) return activity_signal, info # ============================================================================= # Methods # ============================================================================= def _emg_activation_threshold(emg_amplitude, threshold="default"): if threshold == "default": threshold = (1 / 10) * np.std(emg_amplitude) if threshold > np.max(emg_amplitude): raise ValueError( "NeuroKit error: emg_activation(): the threshold specified exceeds the maximum of the signal" "amplitude." ) activity = signal_binarize(emg_amplitude, method="threshold", threshold=threshold) return activity def _emg_activation_mixture(emg_amplitude, threshold="default"): if threshold == "default": threshold = 0.33 activity = signal_binarize(emg_amplitude, method="mixture", threshold=threshold) return activity def _emg_activation_pelt(emg_cleaned, threshold="default", duration_min=0.05, **kwargs): if threshold == "default": threshold = None # Get changepoints changepoints = signal_changepoints(emg_cleaned, change="var", show=False, **kwargs) # Add first point if changepoints[0] != 0: changepoints = np.append(0, changepoints) # Sanitize lengths = np.append(0, np.diff(changepoints)) changepoints = changepoints[1:][lengths[1:] > duration_min] # reèAdd first point if changepoints[0] != 0: changepoints = np.append(0, changepoints) binary = np.full(len(emg_cleaned), np.nan) binary[changepoints[0::2]] = 0 binary[changepoints[1::2]] = 1 activity = pd.Series(binary).fillna(method="ffill").values # Label as 1 to parts that have the larger SD (likely to be activations) if emg_cleaned[activity == 1].std() > emg_cleaned[activity == 0].std(): activity = np.abs(activity - 1) activity[0] = 0 activity[-1] = 0 return activity def _emg_activation_biosppy(emg_cleaned, sampling_rate=1000, size=0.05, threshold="default"): """Adapted from `find_onsets` in Biosppy.""" # check inputs if emg_cleaned is None: raise TypeError("Please specify an input signal.") # full-wave rectification fwlo = np.abs(emg_cleaned) # smooth size = int(sampling_rate * size) mvgav = signal_smooth(fwlo, method="convolution", kernel="boxzen", size=size) # threshold if threshold == "default": aux = np.abs(mvgav) threshold = 1.2 * np.mean(aux) # find onsets # length = len(signal) # start = np.nonzero(mvgav > threshold)[0] # stop = np.nonzero(mvgav <= threshold)[0] # onsets = np.union1d(np.intersect1d(start - 1, stop), # np.intersect1d(start + 1, stop)) # if np.any(onsets): # if onsets[-1] >= length: # onsets[-1] = length - 1 activity = signal_binarize(mvgav, method="threshold", threshold=threshold) return activity def _emg_activation_silva(emg_cleaned, size=20, threshold_size=22, threshold="default"): """Follows the approach by Silva et al. 2012, adapted from `Biosppy`.""" if threshold_size <= size: raise ValueError( "NeuroKit error: emg_activation(): The window size for calculation of the " "adaptive threshold must be bigger than the detection window size." ) if threshold == "default": threshold = 0.05 # subtract baseline offset signal_zero_mean = emg_cleaned - np.mean(emg_cleaned) # full-wave rectification fwlo = np.abs(signal_zero_mean) # moving average for calculating the test function tf_mvgav = np.convolve(fwlo, np.ones((size,)) / size, mode="valid") # moving average for calculating the adaptive threshold threshold_mvgav = np.convolve(fwlo, np.ones((threshold_size,)) / threshold_size, mode="valid") onset_time_list = [] offset_time_list = [] onset = False for k in range(0, len(threshold_mvgav)): if onset is True: # an onset was previously detected, look for offset time if tf_mvgav[k] < threshold_mvgav[k] and tf_mvgav[k] < threshold: offset_time_list.append(k) onset = False # the offset has been detected, and we can look for another activation else: # we only look for another onset if a previous offset was detected if tf_mvgav[k] >= threshold_mvgav[k] and tf_mvgav[k] >= threshold: onset_time_list.append(k) onset = True onsets = np.union1d(onset_time_list, offset_time_list) # adjust indices because of moving average onsets += int(size / 2) binary = np.full(len(emg_cleaned), np.nan) binary[onsets[0::2]] = 0 binary[onsets[1::2]] = 1 activity = pd.Series(binary).fillna(method="bfill").values activity = pd.Series(activity).fillna(0) return activity # ============================================================================= # Internals # ============================================================================= def _emg_activation_activations(activity, duration_min=0.05): activations = events_find( activity, threshold=0.5, threshold_keep="above", duration_min=duration_min ) activations["offset"] = activations["onset"] + activations["duration"] baseline = events_find( activity == 0, threshold=0.5, threshold_keep="above", duration_min=duration_min ) baseline["offset"] = baseline["onset"] + baseline["duration"] # Cross-comparison valid = np.isin(activations["onset"], baseline["offset"]) onsets = activations["onset"][valid] offsets = activations["offset"][valid] # make sure offset indices are within length of signal offsets = offsets[offsets < len(activity)] new_activity = np.array([]) for x, y in zip(onsets, offsets): activated = np.arange(x, y) new_activity = np.append(new_activity, activated) # Prepare Output. info = {"EMG_Onsets": onsets, "EMG_Offsets": offsets, "EMG_Activity": new_activity} return info
15,435
34.731481
118
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_simulate.py
# -*- coding: utf-8 -*- import numpy as np from ..misc import check_random_state from ..signal import signal_resample def emg_simulate( duration=10, length=None, sampling_rate=1000, noise=0.01, burst_number=1, burst_duration=1.0, random_state=None, ): """**Simulate an EMG signal** Generate an artificial (synthetic) EMG signal of a given duration and sampling rate. Parameters ---------- duration : int Desired recording length in seconds. sampling_rate : int The desired sampling rate (in Hz, i.e., samples/second). length : int The desired length of the signal (in samples). noise : float Noise level (gaussian noise). burst_number : int Desired number of bursts of activity (active muscle periods). burst_duration : float or list Duration of the bursts. Can be a float (each burst will have the same duration) or a list of durations for each bursts. random_state : None, int, numpy.random.RandomState or numpy.random.Generator Seed for the random number generator. See for ``misc.check_random_state`` for further information. Returns ---------- array Vector containing the EMG signal. Examples ---------- .. ipython:: python import neurokit2 as nk import pandas as pd emg = nk.emg_simulate(duration=10, burst_number=3) @savefig p_emg_simulate1.png scale=100% fig = nk.signal_plot(emg) @suppress plt.close() See Also -------- ecg_simulate, rsp_simulate, eda_simulate, ppg_simulate References ----------- This function is based on `this script <https://scientificallysound.org/2016/08/11/python-analysing-emg-signals-part-1/>`_. """ # Seed the random generator for reproducible results rng = check_random_state(random_state) # Generate number of samples automatically if length is unspecified if length is None: length = duration * sampling_rate # Sanity checks if isinstance(burst_duration, (int, float)): burst_duration = np.repeat(burst_duration, burst_number) if len(burst_duration) > burst_number: raise ValueError( "NeuroKit error: emg_simulate(): 'burst_duration' cannot be longer than the value of 'burst_number'" ) total_duration_bursts = np.sum(burst_duration) if total_duration_bursts > duration: raise ValueError( "NeuroKit error: emg_simulate(): The total duration of bursts cannot exceed the total duration" ) # Generate bursts bursts = [] for burst in range(burst_number): bursts += [list(rng.uniform(-1, 1, size=int(1000 * burst_duration[burst])) + 0.08)] # Generate quiet n_quiet = burst_number + 1 # number of quiet periods (in between bursts) duration_quiet = (duration - total_duration_bursts) / n_quiet # duration of each quiet period quiets = [] for quiet in range(n_quiet): # pylint: disable=W0612 quiets += [list(rng.uniform(-0.05, 0.05, size=int(1000 * duration_quiet)) + 0.08)] # Merge the two emg = [] for i in range(len(quiets)): # pylint: disable=C0200 emg += quiets[i] if i < len(bursts): emg += bursts[i] emg = np.array(emg) # Add random (gaussian distributed) noise emg += rng.normal(0, noise, len(emg)) # Resample emg = signal_resample( emg, sampling_rate=1000, desired_length=length, desired_sampling_rate=sampling_rate ) return emg
3,570
29.008403
112
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_process.py
# -*- coding: utf-8 -*- import pandas as pd from ..misc.report import create_report from ..signal import signal_sanitize from .emg_activation import emg_activation from .emg_amplitude import emg_amplitude from .emg_clean import emg_clean from .emg_methods import emg_methods from .emg_plot import emg_plot def emg_process(emg_signal, sampling_rate=1000, report=None, **kwargs): """**Process a electromyography (EMG) signal** Convenience function that automatically processes an electromyography signal. Parameters ---------- emg_signal : Union[list, np.array, pd.Series] The raw electromyography channel. sampling_rate : int The sampling frequency of ``emg_signal`` (in Hz, i.e., samples/second). report : str The filename of a report containing description and figures of processing (e.g. ``"myreport.html"``). Needs to be supplied if a report file should be generated. Defaults to ``None``. Can also be ``"text"`` to just print the text in the console without saving anything. **kwargs Other arguments to be passed to specific methods. For more information, see :func:`.emg_methods`. Returns ------- signals : DataFrame A DataFrame of same length as ``emg_signal`` containing the following columns: * ``"EMG_Raw"``: the raw signal. * ``"EMG_Clean"``: the cleaned signal. * ``"EMG_Amplitude"``: the signal amplitude, or the activation level of the signal. * ``"EMG_Activity"``: the activity of the signal for which amplitude exceeds the threshold specified,marked as "1" in a list of zeros. * ``"EMG_Onsets"``: the onsets of the amplitude, marked as "1" in a list of zeros. * ``"EMG_Offsets"``: the offsets of the amplitude, marked as "1" in a list of zeros. info : dict A dictionary containing the information of each amplitude onset, offset, and peak activity (see :func:`emg_activation`), as well as the signals' sampling rate. See Also -------- emg_clean, emg_amplitude, emg_plot Examples -------- .. ipython:: python import neurokit2 as nk emg = nk.emg_simulate(duration=10, sampling_rate=1000, burst_number=3) signals, info = nk.emg_process(emg, sampling_rate=1000) @savefig p_emg_process1.png scale=100% nk.emg_plot(signals) @suppress plt.close() """ # Sanitize input emg_signal = signal_sanitize(emg_signal) methods = emg_methods(sampling_rate=sampling_rate, **kwargs) # Clean signal emg_cleaned = emg_clean( emg_signal, sampling_rate=sampling_rate, method=methods["method_cleaning"] ) # Get amplitude amplitude = emg_amplitude(emg_cleaned) # Get onsets, offsets, and periods of activity activity_signal, info = emg_activation( emg_amplitude=amplitude, emg_cleaned=emg_cleaned, sampling_rate=sampling_rate, method=methods["method_activation"], **methods["kwargs_activation"] ) info["sampling_rate"] = sampling_rate # Add sampling rate in dict info # Prepare output signals = pd.DataFrame( {"EMG_Raw": emg_signal, "EMG_Clean": emg_cleaned, "EMG_Amplitude": amplitude} ) signals = pd.concat([signals, activity_signal], axis=1) if report is not None: # Generate report containing description and figures of processing if ".html" in str(report): fig = emg_plot(signals, sampling_rate=sampling_rate, static=False) else: fig = None create_report(file=report, signals=signals, info=methods, fig=fig) return signals, info
3,693
33.523364
98
py
NeuroKit
NeuroKit-master/neurokit2/emg/__init__.py
"""Submodule for NeuroKit.""" from .emg_activation import emg_activation from .emg_amplitude import emg_amplitude from .emg_analyze import emg_analyze from .emg_clean import emg_clean from .emg_eventrelated import emg_eventrelated from .emg_intervalrelated import emg_intervalrelated from .emg_plot import emg_plot from .emg_process import emg_process from .emg_simulate import emg_simulate __all__ = [ "emg_simulate", "emg_clean", "emg_amplitude", "emg_process", "emg_plot", "emg_activation", "emg_eventrelated", "emg_intervalrelated", "emg_analyze", ]
593
22.76
52
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_intervalrelated.py
# -*- coding: utf-8 -*- import numpy as np import pandas as pd def emg_intervalrelated(data): """**EMG Analysis for Interval-related Data** Performs EMG analysis on longer periods of data (typically > 10 seconds), such as resting-state data. Parameters ---------- data : Union[dict, pd.DataFrame] A DataFrame containing the different processed signal(s) as different columns, typically generated by ``emg_process()`` or ``bio_process()``. Can also take a dict containing sets of separately processed DataFrames. Returns ------- DataFrame A dataframe containing the analyzed EMG features. The analyzed features consist of the following: * ``"EMG_Activation_N"``: the number of bursts of muscular activity. * ``"EMG_Amplitude_Mean"``: the mean amplitude of the muscular activity. See Also -------- .bio_process, emg_eventrelated Examples ---------- .. ipython:: python import neurokit2 as nk # Example with simulated data emg = nk.emg_simulate(duration=40, sampling_rate=1000, burst_number=3) emg_signals, info = nk.emg_process(emg, sampling_rate=1000) # Single dataframe is passed nk.emg_intervalrelated(emg_signals) epochs = nk.epochs_create(emg_signals, events=[0, 20000], sampling_rate=1000, epochs_end=20) nk.emg_intervalrelated(epochs) """ intervals = {} # Format input if isinstance(data, pd.DataFrame): activity_cols = [col for col in data.columns if "EMG_Onsets" in col] if len(activity_cols) == 1: intervals["Activation_N"] = data[activity_cols[0]].values.sum() else: raise ValueError( "NeuroKit error: emg_intervalrelated(): Wrong" "input, we couldn't extract activity bursts." "Please make sure your DataFrame" "contains an `EMG_Onsets` column." ) amplitude_cols = ["EMG_Amplitude", "EMG_Activity"] len([col in data.columns for col in amplitude_cols]) if len(amplitude_cols) == 2: data_bursts = data.loc[data["EMG_Activity"] == 1] intervals["Amplitude_Mean"] = data_bursts["EMG_Amplitude"].values.mean() else: raise ValueError( "NeuroKit error: emg_intervalrelated(): Wrong" "input, we couldn't extract EMG amplitudes." "Please make sure your DataFrame contains both" "`EMG_Amplitude` and `EMG_Activity` columns." ) emg_intervals = pd.DataFrame.from_dict(intervals, orient="index").T.add_prefix("EMG_") elif isinstance(data, dict): for index in data: intervals[index] = {} # Initialize empty container # Add label info intervals[index]["Label"] = data[index]["Label"].iloc[0] intervals[index] = _emg_intervalrelated_formatinput(data[index], intervals[index]) emg_intervals = pd.DataFrame.from_dict(intervals, orient="index") return emg_intervals # ============================================================================= # Internals # ============================================================================= def _emg_intervalrelated_formatinput(interval, output={}): """Format input for dictionary.""" # Sanitize input colnames = interval.columns.values if len([i for i in colnames if "EMG_Onsets" in i]) == 0: raise ValueError( "NeuroKit error: emg_intervalrelated(): Wrong" "input, we couldn't extract activity bursts." "Please make sure your DataFrame" "contains an `EMG_Onsets` column." ) activity_cols = ["EMG_Amplitude", "EMG_Activity"] if len([i in colnames for i in activity_cols]) != 2: raise ValueError( "NeuroKit error: emg_intervalrelated(): Wrong" "input, we couldn't extract EMG amplitudes." "Please make sure your DataFrame contains both" "`EMG_Amplitude` and `EMG_Activity` columns." ) bursts = interval["EMG_Onsets"].values data_bursts = interval.loc[interval["EMG_Activity"] == 1] output["EMG_Activation_N"] = np.sum(bursts) output["EMG_Amplitude_Mean"] = data_bursts["EMG_Amplitude"].values.mean() return output
4,366
35.090909
105
py
NeuroKit
NeuroKit-master/neurokit2/emg/emg_clean.py
# -*- coding: utf-8 -*- from warnings import warn import numpy as np import pandas as pd import scipy.signal from ..misc import NeuroKitWarning, as_vector from ..signal import signal_detrend def emg_clean(emg_signal, sampling_rate=1000, method="biosppy"): """**Preprocess an electromyography (emg) signal** Clean an EMG signal using a set of parameters. Only one method is available at the moment. * **BioSPPy**: fourth order 100 Hz highpass Butterworth filter followed by a constant detrending. Parameters ---------- emg_signal : Union[list, np.array, pd.Series] The raw EMG channel. sampling_rate : int The sampling frequency of ``emg_signal`` (in Hz, i.e., samples/second). Defaults to 1000. method : str The processing pipeline to apply. Can be one of ``"biosppy"`` or ``"none"``. Defaults to ``"biosppy"``. If ``"none"`` is passed, the raw signal will be returned without any cleaning. Returns ------- array Vector containing the cleaned EMG signal. See Also -------- emg_amplitude, emg_process, emg_plot Examples -------- .. ipython:: python import pandas as pd import neurokit2 as nk emg = nk.emg_simulate(duration=10, sampling_rate=1000) signals = pd.DataFrame({"EMG_Raw": emg, "EMG_Cleaned":nk.emg_clean(emg, sampling_rate=1000)}) @savefig p_emg_clean1.png scale=100% fig = signals.plot() @suppress plt.close() """ emg_signal = as_vector(emg_signal) # Missing data n_missing = np.sum(np.isnan(emg_signal)) if n_missing > 0: warn( "There are " + str(n_missing) + " missing data points in your signal." " Filling missing values by using the forward filling method.", category=NeuroKitWarning, ) emg_signal = _emg_clean_missing(emg_signal) method = str(method).lower() if method in ["none"]: clean = emg_signal elif method in ["biosppy"]: clean = _emg_clean_biosppy(emg_signal, sampling_rate=sampling_rate) else: raise ValueError( "NeuroKit error: emg_clean(): 'method' should be one of 'biosppy' or 'none'." ) return clean # ============================================================================= # Handle missing data # ============================================================================= def _emg_clean_missing(emg_signal): emg_signal = pd.DataFrame.pad(pd.Series(emg_signal)) return emg_signal # ============================================================================= # BioSPPy # ============================================================================= def _emg_clean_biosppy(emg_signal, sampling_rate=1000): # Parameters order = 4 frequency = 100 frequency = ( 2 * np.array(frequency) / sampling_rate ) # Normalize frequency to Nyquist Frequency (Fs/2). # Filtering b, a = scipy.signal.butter(N=order, Wn=frequency, btype="highpass", analog=False) filtered = scipy.signal.filtfilt(b, a, emg_signal) # Baseline detrending clean = signal_detrend(filtered, order=0) return clean
3,220
28.550459
99
py
ULR
ULR-main/dual-encoder/L2/utils.py
import os from transformers.data.processors.utils import DataProcessor, InputExample from transformers import PreTrainedTokenizer from tqdm import tqdm import random import code from typing import List, Optional, Union import json from dataclasses import dataclass import logging logger = logging.getLogger(__name__) @dataclass class InputFeatures: """ A single set of features of data. Property names are the same names as the corresponding inputs to a model. Args: input_ids: Indices of input sequence tokens in the vocabulary. attention_mask: Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``: Usually ``1`` for tokens that are NOT MASKED, ``0`` for MASKED (padded) tokens. token_type_ids: (Optional) Segment token indices to indicate first and second portions of the inputs. Only some models use them. label: (Optional) Label corresponding to the input. Int for classification problems, float for regression problems. """ input_ids_a: List[int] input_ids_b: List[int] attention_mask_a: Optional[List[int]] = None token_type_ids_a: Optional[List[int]] = None attention_mask_b: Optional[List[int]] = None token_type_ids_b: Optional[List[int]] = None label: Optional[Union[int, float]] = None def to_json_string(self): """Serializes this instance to a JSON string.""" return json.dumps(dataclasses.asdict(self)) + "\n" def dual_encoder_convert_examples_to_features( examples: List[InputExample], tokenizer: PreTrainedTokenizer, max_length: Optional[int] = None, task=None, label_list=None, output_mode=None, ): if max_length is None: max_length = tokenizer.max_len label_map = {label: i for i, label in enumerate(label_list)} def label_from_example(example: InputExample) -> Union[int, float, None]: if example.label is None: return None if output_mode == "classification": return label_map[example.label] elif output_mode == "regression": return float(example.label) raise KeyError(output_mode) labels = [label_from_example(example) for example in examples] batch_encoding_a = tokenizer.batch_encode_plus( [(example.text_a) for example in examples], max_length=max_length, pad_to_max_length=True, ) batch_encoding_b = tokenizer.batch_encode_plus( [(example.text_b) for example in examples], max_length=max_length, pad_to_max_length=True, ) features = [] for i in range(len(examples)): inputs = {} for k in batch_encoding_a: inputs[k + "_a"] = batch_encoding_a[k][i] inputs[k + "_b"] = batch_encoding_b[k][i] feature = InputFeatures(**inputs, label=labels[i]) features.append(feature) for i, example in enumerate(examples[:5]): logger.info("*** Example ***") logger.info("guid: %s" % (example.guid)) logger.info("features: %s" % features[i]) return features class NatcatProcessor(DataProcessor): """Processor for the Natcat data set.""" def __init__(self): super(NatcatProcessor, self).__init__() def get_examples(self, filepath): """See base class.""" """ filepath: the file of article-category pairs """ examples = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split("\t") pos_cat = line[0] neg_cats = line[1:-1] article = line[-1] for neg_cat in neg_cats: examples.append(InputExample(guid=i, text_a=pos_cat, text_b=article, label='1')) i += 1 examples.append(InputExample(guid=i, text_a=neg_cat, text_b=article, label='0')) i += 1 return examples def get_labels(self): """See base class.""" return ["0", "1"] class EvalProcessor: def __init__(self, cat_file_path): super(EvalProcessor, self).__init__() self.cats = [] with open(cat_file_path) as fin: for line in fin: self.cats.append(line.strip()) def get_examples(self, filepath): """See base class.""" """ filepath: the file of the evaluation dataset """ examples = [] labels = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split(",", 1) if line[0].startswith("'") or line[0].startswith('"'): line[0] = line[0][1:-1] label = int(line[0]) - 1 text = " ".join(line[1][1:-1].split()[:128]) if text.strip() == "": text = "N/A" for cat in self.cats: i += 1 if label >= len(self.cats): examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=1)) else: if cat == self.cats[label]: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=1)) else: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=0)) return examples def get_labels(self): return [0, 1] processors = { "natcat": NatcatProcessor, "eval": EvalProcessor, } output_modes = { "natcat": "classification", "eval": "classification", } def simple_accuracy(preds, labels): return (preds == labels).mean() def compute_metrics(task_name, preds, labels): if task_name in ["natcat"]: return {"acc": simple_accuracy(preds, labels)} if task_name in ["eval"]: return {"acc": simple_accuracy(preds, labels)} class DataFiles: def __init__(self, directory): self.all_files = [os.path.join(directory, f) for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f)) and f.endswith(".data")] self.todo_files = self.all_files def next(self): if len(self.todo_files) == 0: return None return self.todo_files.pop() def save(self, file_path): with open(file_path, "w") as fout: for f in self.todo_files: fout.write(f + "\n") def load(self, file_path): self.todo_files = [] with open(file_path) as fin: for f in fin: self.todo_files.append(f.strip())
6,812
31.913043
154
py
ULR
ULR-main/dual-encoder/L2/compute_acc_kmeans.py
import numpy as np import sys from sklearn.metrics import f1_score from sklearn.kernel_approximation import RBFSampler import code if len(sys.argv) not in [4]: print("Usage: python compute_acc.py test_file text_embeddings category_embeddings ") exit(-1) test_file = sys.argv[1] text_file = sys.argv[2] cat_file = sys.argv[3] with open(test_file) as fin: labels = [] for line in fin: label, _ = line.split(",", 1) if label.startswith("'") or label.startswith('"'): label = label[1:-1] labels.append(int(label) - 1) labels = np.array(labels) text_embeddings = np.loadtxt(text_file) cat_embeddings = np.loadtxt(cat_file) num_cats = cat_embeddings.shape[0] def euclidean(a, b): return np.sum((a - b) * (a-b), -1) centroids = cat_embeddings all_accs = [] all_scores = [] for k in range(100): centroids = np.expand_dims(centroids, 0) text_embeddings_unsqueezed = np.expand_dims(text_embeddings, 1) scores = euclidean(text_embeddings_unsqueezed, centroids) avg_score = np.mean(scores.min(1)) all_scores.append(avg_score) new_preds = scores.argmin(1) if k > 0: change_of_preds = np.sum(preds != new_preds) print("change of preds: ", change_of_preds) print("average score: ", avg_score) # code.interact(local=locals()) preds = new_preds acc = np.sum(preds == labels) / len(labels) all_accs.append(acc) print("after {} iterations of k means accuracy: {}".format(k, acc)) centroids = [] for i in range(num_cats): centroid = text_embeddings[preds == i] if centroid.shape[0] > 0: centroids.append(centroid.mean(0, keepdims=True)) else: centroids.append(text_embeddings.mean(0, keepdims=True)) centroids = np.concatenate(centroids, 0) centroids = (centroids + cat_embeddings) / 2 if k > 0 and change_of_preds == 0: break min_index = np.argmin(all_scores) print("min avg score: ", all_scores[min_index], ", the corresponding accuracy: ", all_accs[min_index], ", max accuracy: ", max(all_accs))
2,121
28.068493
137
py
ULR
ULR-main/dual-encoder/L2/eval_downstream_task.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import code import numpy as np import torch import torch.nn.functional as F from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertModel, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaModel, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from utils import dual_encoder_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertModel, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaModel, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def _sorted_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> List[str]: ordering_and_checkpoint_path = [] glob_checkpoints = glob.glob(os.path.join(args.output_dir, "{}-*".format(checkpoint_prefix))) for path in glob_checkpoints: if use_mtime: ordering_and_checkpoint_path.append((os.path.getmtime(path), path)) else: regex_match = re.match(".*{}-([0-9]+)".format(checkpoint_prefix), path) if regex_match and regex_match.groups(): ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path)) checkpoints_sorted = sorted(ordering_and_checkpoint_path) checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted] return checkpoints_sorted def _rotate_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> None: if not args.save_total_limit: return if args.save_total_limit <= 0: return # Check if we should delete older checkpoint(s) checkpoints_sorted = _sorted_checkpoints(args, checkpoint_prefix, use_mtime) if len(checkpoints_sorted) <= args.save_total_limit: return number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit) checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete] for checkpoint in checkpoints_to_be_deleted: logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint)) shutil.rmtree(checkpoint) def evaluate(args, model, tokenizer, file_path, cat_file_path, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = (args.task_name,) eval_outputs_dirs = (args.output_dir,) cats = [] with open(cat_file_path) as fin: for line in fin: cats.append(line.strip()) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, file_path, cat_file_path) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = [] text_embeddings = [] category_embeddings = [] for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs_a = {"input_ids": batch[0], "attention_mask": batch[1]} inputs_b = {"input_ids": batch[3], "attention_mask": batch[4]} labels = batch[6] if args.model_type != "distilbert": inputs_a["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids inputs_b["token_type_ids"] = ( batch[5] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs_a = model(**inputs_a) outputs_b = model(**inputs_b) # code.interact(local=locals()) if args.bert_representation == "pool": rep1 = outputs_a[1] rep2 = outputs_b[1] elif args.bert_representation == "avg": rep1 = torch.sum(outputs_a[0] * batch[1].unsqueeze(-1), 1) / (torch.sum(batch[1], 1, keepdim=True) + 1e-8) rep2 = torch.sum(outputs_b[0] * batch[4].unsqueeze(-1), 1) / (torch.sum(batch[4], 1, keepdim=True) + 1e-8) score = torch.sum((rep1 - rep2) * (rep1 - rep2), -1) category_embeddings.append(rep1.data.cpu().numpy()) text_embeddings.append(rep2.data.cpu().numpy()) nb_eval_steps += 1 if preds is None: preds.append(score.detach().cpu().numpy().reshape((-1))) else: preds.append(score.detach().cpu().numpy().reshape((-1))) preds = np.concatenate(preds, 0) preds = preds.reshape((-1, len(cats))) logger.info("save prediction file to eval_output_dir") out_file_name = ".".join(file_path.split("/")[-2:]) text_embeddings = np.concatenate(text_embeddings, 0) category_embeddings = np.concatenate(category_embeddings, 0) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".text.txt"), text_embeddings[::len(cats)]) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".category.txt"), category_embeddings[:len(cats)]) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".preds.txt"), preds) return results def load_and_cache_examples(args, task, tokenizer, file_path, cat_file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task](cat_file_path=cat_file_path) output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() # if args.task_name == "nyt": if args.multi_class: examples = (processor.get_examples(file_path, args.label_filepath)) else: examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, ) # Convert to Tensors and build dataset all_input_ids_a = torch.tensor([f.input_ids_a for f in features], dtype=torch.long) all_attention_mask_a = torch.tensor([f.attention_mask_a for f in features], dtype=torch.long) all_token_type_ids_a = torch.tensor([f.token_type_ids_a if f.token_type_ids_a is not None else [0]*len(f.attention_mask_a) for f in features], dtype=torch.long) all_input_ids_b = torch.tensor([f.input_ids_b for f in features], dtype=torch.long) all_attention_mask_b = torch.tensor([f.attention_mask_b for f in features], dtype=torch.long) all_token_type_ids_b = torch.tensor([f.token_type_ids_b if f.token_type_ids_b is not None else [0]*len(f.attention_mask_b) for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids_a, all_attention_mask_a, all_token_type_ids_a, all_input_ids_b, all_attention_mask_b, all_token_type_ids_b, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--eval_data_dir", default=None, type=str, required=True, help="The directory containing the evaluation dataset", ) parser.add_argument( "--eval_data_file", default=None, type=str, required=True, help="The directory containing the evaluation dataset", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--label_filepath", default=None, type=str, help="Path to the label file for the nyt dataset", ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--bert_representation", default="pool", choices=["avg", "pool"], type=str, help="The BERT representation type", ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument("--multi_class", action="store_true", help="Whether it is a multi class classfication task.") parser.add_argument( "--all_cats_file", default=None, type=str, help="The file containing all category names", ) # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) file_path = os.path.join(args.eval_data_dir, "test.csv") # if args.task_name == "nyt": if args.multi_class: file_path = os.path.join(args.eval_data_dir, "test.doc.txt") if args.all_cats_file is not None: all_cats_file = args.all_cats_file else: all_cats_file = os.path.join(args.eval_data_dir, "classes.txt.acl") checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, args.eval_data_file, all_cats_file, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
23,469
39.25729
164
py
ULR
ULR-main/dual-encoder/L2/train_natcat.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import code import numpy as np import torch from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange import torch.nn.functional as F from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertModel, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaModel, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from utils import dual_encoder_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertModel, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaModel, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def _sorted_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> List[str]: ordering_and_checkpoint_path = [] glob_checkpoints = glob.glob(os.path.join(args.output_dir, "{}-*".format(checkpoint_prefix))) for path in glob_checkpoints: if use_mtime: ordering_and_checkpoint_path.append((os.path.getmtime(path), path)) else: regex_match = re.match(".*{}-([0-9]+)".format(checkpoint_prefix), path) if regex_match and regex_match.groups(): ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path)) checkpoints_sorted = sorted(ordering_and_checkpoint_path) checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted] return checkpoints_sorted def _rotate_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> None: if not args.save_total_limit: return if args.save_total_limit <= 0: return # Check if we should delete older checkpoint(s) checkpoints_sorted = _sorted_checkpoints(args, checkpoint_prefix, use_mtime) if len(checkpoints_sorted) <= args.save_total_limit: return number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit) checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete] for checkpoint in checkpoints_to_be_deleted: logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint)) shutil.rmtree(checkpoint) def train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=0): """ Train the model """ tr_loss, logging_loss = 0.0, 0.0 model.zero_grad() set_seed(args) # Added here for reproductibility epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0]) for step, batch in enumerate(epoch_iterator): # Skip past any already trained steps if resuming training model.train() batch = tuple(t.to(args.device) for t in batch) inputs_a = {"input_ids": batch[0], "attention_mask": batch[1]} inputs_b = {"input_ids": batch[3], "attention_mask": batch[4]} labels = batch[6] if args.model_type != "distilbert": inputs_a["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids inputs_b["token_type_ids"] = ( batch[5] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs_a = model(**inputs_a) outputs_b = model(**inputs_b) if args.bert_representation == "pool": rep1 = outputs_a[1] rep2 = outputs_b[1] # now the score will be between -1 and 1 elif args.bert_representation == "avg": rep1 = torch.sum(outputs_a[0] * batch[1].unsqueeze(-1), 1) / (torch.sum(batch[1], 1, keepdim=True) + 1e-8) rep2 = torch.sum(outputs_b[0] * batch[4].unsqueeze(-1), 1) / (torch.sum(batch[4], 1, keepdim=True) + 1e-8) score = torch.sum((rep1 - rep2) * (rep1 - rep2), -1) score = score.view(-1, 2) # code.interact(local=locals()) score = score[:, 0] + args.margin - score[:, 1] score[score <= 0] = 0 loss = score.mean() #code.interact(local=locals()) # if label is 1, we want score to be high, so we negate it # if label is 0, we want the score to be low, so we keep the sign # loss = 2 - (labels - 0.5) * 2 * cosine_score # loss = (1 - 2 * labels) * cosine_score # This loss function does not give good model performances # loss = labels * 2 - (labels - 0.5) * 2 * (cosine_score + 1) if args.n_gpu > 1: loss = loss.mean() # mean() to average on multi-gpu parallel training if args.gradient_accumulation_steps > 1: loss = loss / args.gradient_accumulation_steps loss = loss.mean() if args.fp16: with amp.scale_loss(loss, optimizer) as scaled_loss: scaled_loss.backward() else: loss.backward() tr_loss += loss.item() if (step + 1) % args.gradient_accumulation_steps == 0: if args.fp16: torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm) else: torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm) optimizer.step() scheduler.step() # Update learning rate schedule model.zero_grad() global_step += 1 if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0: logs = {} if ( args.local_rank == -1 and args.evaluate_during_training ): # Only evaluate when single GPU otherwise metrics may not average well results = evaluate(args, model, tokenizer) for key, value in results.items(): eval_key = "eval_{}".format(key) logs[eval_key] = value loss_scalar = (tr_loss - logging_loss) / args.logging_steps learning_rate_scalar = scheduler.get_lr()[0] logs["learning_rate"] = learning_rate_scalar logs["loss"] = loss_scalar logging_loss = tr_loss for key, value in logs.items(): tb_writer.add_scalar(key, value, global_step) print(json.dumps({**logs, **{"step": global_step}})) if args.max_steps > 0 and global_step > args.max_steps: epoch_iterator.close() break return global_step, tr_loss / global_step, optimizer, scheduler def evaluate(args, model, tokenizer, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,) eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=True) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = None out_label_ids = None for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) tmp_eval_loss, logits = outputs[:2] eval_loss += tmp_eval_loss.mean().item() nb_eval_steps += 1 if preds is None: preds = logits.detach().cpu().numpy() out_label_ids = inputs["labels"].detach().cpu().numpy() else: preds = np.append(preds, logits.detach().cpu().numpy(), axis=0) out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0) eval_loss = eval_loss / nb_eval_steps preds = np.squeeze(preds) result = compute_metrics(eval_task, preds, out_label_ids) results.update(result) output_eval_file = os.path.join(eval_output_dir, prefix, "eval_results.txt") with open(output_eval_file, "w") as writer: logger.info("***** Eval results {} *****".format(prefix)) for key in sorted(result.keys()): logger.info(" %s = %s", key, str(result[key])) writer.write("%s = %s\n" % (key, str(result[key]))) return results def load_and_cache_examples(args, task, tokenizer, file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task]() output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, ) # Convert to Tensors and build dataset all_input_ids_a = torch.tensor([f.input_ids_a for f in features], dtype=torch.long) all_attention_mask_a = torch.tensor([f.attention_mask_a for f in features], dtype=torch.long) all_token_type_ids_a = torch.tensor([f.token_type_ids_a if f.token_type_ids_a is not None else [0]*len(f.attention_mask_a) for f in features], dtype=torch.long) all_input_ids_b = torch.tensor([f.input_ids_b for f in features], dtype=torch.long) all_attention_mask_b = torch.tensor([f.attention_mask_b for f in features], dtype=torch.long) all_token_type_ids_b = torch.tensor([f.token_type_ids_b if f.token_type_ids_b is not None else [0]*len(f.attention_mask_b) for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids_a, all_attention_mask_a, all_token_type_ids_a, all_input_ids_b, all_attention_mask_b, all_token_type_ids_b, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--data_dir", default=None, type=str, required=True, help="The input data dir. Should contain the .tsv files (or other data files) for the task.", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument( "--bert_representation", default="pool", choices=["avg", "pool"], type=str, help="The BERT representation type", ) parser.add_argument( "--margin", default=0.5, type=float, help="The margin to train with hinge loss", ) # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) processor = processors[args.task_name]() args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Training if args.do_train: datafiles = DataFiles(args.data_dir) if os.path.isfile(os.path.join(args.model_name_or_path, "datafiles.txt")): datafiles.load(os.path.join(args.model_name_or_path, "datafiles.txt")) global_step = 0 shard_count = 0 if args.local_rank in [-1, 0]: tb_writer = SummaryWriter() while True: todo_file = datafiles.next() if not todo_file: break if args.local_rank == 0: torch.distributed.barrier() train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, todo_file) args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu) train_sampler = SequentialSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) if shard_count == 0: # if this is the first shard, create the optimizer or load from the previous checkpoint # Prepare optimizer and schedule (linear warmup and decay) no_decay = ["bias", "LayerNorm.weight"] optimizer_grouped_parameters = [ { "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], "weight_decay": args.weight_decay, }, {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0}, ] t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs * len(datafiles.all_files) # 280 shards of data files in total optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) scheduler = get_linear_schedule_with_warmup( optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total ) # Check if saved optimizer or scheduler states exist if os.path.isfile(os.path.join(args.model_name_or_path, "optimizer.pt")) and os.path.isfile( os.path.join(args.model_name_or_path, "scheduler.pt") ): logger.info("loading optimizer and scheduler from %s", args.model_name_or_path) # Load in optimizer and scheduler states optimizer.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "optimizer.pt"))) scheduler.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "scheduler.pt"))) if args.fp16: try: from apex import amp except ImportError: raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.") model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True, ) # Train! logger.info("***** Running training *****") logger.info(" Num examples = %d", len(train_dataset)) logger.info(" Num Epochs = %d", args.num_train_epochs) logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size) logger.info( " Total train batch size (w. parallel, distributed & accumulation) = %d", args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1), ) logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps) logger.info(" Total optimization steps = %d", t_total) if shard_count == 0: # Check if continuing training from a checkpoint if os.path.exists(args.model_name_or_path): # set global_step to global_step of last saved checkpoint from model path try: global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0]) except ValueError: global_step = 0 epochs_trained = global_step // (len(train_dataloader) // args.gradient_accumulation_steps) logger.info(" Continuing training from checkpoint %s", args.model_name_or_path) logger.info(" Continuing training from global step %d", global_step) global_step, tr_loss, optimizer, scheduler = train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=global_step) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Save model checkpoint output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(output_dir) tokenizer.save_pretrained(output_dir) torch.save(args, os.path.join(output_dir, "training_args.bin")) logger.info("Saving model checkpoint to %s", output_dir) torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt")) torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt")) datafiles.save(os.path.join(output_dir, "datafiles.txt")) logger.info("Saving optimizer and scheduler states to %s", output_dir) _rotate_checkpoints(args, "checkpoint") shard_count += 1 if args.local_rank in [-1, 0]: tb_writer.close() # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained() if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): # Create output directory if needed if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]: os.makedirs(args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, "training_args.bin")) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir) model.to(args.device) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
33,817
41.753477
177
py
ULR
ULR-main/dual-encoder/cosine/utils.py
import os from transformers.data.processors.utils import DataProcessor, InputExample from transformers import PreTrainedTokenizer from tqdm import tqdm import random import code from typing import List, Optional, Union import json from dataclasses import dataclass import logging logger = logging.getLogger(__name__) @dataclass class InputFeatures: """ A single set of features of data. Property names are the same names as the corresponding inputs to a model. Args: input_ids: Indices of input sequence tokens in the vocabulary. attention_mask: Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``: Usually ``1`` for tokens that are NOT MASKED, ``0`` for MASKED (padded) tokens. token_type_ids: (Optional) Segment token indices to indicate first and second portions of the inputs. Only some models use them. label: (Optional) Label corresponding to the input. Int for classification problems, float for regression problems. """ input_ids_a: List[int] input_ids_b: List[int] attention_mask_a: Optional[List[int]] = None token_type_ids_a: Optional[List[int]] = None attention_mask_b: Optional[List[int]] = None token_type_ids_b: Optional[List[int]] = None label: Optional[Union[int, float]] = None def to_json_string(self): """Serializes this instance to a JSON string.""" return json.dumps(dataclasses.asdict(self)) + "\n" def dual_encoder_convert_examples_to_features( examples: List[InputExample], tokenizer: PreTrainedTokenizer, max_length: Optional[int] = None, task=None, label_list=None, output_mode=None, ): if max_length is None: max_length = tokenizer.max_len label_map = {label: i for i, label in enumerate(label_list)} def label_from_example(example: InputExample) -> Union[int, float, None]: if example.label is None: return None if output_mode == "classification": return label_map[example.label] elif output_mode == "regression": return float(example.label) raise KeyError(output_mode) labels = [label_from_example(example) for example in examples] batch_encoding_a = tokenizer.batch_encode_plus( [(example.text_a) for example in examples], max_length=max_length, pad_to_max_length=True, ) batch_encoding_b = tokenizer.batch_encode_plus( [(example.text_b) for example in examples], max_length=max_length, pad_to_max_length=True, ) features = [] for i in range(len(examples)): inputs = {} for k in batch_encoding_a: inputs[k + "_a"] = batch_encoding_a[k][i] inputs[k + "_b"] = batch_encoding_b[k][i] feature = InputFeatures(**inputs, label=labels[i]) features.append(feature) for i, example in enumerate(examples[:5]): logger.info("*** Example ***") logger.info("guid: %s" % (example.guid)) logger.info("features: %s" % features[i]) return features class NatcatProcessor(DataProcessor): """Processor for the Natcat data set.""" def __init__(self): super(NatcatProcessor, self).__init__() def get_examples(self, filepath): """See base class.""" """ filepath: the file of article-category pairs """ examples = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split("\t") pos_cats = line[:1] neg_cats = line[len(pos_cats):-1] article = line[-1] for pos_cat in pos_cats: examples.append(InputExample(guid=i, text_a=pos_cat, text_b=article, label='1')) i += 1 for neg_cat in neg_cats: examples.append(InputExample(guid=i, text_a=neg_cat, text_b=article, label='0')) i += 1 return examples def get_labels(self): """See base class.""" return ["0", "1"] class EvalProcessor: def __init__(self, cat_file_path): super(EvalProcessor, self).__init__() self.cats = [] with open(cat_file_path) as fin: for line in fin: self.cats.append(line.strip()) def get_examples(self, filepath): """See base class.""" """ filepath: the file of the evaluation dataset """ examples = [] labels = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split(",", 1) if line[0].startswith("'") or line[0].startswith('"'): line[0] = line[0][1:-1] label = int(line[0]) - 1 text = " ".join(line[1][1:-1].split()[:128]) if text.strip() == "": text = "N/A" for cat in self.cats: i += 1 if label >= len(self.cats): examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=1)) else: if cat == self.cats[label]: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=1)) else: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=0)) return examples def get_labels(self): return [0, 1] processors = { "natcat": NatcatProcessor, "eval": EvalProcessor, } output_modes = { "natcat": "classification", "eval": "classification", } def simple_accuracy(preds, labels): return (preds == labels).mean() def compute_metrics(task_name, preds, labels): if task_name in ["natcat"]: return {"acc": simple_accuracy(preds, labels)} if task_name in ["eval"]: return {"acc": simple_accuracy(preds, labels)} class DataFiles: def __init__(self, directory): self.all_files = [os.path.join(directory, f) for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f)) and f.endswith(".data")] self.todo_files = self.all_files def next(self): if len(self.todo_files) == 0: return None return self.todo_files.pop() def save(self, file_path): with open(file_path, "w") as fout: for f in self.todo_files: fout.write(f + "\n") def load(self, file_path): self.todo_files = [] with open(file_path) as fin: for f in fin: self.todo_files.append(f.strip())
6,865
32.009615
154
py
ULR
ULR-main/dual-encoder/cosine/compute_acc_kmeans_cosine.py
import numpy as np import sys from sklearn.metrics import f1_score import code import random random.seed(1) if len(sys.argv) not in [4]: print("Usage: python compute_acc.py test_file text_embeddings category_embeddings ") exit(-1) test_file = sys.argv[1] text_file = sys.argv[2] cat_file = sys.argv[3] with open(test_file) as fin: labels = [] for line in fin: label, _ = line.split(",", 1) if label.startswith("'") or label.startswith('"'): label = label[1:-1] labels.append(int(label) - 1) labels = np.array(labels) def normalize(x): return x / np.sum(x * x, -1, keepdims=True) text_embeddings = np.loadtxt(text_file) cat_embeddings = np.loadtxt(cat_file) num_cats = cat_embeddings.shape[0] text_embeddings = normalize(text_embeddings) cat_embeddings = normalize(cat_embeddings) def dot_product(a, b): return np.sum(a * b, -1) centroids = cat_embeddings all_avg_scores = [] all_accs = [] for k in range(100): centroids = np.expand_dims(centroids, 0) text_embeddings_unsqueezed = np.expand_dims(text_embeddings, 1) scores = dot_product(text_embeddings_unsqueezed, centroids) avg_score = np.mean(scores.max(1)) all_avg_scores.append(avg_score) new_preds = scores.argmax(1) if k > 0: change_of_preds = np.sum(preds != new_preds) print("change of preds: ", change_of_preds) preds = new_preds acc = np.sum(preds == labels) / len(labels) all_accs.append(acc) print("average score: ", avg_score) print("after {} iterations of k means accuracy: {}".format(k, acc)) centroids = [] for i in range(num_cats): centroid = text_embeddings[preds == i] if centroid.shape[0] > 0: centroids.append(centroid.mean(0, keepdims=True)) else: centroids.append(text_embeddings.mean(0, keepdims=True)) centroids = np.concatenate(centroids, 0) centroids = (centroids + cat_embeddings) / 2 if k > 0 and change_of_preds == 0: break max_index = np.argmax(all_avg_scores) print("max avg score: ", all_avg_scores[max_index], ", the corresponding accuracy: ", all_accs[max_index], ", max accuracy: ", max(all_accs))
2,231
24.953488
141
py
ULR
ULR-main/dual-encoder/cosine/eval_downstream_task.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import code import numpy as np import torch import torch.nn.functional as F from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertModel, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaModel, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from utils import dual_encoder_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertModel, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaModel, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def evaluate(args, model, tokenizer, file_path, cat_file_path, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = (args.task_name,) eval_outputs_dirs = (args.output_dir,) cats = [] with open(cat_file_path) as fin: for line in fin: cats.append(line.strip()) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, file_path, cat_file_path) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = [] text_embeddings = [] category_embeddings = [] for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs_a = {"input_ids": batch[0], "attention_mask": batch[1]} inputs_b = {"input_ids": batch[3], "attention_mask": batch[4]} labels = batch[6] if args.model_type != "distilbert": inputs_a["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids inputs_b["token_type_ids"] = ( batch[5] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs_a = model(**inputs_a) outputs_b = model(**inputs_b) # code.interact(local=locals()) if args.bert_representation == "pool": rep1 = outputs_a[1] rep2 = outputs_b[1] elif args.bert_representation == "avg": rep1 = torch.sum(outputs_a[0] * batch[1].unsqueeze(-1), 1) / (torch.sum(batch[1], 1, keepdim=True) + 1e-8) rep2 = torch.sum(outputs_b[0] * batch[4].unsqueeze(-1), 1) / (torch.sum(batch[4], 1, keepdim=True) + 1e-8) if args.similarity_function == "dot": score = torch.sum(rep1 * rep2, -1) # now the score will be between -1 and 1 elif args.similarity_function == "cosine": score = F.cosine_similarity(rep1, rep2) # now the score will be between -1 and 1 category_embeddings.append(rep1.data.cpu().numpy()) text_embeddings.append(rep2.data.cpu().numpy()) nb_eval_steps += 1 if preds is None: preds.append(score.detach().cpu().numpy().reshape((-1))) else: preds.append(score.detach().cpu().numpy().reshape((-1))) preds = np.concatenate(preds, 0) preds = preds.reshape((-1, len(cats))) logger.info("save prediction file to eval_output_dir") out_file_name = ".".join(file_path.split("/")[-2:]) text_embeddings = np.concatenate(text_embeddings, 0) category_embeddings = np.concatenate(category_embeddings, 0) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".text.txt"), text_embeddings[::len(cats)]) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".category.txt"), category_embeddings[:len(cats)]) np.savetxt(os.path.join(eval_output_dir, out_file_name + ".preds.txt"), preds) return results def load_and_cache_examples(args, task, tokenizer, file_path, cat_file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task](cat_file_path=cat_file_path) output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() # if args.task_name == "nyt": if args.multi_class: examples = (processor.get_examples(file_path, args.label_filepath)) else: examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, ) # Convert to Tensors and build dataset all_input_ids_a = torch.tensor([f.input_ids_a for f in features], dtype=torch.long) all_attention_mask_a = torch.tensor([f.attention_mask_a for f in features], dtype=torch.long) all_token_type_ids_a = torch.tensor([f.token_type_ids_a if f.token_type_ids_a is not None else [0]*len(f.attention_mask_a) for f in features], dtype=torch.long) all_input_ids_b = torch.tensor([f.input_ids_b for f in features], dtype=torch.long) all_attention_mask_b = torch.tensor([f.attention_mask_b for f in features], dtype=torch.long) all_token_type_ids_b = torch.tensor([f.token_type_ids_b if f.token_type_ids_b is not None else [0]*len(f.attention_mask_b) for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids_a, all_attention_mask_a, all_token_type_ids_a, all_input_ids_b, all_attention_mask_b, all_token_type_ids_b, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--eval_data_dir", default=None, type=str, required=True, help="The directory containing the evaluation dataset", ) parser.add_argument( "--eval_data_file", default=None, type=str, required=True, help="The directory containing the evaluation dataset", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--label_filepath", default=None, type=str, help="Path to the label file for the nyt dataset", ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--bert_representation", default="pool", choices=["avg", "pool"], type=str, help="The BERT representation type", ) parser.add_argument( "--similarity_function", default="pool", choices=["dot", "cosine"], type=str, help="The similarity scoring function", ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument("--multi_class", action="store_true", help="Whether it is a multi class classfication task.") parser.add_argument( "--all_cats_file", default=None, type=str, help="The file containing all category names", ) # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) file_path = os.path.join(args.eval_data_dir, "test.csv") # if args.task_name == "nyt": if args.all_cats_file is not None: all_cats_file = args.all_cats_file else: all_cats_file = os.path.join(args.eval_data_dir, "classes.txt.acl") checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) # result = evaluate(args, model, tokenizer, file_path, all_cats_file, prefix=prefix) result = evaluate(args, model, tokenizer, args.eval_data_file, all_cats_file, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
22,345
39.190647
164
py
ULR
ULR-main/dual-encoder/cosine/train_natcat.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import code import numpy as np import torch from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange import torch.nn.functional as F from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertModel, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaModel, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from utils import dual_encoder_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertModel, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaModel, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def _sorted_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> List[str]: ordering_and_checkpoint_path = [] glob_checkpoints = glob.glob(os.path.join(args.output_dir, "{}-*".format(checkpoint_prefix))) for path in glob_checkpoints: if use_mtime: ordering_and_checkpoint_path.append((os.path.getmtime(path), path)) else: regex_match = re.match(".*{}-([0-9]+)".format(checkpoint_prefix), path) if regex_match and regex_match.groups(): ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path)) checkpoints_sorted = sorted(ordering_and_checkpoint_path) checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted] return checkpoints_sorted def _rotate_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> None: if not args.save_total_limit: return if args.save_total_limit <= 0: return # Check if we should delete older checkpoint(s) checkpoints_sorted = _sorted_checkpoints(args, checkpoint_prefix, use_mtime) if len(checkpoints_sorted) <= args.save_total_limit: return number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit) checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete] for checkpoint in checkpoints_to_be_deleted: logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint)) shutil.rmtree(checkpoint) def train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=0): """ Train the model """ tr_loss, logging_loss = 0.0, 0.0 model.zero_grad() set_seed(args) # Added here for reproductibility epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0]) if args.similarity_function == "dot": crit = torch.nn.BCEWithLogitsLoss() elif args.similarity_function == "euclidean": pass for step, batch in enumerate(epoch_iterator): # Skip past any already trained steps if resuming training model.train() batch = tuple(t.to(args.device) for t in batch) inputs_a = {"input_ids": batch[0], "attention_mask": batch[1]} inputs_b = {"input_ids": batch[3], "attention_mask": batch[4]} labels = batch[6] if args.model_type != "distilbert": inputs_a["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids inputs_b["token_type_ids"] = ( batch[5] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs_a = model(**inputs_a) outputs_b = model(**inputs_b) if args.bert_representation == "pool": rep1 = outputs_a[1] rep2 = outputs_b[1] # now the score will be between -1 and 1 elif args.bert_representation == "avg": rep1 = torch.sum(outputs_a[0] * batch[1].unsqueeze(-1), 1) / (torch.sum(batch[1], 1, keepdim=True) + 1e-8) rep2 = torch.sum(outputs_b[0] * batch[4].unsqueeze(-1), 1) / (torch.sum(batch[4], 1, keepdim=True) + 1e-8) if args.similarity_function == "dot": score = torch.sum(rep1 * rep2, -1) # now the score will be between -1 and 1 loss = crit(score, labels.float()) elif args.similarity_function == "euclidean": score = torch.sum((rep1 - rep2) * (rep1 - rep2), -1) #code.interact(local=locals()) # if label is 1, we want score to be high, so we negate it # if label is 0, we want the score to be low, so we keep the sign # loss = 2 - (labels - 0.5) * 2 * cosine_score # loss = (1 - 2 * labels) * cosine_score # This loss function does not give good model performances # loss = labels * 2 - (labels - 0.5) * 2 * (cosine_score + 1) if args.n_gpu > 1: loss = loss.mean() # mean() to average on multi-gpu parallel training if args.gradient_accumulation_steps > 1: loss = loss / args.gradient_accumulation_steps loss = loss.mean() if args.fp16: with amp.scale_loss(loss, optimizer) as scaled_loss: scaled_loss.backward() else: loss.backward() tr_loss += loss.item() if (step + 1) % args.gradient_accumulation_steps == 0: if args.fp16: torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm) else: torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm) optimizer.step() scheduler.step() # Update learning rate schedule model.zero_grad() global_step += 1 if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0: logs = {} if ( args.local_rank == -1 and args.evaluate_during_training ): # Only evaluate when single GPU otherwise metrics may not average well results = evaluate(args, model, tokenizer) for key, value in results.items(): eval_key = "eval_{}".format(key) logs[eval_key] = value loss_scalar = (tr_loss - logging_loss) / args.logging_steps learning_rate_scalar = scheduler.get_lr()[0] logs["learning_rate"] = learning_rate_scalar logs["loss"] = loss_scalar logging_loss = tr_loss for key, value in logs.items(): tb_writer.add_scalar(key, value, global_step) print(json.dumps({**logs, **{"step": global_step}})) if args.max_steps > 0 and global_step > args.max_steps: epoch_iterator.close() break return global_step, tr_loss / global_step, optimizer, scheduler def evaluate(args, model, tokenizer, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,) eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=True) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = None out_label_ids = None for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) tmp_eval_loss, logits = outputs[:2] eval_loss += tmp_eval_loss.mean().item() nb_eval_steps += 1 if preds is None: preds = logits.detach().cpu().numpy() out_label_ids = inputs["labels"].detach().cpu().numpy() else: preds = np.append(preds, logits.detach().cpu().numpy(), axis=0) out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0) eval_loss = eval_loss / nb_eval_steps preds = np.squeeze(preds) result = compute_metrics(eval_task, preds, out_label_ids) results.update(result) output_eval_file = os.path.join(eval_output_dir, prefix, "eval_results.txt") with open(output_eval_file, "w") as writer: logger.info("***** Eval results {} *****".format(prefix)) for key in sorted(result.keys()): logger.info(" %s = %s", key, str(result[key])) writer.write("%s = %s\n" % (key, str(result[key]))) return results def load_and_cache_examples(args, task, tokenizer, file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task]() output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, ) # Convert to Tensors and build dataset all_input_ids_a = torch.tensor([f.input_ids_a for f in features], dtype=torch.long) all_attention_mask_a = torch.tensor([f.attention_mask_a for f in features], dtype=torch.long) all_token_type_ids_a = torch.tensor([f.token_type_ids_a if f.token_type_ids_a is not None else [0]*len(f.attention_mask_a) for f in features], dtype=torch.long) all_input_ids_b = torch.tensor([f.input_ids_b for f in features], dtype=torch.long) all_attention_mask_b = torch.tensor([f.attention_mask_b for f in features], dtype=torch.long) all_token_type_ids_b = torch.tensor([f.token_type_ids_b if f.token_type_ids_b is not None else [0]*len(f.attention_mask_b) for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids_a, all_attention_mask_a, all_token_type_ids_a, all_input_ids_b, all_attention_mask_b, all_token_type_ids_b, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--data_dir", default=None, type=str, required=True, help="The input data dir. Should contain the .tsv files (or other data files) for the task.", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument( "--bert_representation", default="pool", choices=["avg", "pool"], type=str, help="The BERT representation type", ) parser.add_argument( "--similarity_function", default="pool", choices=["dot"], type=str, help="The similarity scoring function", ) # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) processor = processors[args.task_name]() args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Training if args.do_train: datafiles = DataFiles(args.data_dir) if os.path.isfile(os.path.join(args.model_name_or_path, "datafiles.txt")): datafiles.load(os.path.join(args.model_name_or_path, "datafiles.txt")) global_step = 0 shard_count = 0 if args.local_rank in [-1, 0]: tb_writer = SummaryWriter() while True: todo_file = datafiles.next() if not todo_file: break if args.local_rank == 0: torch.distributed.barrier() train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, todo_file) args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu) train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) if shard_count == 0: # if this is the first shard, create the optimizer or load from the previous checkpoint # Prepare optimizer and schedule (linear warmup and decay) no_decay = ["bias", "LayerNorm.weight"] optimizer_grouped_parameters = [ { "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], "weight_decay": args.weight_decay, }, {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0}, ] t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs * len(datafiles.all_files) # 280 shards of data files in total optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) scheduler = get_linear_schedule_with_warmup( optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total ) # Check if saved optimizer or scheduler states exist if os.path.isfile(os.path.join(args.model_name_or_path, "optimizer.pt")) and os.path.isfile( os.path.join(args.model_name_or_path, "scheduler.pt") ): logger.info("loading optimizer and scheduler from %s", args.model_name_or_path) # Load in optimizer and scheduler states optimizer.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "optimizer.pt"))) scheduler.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "scheduler.pt"))) if args.fp16: try: from apex import amp except ImportError: raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.") model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True, ) # Train! logger.info("***** Running training *****") logger.info(" Num examples = %d", len(train_dataset)) logger.info(" Num Epochs = %d", args.num_train_epochs) logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size) logger.info( " Total train batch size (w. parallel, distributed & accumulation) = %d", args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1), ) logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps) logger.info(" Total optimization steps = %d", t_total) if shard_count == 0: # Check if continuing training from a checkpoint if os.path.exists(args.model_name_or_path): # set global_step to global_step of last saved checkpoint from model path try: global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0]) except ValueError: global_step = 0 epochs_trained = global_step // (len(train_dataloader) // args.gradient_accumulation_steps) logger.info(" Continuing training from checkpoint %s", args.model_name_or_path) logger.info(" Continuing training from global step %d", global_step) global_step, tr_loss, optimizer, scheduler = train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=global_step) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Save model checkpoint output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(output_dir) tokenizer.save_pretrained(output_dir) torch.save(args, os.path.join(output_dir, "training_args.bin")) logger.info("Saving model checkpoint to %s", output_dir) torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt")) torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt")) datafiles.save(os.path.join(output_dir, "datafiles.txt")) logger.info("Saving optimizer and scheduler states to %s", output_dir) _rotate_checkpoints(args, "checkpoint") shard_count += 1 if args.local_rank in [-1, 0]: tb_writer.close() # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained() if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): # Create output directory if needed if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]: os.makedirs(args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, "training_args.bin")) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir) model.to(args.device) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
34,036
41.813836
177
py
ULR
ULR-main/single-encoder/utils.py
import os from transformers.data.processors.utils import DataProcessor, InputExample from tqdm import tqdm import random import code class NatcatProcessor(DataProcessor): """Processor for the Natcat data set.""" def __init__(self): super(NatcatProcessor, self).__init__() def get_examples(self, filepath): """See base class.""" """ filepath: the file of article-category pairs """ examples = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split("\t") pos_cats = line[:1] neg_cats = line[len(pos_cats):-1] article = line[-1] for pos_cat in pos_cats: examples.append(InputExample(guid=i, text_a=pos_cat, text_b=article, label='1')) i += 1 for neg_cat in neg_cats: examples.append(InputExample(guid=i, text_a=neg_cat, text_b=article, label='0')) i += 1 return examples def get_labels(self): """See base class.""" return ["0", "1"] class EvalProcessor: def __init__(self, cat_file_path): super(EvalProcessor, self).__init__() self.cats = [] with open(cat_file_path) as fin: for line in fin.read().strip().split("\n"): self.cats.append(line.strip()) def get_examples(self, filepath): """See base class.""" """ filepath: the file of the evaluation dataset """ examples = [] labels = [] i = 0 with open(filepath) as fin: lines = fin.read().strip().split("\n") for line in tqdm(lines): line = line.strip().split(",", 1) if line[0].startswith("'") or line[0].startswith('"'): line[0] = line[0][1:-1] label = int(line[0]) - 1 text = " ".join(line[1][1:-1].split()[:128]) if text.strip() == "": text = "N/A" for cat in self.cats: i += 1 if cat == self.cats[label]: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=1)) else: examples.append(InputExample(guid=i, text_a=cat, text_b=text, label=0)) return examples def get_labels(self): return [0, 1] processors = { "natcat": NatcatProcessor, "eval": EvalProcessor, } output_modes = { "natcat": "classification", "eval": "classification", } def simple_accuracy(preds, labels): return (preds == labels).mean() def compute_metrics(task_name, preds, labels): if task_name in ["wikicat"]: return {"acc": simple_accuracy(preds, labels)} if task_name in ["eval"]: return {"acc": simple_accuracy(preds, labels)} class DataFiles: def __init__(self, directory): self.all_files = [os.path.join(directory, f) for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f)) and f.endswith(".data")] self.todo_files = self.all_files def next(self): if len(self.todo_files) == 0: return None return self.todo_files.pop() def save(self, file_path): with open(file_path, "w") as fout: for f in self.todo_files: fout.write(f + "\n") def load(self, file_path): self.todo_files = [] with open(file_path) as fin: for f in fin: self.todo_files.append(f.strip())
3,716
29.975
154
py
ULR
ULR-main/single-encoder/compute_acc_kmeans.py
import numpy as np import sys from sklearn.metrics import f1_score from scipy.spatial import distance from scipy.special import softmax, kl_div import code if len(sys.argv) != 3: print("Usage: python compute_acc.py preds_file test_file") exit(-1) test_file = sys.argv[1] preds_file = sys.argv[2] with open(test_file) as fin: labels = [] for line in fin: label, _ = line.split(",", 1) if label.startswith("'") or label.startswith('"'): label = label[1:-1] labels.append(int(label) - 1) labels = np.array(labels) preds = np.loadtxt(preds_file) num_cats = preds.shape[1] num_ins = preds.shape[0] probs = softmax(preds, axis=1) preds = preds.argmax(1) acc = np.sum(preds == labels) / len(labels) print("accuracy of {}: {}".format(preds_file, acc)) def kl_divergence(p, q): return np.sum(np.where(p != 0, p * np.log(p / q), 0), -1) centroids = np.zeros((num_cats, num_cats)) cat_embeddings = np.zeros((num_cats, num_cats)) for i in range(num_cats): centroids[i][i] = 1 cat_embeddings[i][i] = 1 all_avg_scores = [] all_accs = [] for k in range(100): old_preds = preds.copy() scores = np.zeros((num_ins, num_cats)) for i in range(num_ins): for j in range(num_cats): scores[i, j] = distance.jensenshannon(probs[i], centroids[j]) preds = scores.argmin(1) avg_score = np.mean(scores.min(1)) all_avg_scores.append(avg_score) acc = np.sum(preds == labels) / len(labels) all_accs.append(acc) num_updates = np.sum(preds != old_preds) print("iteration: ", k, ", accuracy: ", acc, ", number of updates: ", num_updates) if num_updates == 0 and k > 0: break new_centroids = [] for j in range(num_cats): new_centroid = probs[preds == j] if len(new_centroid) == 0: new_centroid = centroids[j][None, :] else: new_centroid = new_centroid.mean(0, keepdims=True) new_centroids.append(new_centroid) centroids = np.concatenate(new_centroids, axis=0)
2,057
26.078947
86
py
ULR
ULR-main/single-encoder/eval_downstream_task.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import code import numpy as np import torch from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertForSequenceClassification, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from transformers import glue_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertForSequenceClassification, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def _sorted_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> List[str]: ordering_and_checkpoint_path = [] glob_checkpoints = glob.glob(os.path.join(args.output_dir, "{}-*".format(checkpoint_prefix))) for path in glob_checkpoints: if use_mtime: ordering_and_checkpoint_path.append((os.path.getmtime(path), path)) else: regex_match = re.match(".*{}-([0-9]+)".format(checkpoint_prefix), path) if regex_match and regex_match.groups(): ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path)) checkpoints_sorted = sorted(ordering_and_checkpoint_path) checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted] return checkpoints_sorted def _rotate_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> None: if not args.save_total_limit: return if args.save_total_limit <= 0: return # Check if we should delete older checkpoint(s) checkpoints_sorted = _sorted_checkpoints(args, checkpoint_prefix, use_mtime) if len(checkpoints_sorted) <= args.save_total_limit: return number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit) checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete] for checkpoint in checkpoints_to_be_deleted: logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint)) shutil.rmtree(checkpoint) def evaluate(args, model, tokenizer, file_path, cat_file_path, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = (args.task_name,) eval_outputs_dirs = (args.output_dir,) cats = [] with open(cat_file_path) as fin: for line in fin: cats.append(line.strip()) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, file_path, cat_file_path) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = None out_label_ids = None for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) tmp_eval_loss, logits = outputs[:2] eval_loss += tmp_eval_loss.mean().item() nb_eval_steps += 1 if preds is None: preds = logits.detach().cpu().numpy() out_label_ids = inputs["labels"].detach().cpu().numpy() else: preds = np.append(preds, logits.detach().cpu().numpy(), axis=0) out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0) preds = preds[:,1] - preds[:,0] # try argmax over pos scores eval_loss = eval_loss / nb_eval_steps preds = np.squeeze(preds) preds = preds.reshape((-1, len(cats))) # code.interact(local=locals()) logger.info("save prediction file to eval_output_dir") # out_file_name = cat_file_path.split("/")[-2] task_name = args.class_file_name.split("/")[-2] np.savetxt(os.path.join(eval_output_dir, task_name + ".preds.txt"), preds) return results def load_and_cache_examples(args, task, tokenizer, file_path, cat_file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task](cat_file_path=cat_file_path) output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() # if args.task_name == "nyt": if args.multi_class: examples = (processor.get_examples(file_path, args.label_filepath)) else: examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, pad_on_left=bool(args.model_type in ["xlnet"]), # pad on the left for xlnet pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0], pad_token_segment_id=4 if args.model_type in ["xlnet"] else 0, ) # Convert to Tensors and build dataset all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long) all_attention_mask = torch.tensor([f.attention_mask for f in features], dtype=torch.long) all_token_type_ids = torch.tensor([f.token_type_ids for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--eval_data_dir", default=None, type=str, help="The directory containing the evaluation dataset", ) parser.add_argument( "--eval_data_file", default=None, type=str, required=True, help="The file containing the evaluation dataset", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--label_filepath", default=None, type=str, help="Path to the label file for the nyt dataset", ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) parser.add_argument("--multi_class", action="store_true", help="Whether it is a multi class classfication task.") parser.add_argument("--class_file_name", type=str, default="classes.txt.acl", help="The file containing all class descriptions") parser.add_argument("--pred_file_suffix", type=str, default=None, help="Suffix after the prediction file") parser.add_argument("--pred_file_prefix", type=str, default=None, help="Prefix before the prediction file") # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) file_path = args.eval_data_file # if args.task_name == "nyt": # all_cats_file = os.path.join(args.eval_data_dir, args.class_file_name) all_cats_file = args.class_file_name checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, file_path, all_cats_file, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
22,069
38.981884
150
py
ULR
ULR-main/single-encoder/train_natcat.py
# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Finetuning the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa, Albert, XLM-RoBERTa).""" import argparse import glob import json import logging import os import re import random import shutil import pickle from typing import Dict, List, Tuple import numpy as np import torch from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset from torch.utils.data.distributed import DistributedSampler from tqdm import tqdm, trange from transformers import ( WEIGHTS_NAME, AdamW, AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer, BertConfig, BertForSequenceClassification, BertTokenizer, DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer, FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer, RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer, XLMConfig, XLMForSequenceClassification, XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer, XLMTokenizer, XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer, get_linear_schedule_with_warmup, ) from transformers import glue_convert_examples_to_features as convert_examples_to_features from utils import compute_metrics from utils import output_modes from utils import processors from utils import DataFiles try: from torch.utils.tensorboard import SummaryWriter except ImportError: from tensorboardX import SummaryWriter logger = logging.getLogger(__name__) ALL_MODELS = sum( ( tuple(conf.pretrained_config_archive_map.keys()) for conf in ( BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig, AlbertConfig, XLMRobertaConfig, FlaubertConfig, ) ), (), ) MODEL_CLASSES = { "bert": (BertConfig, BertForSequenceClassification, BertTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer), "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer), "flaubert": (FlaubertConfig, FlaubertForSequenceClassification, FlaubertTokenizer), } def set_seed(args): random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) if args.n_gpu > 0: torch.cuda.manual_seed_all(args.seed) def _sorted_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> List[str]: ordering_and_checkpoint_path = [] glob_checkpoints = glob.glob(os.path.join(args.output_dir, "{}-*".format(checkpoint_prefix))) for path in glob_checkpoints: if use_mtime: ordering_and_checkpoint_path.append((os.path.getmtime(path), path)) else: regex_match = re.match(".*{}-([0-9]+)".format(checkpoint_prefix), path) if regex_match and regex_match.groups(): ordering_and_checkpoint_path.append((int(regex_match.groups()[0]), path)) checkpoints_sorted = sorted(ordering_and_checkpoint_path) checkpoints_sorted = [checkpoint[1] for checkpoint in checkpoints_sorted] return checkpoints_sorted def _rotate_checkpoints(args, checkpoint_prefix="checkpoint", use_mtime=False) -> None: if not args.save_total_limit: return if args.save_total_limit <= 0: return # Check if we should delete older checkpoint(s) checkpoints_sorted = _sorted_checkpoints(args, checkpoint_prefix, use_mtime) if len(checkpoints_sorted) <= args.save_total_limit: return number_of_checkpoints_to_delete = max(0, len(checkpoints_sorted) - args.save_total_limit) checkpoints_to_be_deleted = checkpoints_sorted[:number_of_checkpoints_to_delete] for checkpoint in checkpoints_to_be_deleted: logger.info("Deleting older checkpoint [{}] due to args.save_total_limit".format(checkpoint)) shutil.rmtree(checkpoint) def train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=0): """ Train the model """ tr_loss, logging_loss = 0.0, 0.0 model.zero_grad() set_seed(args) # Added here for reproductibility epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0]) for step, batch in enumerate(epoch_iterator): # Skip past any already trained steps if resuming training model.train() batch = tuple(t.to(args.device) for t in batch) inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) loss = outputs[0] # model outputs are always tuple in transformers (see doc) if args.n_gpu > 1: loss = loss.mean() # mean() to average on multi-gpu parallel training if args.gradient_accumulation_steps > 1: loss = loss / args.gradient_accumulation_steps if args.fp16: with amp.scale_loss(loss, optimizer) as scaled_loss: scaled_loss.backward() else: loss.backward() tr_loss += loss.item() if (step + 1) % args.gradient_accumulation_steps == 0: if args.fp16: torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm) else: torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm) optimizer.step() scheduler.step() # Update learning rate schedule model.zero_grad() global_step += 1 if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0: logs = {} if ( args.local_rank == -1 and args.evaluate_during_training ): # Only evaluate when single GPU otherwise metrics may not average well results = evaluate(args, model, tokenizer) for key, value in results.items(): eval_key = "eval_{}".format(key) logs[eval_key] = value loss_scalar = (tr_loss - logging_loss) / args.logging_steps learning_rate_scalar = scheduler.get_lr()[0] logs["learning_rate"] = learning_rate_scalar logs["loss"] = loss_scalar logging_loss = tr_loss for key, value in logs.items(): tb_writer.add_scalar(key, value, global_step) print(json.dumps({**logs, **{"step": global_step}})) if args.max_steps > 0 and global_step > args.max_steps: epoch_iterator.close() break return global_step, tr_loss / global_step, optimizer, scheduler def evaluate(args, model, tokenizer, prefix=""): # Loop to handle MNLI double evaluation (matched, mis-matched) eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,) eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,) results = {} for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, evaluate=True) if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]: os.makedirs(eval_output_dir) args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu) # Note that DistributedSampler samples randomly eval_sampler = SequentialSampler(eval_dataset) eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size) # multi-gpu eval if args.n_gpu > 1 and not isinstance(model, torch.nn.DataParallel): model = torch.nn.DataParallel(model) # Eval! logger.info("***** Running evaluation {} *****".format(prefix)) logger.info(" Num examples = %d", len(eval_dataset)) logger.info(" Batch size = %d", args.eval_batch_size) eval_loss = 0.0 nb_eval_steps = 0 preds = None out_label_ids = None for batch in tqdm(eval_dataloader, desc="Evaluating"): model.eval() batch = tuple(t.to(args.device) for t in batch) with torch.no_grad(): inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]} if args.model_type != "distilbert": inputs["token_type_ids"] = ( batch[2] if args.model_type in ["bert", "xlnet", "albert"] else None ) # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids outputs = model(**inputs) tmp_eval_loss, logits = outputs[:2] eval_loss += tmp_eval_loss.mean().item() nb_eval_steps += 1 if preds is None: preds = logits.detach().cpu().numpy() out_label_ids = inputs["labels"].detach().cpu().numpy() else: preds = np.append(preds, logits.detach().cpu().numpy(), axis=0) out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0) eval_loss = eval_loss / nb_eval_steps preds = np.squeeze(preds) result = compute_metrics(eval_task, preds, out_label_ids) results.update(result) output_eval_file = os.path.join(eval_output_dir, prefix, "eval_results.txt") with open(output_eval_file, "w") as writer: logger.info("***** Eval results {} *****".format(prefix)) for key in sorted(result.keys()): logger.info(" %s = %s", key, str(result[key])) writer.write("%s = %s\n" % (key, str(result[key]))) return results def load_and_cache_examples(args, task, tokenizer, file_path): if args.local_rank not in [-1, 0] and not evaluate: torch.distributed.barrier() # Make sure only the first process in distributed training process the dataset, and the others will use the cache processor = processors[task]() output_mode = output_modes[task] # cached_features_file = os.path.join( # args.data_dir, # "cached_{}_{}_{}".format( # file_path, # str(args.max_seq_length), # str(task), # ), # ) # Load data features from cache or dataset file # if os.path.exists(cached_features_file) and not args.overwrite_cache: # logger.info("Loading features from cached file %s", cached_features_file) # features = torch.load(cached_features_file) # else: logger.info("Loading from dataset file at %s", file_path) label_list = processor.get_labels() examples = (processor.get_examples(file_path)) logger.info("Encoding features from dataset file at %s", file_path) # if args.local_rank in [-1, 0]: # logger.info("Saving features into cached file %s", cached_features_file) # torch.save(features, cached_features_file) features = convert_examples_to_features( examples, tokenizer, label_list=label_list, max_length=args.max_seq_length, output_mode=output_mode, pad_on_left=bool(args.model_type in ["xlnet"]), # pad on the left for xlnet pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0], pad_token_segment_id=4 if args.model_type in ["xlnet"] else 0, ) # Convert to Tensors and build dataset all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long) all_attention_mask = torch.tensor([f.attention_mask for f in features], dtype=torch.long) all_token_type_ids = torch.tensor([f.token_type_ids for f in features], dtype=torch.long) all_labels = torch.tensor([f.label for f in features], dtype=torch.long) dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels) return dataset def main(): parser = argparse.ArgumentParser() # Required parameters parser.add_argument( "--data_dir", default=None, type=str, required=True, help="The input data dir. Should contain the .tsv files (or other data files) for the task.", ) parser.add_argument( "--model_type", default=None, type=str, required=True, help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()), ) parser.add_argument( "--model_name_or_path", default=None, type=str, required=True, help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS), ) parser.add_argument( "--task_name", default=None, type=str, required=True, help="The name of the task to train selected in the list: " + ", ".join(processors.keys()), ) parser.add_argument( "--output_dir", default=None, type=str, required=True, help="The output directory where the model predictions and checkpoints will be written.", ) # Other parameters parser.add_argument( "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name", ) parser.add_argument( "--tokenizer_name", default="", type=str, help="Pretrained tokenizer name or path if not the same as model_name", ) parser.add_argument( "--cache_dir", default="", type=str, help="Where do you want to store the pre-trained models downloaded from s3", ) parser.add_argument( "--max_seq_length", default=128, type=int, help="The maximum total input sequence length after tokenization. Sequences longer " "than this will be truncated, sequences shorter will be padded.", ) parser.add_argument("--do_train", action="store_true", help="Whether to run training.") parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.") parser.add_argument( "--evaluate_during_training", action="store_true", help="Run evaluation during training at each logging step.", ) parser.add_argument( "--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model.", ) parser.add_argument( "--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.", ) parser.add_argument( "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.") parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.") parser.add_argument( "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform.", ) parser.add_argument( "--max_steps", default=-1, type=int, help="If > 0: set total number of training steps to perform. Override num_train_epochs.", ) parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.") parser.add_argument("--logging_steps", type=int, default=500, help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.") parser.add_argument( "--eval_all_checkpoints", action="store_true", help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number", ) parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available") parser.add_argument( "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory", ) parser.add_argument( "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets", ) parser.add_argument("--seed", type=int, default=42, help="random seed for initialization") parser.add_argument( "--fp16", action="store_true", help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit", ) parser.add_argument( "--fp16_opt_level", type=str, default="O1", help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." "See details at https://nvidia.github.io/apex/amp.html", ) parser.add_argument( "--save_total_limit", type=int, default=None, help="Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, does not delete by default", ) parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank") parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.") parser.add_argument("--server_port", type=str, default="", help="For distant debugging.") args = parser.parse_args() logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN, ) if ( os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir ): # set to load the latest checkpoint for training args.model_name_or_path = args.output_dir all_model_checkpoints = [ckpt for ckpt in os.listdir(args.model_name_or_path) if os.path.isdir(os.path.join(args.model_name_or_path, ckpt))] all_model_checkpoints = [(ckpt.split("-")[-1] if "-" in ckpt else -1, ckpt) for ckpt in all_model_checkpoints] all_model_checkpoints.sort(reverse=True) args.model_name_or_path = os.path.join(args.model_name_or_path, all_model_checkpoints[0][1]) logger.info("setting to load the model from %s", args.model_name_or_path) # Setup distant debugging if needed if args.server_ip and args.server_port: # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script import ptvsd print("Waiting for debugger attach") ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.wait_for_attach() # Setup CUDA, GPU & distributed training if args.local_rank == -1 or args.no_cuda: device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu") args.n_gpu = 0 if args.no_cuda else torch.cuda.device_count() else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs torch.cuda.set_device(args.local_rank) device = torch.device("cuda", args.local_rank) torch.distributed.init_process_group(backend="nccl") args.n_gpu = 1 args.device = device # Setup logging logger.warning( "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16, ) # Set seed set_seed(args) # Prepare GLUE task args.task_name = args.task_name.lower() if args.task_name not in processors: raise ValueError("Task not found: %s" % (args.task_name)) processor = processors[args.task_name]() args.output_mode = output_modes[args.task_name] num_labels = 2 # Load pretrained model and tokenizer if args.local_rank not in [-1, 0]: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab args.model_type = args.model_type.lower() config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config = config_class.from_pretrained( args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name, cache_dir=args.cache_dir if args.cache_dir else None, ) tokenizer = tokenizer_class.from_pretrained( args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, do_lower_case=args.do_lower_case, cache_dir=args.cache_dir if args.cache_dir else None, ) model = model_class.from_pretrained( args.model_name_or_path, from_tf=bool(".ckpt" in args.model_name_or_path), config=config, cache_dir=args.cache_dir if args.cache_dir else None, ) if args.local_rank == 0: torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab model.to(args.device) logger.info("Training/evaluation parameters %s", args) # Training if args.do_train: datafiles = DataFiles(args.data_dir) if os.path.isfile(os.path.join(args.model_name_or_path, "datafiles.txt")): datafiles.load(os.path.join(args.model_name_or_path, "datafiles.txt")) global_step = 0 shard_count = 0 if args.local_rank in [-1, 0]: tb_writer = SummaryWriter() while True: todo_file = datafiles.next() if not todo_file: break if args.local_rank == 0: torch.distributed.barrier() train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, todo_file) args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu) train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset) train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size) if shard_count == 0: # if this is the first shard, create the optimizer or load from the previous checkpoint # Prepare optimizer and schedule (linear warmup and decay) no_decay = ["bias", "LayerNorm.weight"] optimizer_grouped_parameters = [ { "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], "weight_decay": args.weight_decay, }, {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0}, ] t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs * len(datafiles.all_files) # 280 shards of data files in total optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) scheduler = get_linear_schedule_with_warmup( optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total ) # Check if saved optimizer or scheduler states exist if os.path.isfile(os.path.join(args.model_name_or_path, "optimizer.pt")) and os.path.isfile( os.path.join(args.model_name_or_path, "scheduler.pt") ): logger.info("loading optimizer and scheduler from %s", args.model_name_or_path) # Load in optimizer and scheduler states optimizer.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "optimizer.pt"))) scheduler.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "scheduler.pt"))) if args.fp16: try: from apex import amp except ImportError: raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.") model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level) # multi-gpu training (should be after apex fp16 initialization) if args.n_gpu > 1: model = torch.nn.DataParallel(model) # Distributed training (should be after apex fp16 initialization) if args.local_rank != -1: model = torch.nn.parallel.DistributedDataParallel( model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True, ) # Train! logger.info("***** Running training *****") logger.info(" Num examples = %d", len(train_dataset)) logger.info(" Num Epochs = %d", args.num_train_epochs) logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size) logger.info( " Total train batch size (w. parallel, distributed & accumulation) = %d", args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1), ) logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps) logger.info(" Total optimization steps = %d", t_total) if shard_count == 0: # Check if continuing training from a checkpoint if os.path.exists(args.model_name_or_path): # set global_step to global_step of last saved checkpoint from model path try: global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0]) except ValueError: global_step = 0 epochs_trained = global_step // (len(train_dataloader) // args.gradient_accumulation_steps) logger.info(" Continuing training from checkpoint %s", args.model_name_or_path) logger.info(" Continuing training from global step %d", global_step) global_step, tr_loss, optimizer, scheduler = train(args, train_dataset, train_dataloader, model, tokenizer, optimizer, scheduler, tb_writer, global_step=global_step) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) # Save model checkpoint output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step)) if not os.path.exists(output_dir): os.makedirs(output_dir) model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(output_dir) tokenizer.save_pretrained(output_dir) torch.save(args, os.path.join(output_dir, "training_args.bin")) logger.info("Saving model checkpoint to %s", output_dir) torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt")) torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt")) datafiles.save(os.path.join(output_dir, "datafiles.txt")) logger.info("Saving optimizer and scheduler states to %s", output_dir) _rotate_checkpoints(args, "checkpoint") shard_count += 1 if args.local_rank in [-1, 0]: tb_writer.close() # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained() if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): # Create output directory if needed if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]: os.makedirs(args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir) # Save a trained model, configuration and tokenizer using `save_pretrained()`. # They can then be reloaded using `from_pretrained()` model_to_save = ( model.module if hasattr(model, "module") else model ) # Take care of distributed/parallel training model_to_save.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir) # Good practice: save your training arguments together with the trained model torch.save(args, os.path.join(args.output_dir, "training_args.bin")) # Load a trained model and vocabulary that you have fine-tuned model = model_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir) model.to(args.device) # Evaluation results = {} if args.do_eval and args.local_rank in [-1, 0]: tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) checkpoints = [args.output_dir] if args.eval_all_checkpoints: checkpoints = list( os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True)) ) logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logger.info("Evaluate the following checkpoints: %s", checkpoints) for checkpoint in checkpoints: global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else "" model = model_class.from_pretrained(checkpoint) model.to(args.device) result = evaluate(args, model, tokenizer, prefix=prefix) result = dict((k + "_{}".format(global_step), v) for k, v in result.items()) results.update(result) return results if __name__ == "__main__": main()
31,833
41.902965
177
py
GlobalSfMpy
GlobalSfMpy-main/setup.py
import os import re import subprocess import sys from setuptools import Extension, setup from setuptools.command.build_ext import build_ext # Convert distutils Windows platform specifiers to CMake -A arguments PLAT_TO_CMAKE = { "win32": "Win32", "win-amd64": "x64", "win-arm32": "ARM", "win-arm64": "ARM64", } # A CMakeExtension needs a sourcedir instead of a file list. # The name must be the _single_ output extension from the CMake build. # If you need multiple extensions, see scikit-build. class CMakeExtension(Extension): def __init__(self, name, sourcedir=""): Extension.__init__(self, name, sources=[]) self.sourcedir = os.path.abspath(sourcedir) class CMakeBuild(build_ext): def build_extension(self, ext): extdir = os.path.abspath(os.path.dirname(self.get_ext_fullpath(ext.name))) # required for auto-detection & inclusion of auxiliary "native" libs if not extdir.endswith(os.path.sep): extdir += os.path.sep debug = int(os.environ.get("DEBUG", 0)) if self.debug is None else self.debug cfg = "Debug" if debug else "Release" # CMake lets you override the generator - we need to check this. # Can be set with Conda-Build, for example. cmake_generator = os.environ.get("CMAKE_GENERATOR", "") # Set Python_EXECUTABLE instead if you use PYBIND11_FINDPYTHON # EXAMPLE_VERSION_INFO shows you how to pass a value into the C++ code # from Python. cmake_args = [ f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={extdir}", f"-DPYTHON_EXECUTABLE={sys.executable}", f"-DCMAKE_BUILD_TYPE={cfg}", # not used on MSVC, but no harm ] build_args = [] # Adding CMake arguments set as environment variable # (needed e.g. to build for ARM OSx on conda-forge) if "CMAKE_ARGS" in os.environ: cmake_args += [item for item in os.environ["CMAKE_ARGS"].split(" ") if item] # In this example, we pass in the version to C++. You might not need to. cmake_args += [f"-DEXAMPLE_VERSION_INFO={self.distribution.get_version()}"] if self.compiler.compiler_type != "msvc": # Using Ninja-build since it a) is available as a wheel and b) # multithreads automatically. MSVC would require all variables be # exported for Ninja to pick it up, which is a little tricky to do. # Users can override the generator with CMAKE_GENERATOR in CMake # 3.15+. if not cmake_generator or cmake_generator == "Ninja": try: import ninja # noqa: F401 ninja_executable_path = os.path.join(ninja.BIN_DIR, "ninja") cmake_args += [ "-GNinja", f"-DCMAKE_MAKE_PROGRAM:FILEPATH={ninja_executable_path}", ] except ImportError: pass else: # Single config generators are handled "normally" single_config = any(x in cmake_generator for x in {"NMake", "Ninja"}) # CMake allows an arch-in-generator style for backward compatibility contains_arch = any(x in cmake_generator for x in {"ARM", "Win64"}) # Specify the arch if using MSVC generator, but only if it doesn't # contain a backward-compatibility arch spec already in the # generator name. if not single_config and not contains_arch: cmake_args += ["-A", PLAT_TO_CMAKE[self.plat_name]] # Multi-config generators have a different way to specify configs if not single_config: cmake_args += [ f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY_{cfg.upper()}={extdir}" ] build_args += ["--config", cfg] if sys.platform.startswith("darwin"): # Cross-compile support for macOS - respect ARCHFLAGS if set archs = re.findall(r"-arch (\S+)", os.environ.get("ARCHFLAGS", "")) if archs: cmake_args += ["-DCMAKE_OSX_ARCHITECTURES={}".format(";".join(archs))] # Set CMAKE_BUILD_PARALLEL_LEVEL to control the parallel build level # across all generators. if "CMAKE_BUILD_PARALLEL_LEVEL" not in os.environ: # self.parallel is a Python 3 only way to set parallel jobs by hand # using -j in the build_ext call, not supported by pip or PyPA-build. if hasattr(self, "parallel") and self.parallel: # CMake 3.12+ only. build_args += [f"-j{self.parallel}"] build_temp = os.path.join(self.build_temp, ext.name) if not os.path.exists(build_temp): os.makedirs(build_temp) subprocess.check_call(["cmake", ext.sourcedir] + cmake_args, cwd=build_temp) subprocess.check_call(["cmake", "--build", "."] + build_args, cwd=build_temp) # The information here can also be placed in setup.cfg - better separation of # logic and declaration, and simpler if you include description/version in a file. setup( name="GlobalSfMpy", version="0.0.1", author="Ganlin Zhang", author_email="zgl0315@gmail.com", description="Global Structure from Motion library, based on TheiaSfM, enable two-view uncertainty and customized loss function.", long_description="Global Structure from Motion library, based on TheiaSfM, enable two-view uncertainty and customized loss function.", ext_modules=[CMakeExtension("GlobalSfMpy")], cmdclass={"build_ext": CMakeBuild}, zip_safe=False, extras_require={"test": ["pytest>=6.0"]}, python_requires=">=3.6", )
5,750
41.286765
138
py
GlobalSfMpy
GlobalSfMpy-main/scripts/colmap_database.py
import sys import sqlite3 import numpy as np IS_PYTHON3 = sys.version_info[0] >= 3 MAX_IMAGE_ID = 2**31 - 1 CREATE_CAMERAS_TABLE = """CREATE TABLE IF NOT EXISTS cameras ( camera_id INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL, model INTEGER NOT NULL, width INTEGER NOT NULL, height INTEGER NOT NULL, params BLOB, prior_focal_length INTEGER NOT NULL)""" CREATE_DESCRIPTORS_TABLE = """CREATE TABLE IF NOT EXISTS descriptors ( image_id INTEGER PRIMARY KEY NOT NULL, rows INTEGER NOT NULL, cols INTEGER NOT NULL, data BLOB, FOREIGN KEY(image_id) REFERENCES images(image_id) ON DELETE CASCADE)""" CREATE_IMAGES_TABLE = """CREATE TABLE IF NOT EXISTS images ( image_id INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL, name TEXT NOT NULL UNIQUE, camera_id INTEGER NOT NULL, prior_qw REAL, prior_qx REAL, prior_qy REAL, prior_qz REAL, prior_tx REAL, prior_ty REAL, prior_tz REAL, CONSTRAINT image_id_check CHECK(image_id >= 0 and image_id < {}), FOREIGN KEY(camera_id) REFERENCES cameras(camera_id)) """.format(MAX_IMAGE_ID) CREATE_TWO_VIEW_GEOMETRIES_TABLE = """ CREATE TABLE IF NOT EXISTS two_view_geometries ( pair_id INTEGER PRIMARY KEY NOT NULL, rows INTEGER NOT NULL, cols INTEGER NOT NULL, data BLOB, config INTEGER NOT NULL, F BLOB, E BLOB, H BLOB) """ CREATE_KEYPOINTS_TABLE = """CREATE TABLE IF NOT EXISTS keypoints ( image_id INTEGER PRIMARY KEY NOT NULL, rows INTEGER NOT NULL, cols INTEGER NOT NULL, data BLOB, FOREIGN KEY(image_id) REFERENCES images(image_id) ON DELETE CASCADE) """ CREATE_MATCHES_TABLE = """CREATE TABLE IF NOT EXISTS matches ( pair_id INTEGER PRIMARY KEY NOT NULL, rows INTEGER NOT NULL, cols INTEGER NOT NULL, data BLOB)""" CREATE_NAME_INDEX = \ "CREATE UNIQUE INDEX IF NOT EXISTS index_name ON images(name)" CREATE_ALL = "; ".join([ CREATE_CAMERAS_TABLE, CREATE_IMAGES_TABLE, CREATE_KEYPOINTS_TABLE, CREATE_DESCRIPTORS_TABLE, CREATE_MATCHES_TABLE, CREATE_TWO_VIEW_GEOMETRIES_TABLE, CREATE_NAME_INDEX ]) def image_ids_to_pair_id(image_id1, image_id2): if image_id1 > image_id2: image_id1, image_id2 = image_id2, image_id1 return image_id1 * MAX_IMAGE_ID + image_id2 def pair_id_to_image_ids(pair_id): image_id2 = pair_id % MAX_IMAGE_ID image_id1 = (pair_id - image_id2) / MAX_IMAGE_ID return image_id1, image_id2 def array_to_blob(array): if IS_PYTHON3: return array.tostring() else: return np.getbuffer(array) def blob_to_array(blob, dtype, shape=(-1,)): if IS_PYTHON3: return np.fromstring(blob, dtype=dtype).reshape(*shape) else: return np.frombuffer(blob, dtype=dtype).reshape(*shape) class COLMAPDatabase(sqlite3.Connection): @staticmethod def connect(database_path): return sqlite3.connect(database_path, factory=COLMAPDatabase) def __init__(self, *args, **kwargs): super(COLMAPDatabase, self).__init__(*args, **kwargs) self.create_tables = lambda: self.executescript(CREATE_ALL) self.create_cameras_table = \ lambda: self.executescript(CREATE_CAMERAS_TABLE) self.create_descriptors_table = \ lambda: self.executescript(CREATE_DESCRIPTORS_TABLE) self.create_images_table = \ lambda: self.executescript(CREATE_IMAGES_TABLE) self.create_two_view_geometries_table = \ lambda: self.executescript(CREATE_TWO_VIEW_GEOMETRIES_TABLE) self.create_keypoints_table = \ lambda: self.executescript(CREATE_KEYPOINTS_TABLE) self.create_matches_table = \ lambda: self.executescript(CREATE_MATCHES_TABLE) self.create_name_index = lambda: self.executescript(CREATE_NAME_INDEX) def add_camera(self, model, width, height, params, prior_focal_length=False, camera_id=None): params = np.asarray(params, np.float64) cursor = self.execute( "INSERT INTO cameras VALUES (?, ?, ?, ?, ?, ?)", (camera_id, model, width, height, array_to_blob(params), prior_focal_length)) return cursor.lastrowid def add_image(self, name, camera_id, prior_q=np.full(4, np.NaN), prior_t=np.full(3, np.NaN), image_id=None): cursor = self.execute( "INSERT INTO images VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)", (image_id, name, camera_id, prior_q[0], prior_q[1], prior_q[2], prior_q[3], prior_t[0], prior_t[1], prior_t[2])) return cursor.lastrowid def add_keypoints(self, image_id, keypoints): assert(len(keypoints.shape) == 2) assert(keypoints.shape[1] in [2, 4, 6]) keypoints = np.asarray(keypoints, np.float32) self.execute( "INSERT INTO keypoints VALUES (?, ?, ?, ?)", (image_id,) + keypoints.shape + (array_to_blob(keypoints),)) def add_descriptors(self, image_id, descriptors): descriptors = np.ascontiguousarray(descriptors, np.uint8) self.execute( "INSERT INTO descriptors VALUES (?, ?, ?, ?)", (image_id,) + descriptors.shape + (array_to_blob(descriptors),)) def add_matches(self, image_id1, image_id2, matches): assert(len(matches.shape) == 2) assert(matches.shape[1] == 2) if image_id1 > image_id2: matches = matches[:,::-1] pair_id = image_ids_to_pair_id(image_id1, image_id2) matches = np.asarray(matches, np.uint32) self.execute( "INSERT INTO matches VALUES (?, ?, ?, ?)", (pair_id,) + matches.shape + (array_to_blob(matches),)) def add_two_view_geometry(self, image_id1, image_id2, matches, F=np.eye(3), E=np.eye(3), H=np.eye(3), config=2): assert(len(matches.shape) == 2) assert(matches.shape[1] == 2) if image_id1 > image_id2: matches = matches[:,::-1] pair_id = image_ids_to_pair_id(image_id1, image_id2) matches = np.asarray(matches, np.uint32) F = np.asarray(F, dtype=np.float64) E = np.asarray(E, dtype=np.float64) H = np.asarray(H, dtype=np.float64) self.execute( "INSERT INTO two_view_geometries VALUES (?, ?, ?, ?, ?, ?, ?, ?)", (pair_id,) + matches.shape + (array_to_blob(matches), config, array_to_blob(F), array_to_blob(E), array_to_blob(H))) def example_usage(): import os import argparse parser = argparse.ArgumentParser() parser.add_argument("--database_path", default="database.db") args = parser.parse_args() if os.path.exists(args.database_path): print("ERROR: database path already exists -- will not modify it.") return # Open the database. db = COLMAPDatabase.connect(args.database_path) # For convenience, try creating all the tables upfront. db.create_tables() # Create dummy cameras. model1, width1, height1, params1 = \ 0, 1024, 768, np.array((1024., 512., 384.)) model2, width2, height2, params2 = \ 2, 1024, 768, np.array((1024., 512., 384., 0.1)) camera_id1 = db.add_camera(model1, width1, height1, params1) camera_id2 = db.add_camera(model2, width2, height2, params2) # Create dummy images. image_id1 = db.add_image("image1.png", camera_id1) image_id2 = db.add_image("image2.png", camera_id1) image_id3 = db.add_image("image3.png", camera_id2) image_id4 = db.add_image("image4.png", camera_id2) # Create dummy keypoints. # # Note that COLMAP supports: # - 2D keypoints: (x, y) # - 4D keypoints: (x, y, theta, scale) # - 6D affine keypoints: (x, y, a_11, a_12, a_21, a_22) num_keypoints = 1000 keypoints1 = np.random.rand(num_keypoints, 2) * (width1, height1) keypoints2 = np.random.rand(num_keypoints, 2) * (width1, height1) keypoints3 = np.random.rand(num_keypoints, 2) * (width2, height2) keypoints4 = np.random.rand(num_keypoints, 2) * (width2, height2) db.add_keypoints(image_id1, keypoints1) db.add_keypoints(image_id2, keypoints2) db.add_keypoints(image_id3, keypoints3) db.add_keypoints(image_id4, keypoints4) # Create dummy matches. M = 50 matches12 = np.random.randint(num_keypoints, size=(M, 2)) matches23 = np.random.randint(num_keypoints, size=(M, 2)) matches34 = np.random.randint(num_keypoints, size=(M, 2)) db.add_matches(image_id1, image_id2, matches12) db.add_matches(image_id2, image_id3, matches23) db.add_matches(image_id3, image_id4, matches34) # Commit the data to the file. db.commit() # Read and check cameras. rows = db.execute("SELECT * FROM cameras") camera_id, model, width, height, params, prior = next(rows) params = blob_to_array(params, np.float64) assert camera_id == camera_id1 assert model == model1 and width == width1 and height == height1 assert np.allclose(params, params1) camera_id, model, width, height, params, prior = next(rows) params = blob_to_array(params, np.float64) assert camera_id == camera_id2 assert model == model2 and width == width2 and height == height2 assert np.allclose(params, params2) # Read and check keypoints. keypoints = dict( (image_id, blob_to_array(data, np.float32, (-1, 2))) for image_id, data in db.execute( "SELECT image_id, data FROM keypoints")) assert np.allclose(keypoints[image_id1], keypoints1) assert np.allclose(keypoints[image_id2], keypoints2) assert np.allclose(keypoints[image_id3], keypoints3) assert np.allclose(keypoints[image_id4], keypoints4) # Read and check matches. pair_ids = [image_ids_to_pair_id(*pair) for pair in ((image_id1, image_id2), (image_id2, image_id3), (image_id3, image_id4))] matches = dict( (pair_id_to_image_ids(pair_id), blob_to_array(data, np.uint32, (-1, 2))) for pair_id, data in db.execute("SELECT pair_id, data FROM matches") ) assert np.all(matches[(image_id1, image_id2)] == matches12) assert np.all(matches[(image_id2, image_id3)] == matches23) assert np.all(matches[(image_id3, image_id4)] == matches34) # Clean up. db.close() if os.path.exists(args.database_path): os.remove(args.database_path) if __name__ == "__main__": example_usage()
10,542
31.946875
89
py
GlobalSfMpy
GlobalSfMpy-main/scripts/sfm_pipeline.py
import sys import yaml import os import shutil sys.path.append('../build') import GlobalSfMpy as sfm from loss_functions import * def sfm_with_1dsfm_dataset(flagfile,dataset_path, loss_func_rotation, loss_func_position, rotation_error_type=sfm.RotationErrorType.ANGLE_AXIS, position_error_type=sfm.PositionErrorType.BASELINE, onlyRotationAvg=False): return sfm_pipeline(flagfile,dataset_path, loss_func_rotation, loss_func_position, rotation_error_type, position_error_type, onlyRotationAvg, use1DSfM = True) def sfm_pipeline(flagfile,dataset_path, loss_func_rotation, loss_func_position, rotation_error_type=sfm.RotationErrorType.ANGLE_AXIS, position_error_type=sfm.PositionErrorType.BASELINE, onlyRotationAvg=False, use1DSfM = False): if(use1DSfM): options = sfm.ReconstructionBuilderOptions() sfm.load_1DSFM_config(flagfile,options) reconstruction = sfm.Reconstruction() view_graph = sfm.ViewGraph() rot_covariances = sfm.MapEdgesCovariance() sfm.Read1DSFM(dataset_path,reconstruction,view_graph,rot_covariances) reconstruction_builder = sfm.ReconstructionBuilder(options,reconstruction,view_graph) else: database = sfm.FeaturesAndMatchesDatabase( dataset_path+"/database") options = sfm.ReconstructionBuilderOptions() sfm.load_1DSFM_config(flagfile,options) rot_covariances = sfm.MapEdgesCovariance() sfm.ReadCovariance(dataset_path,rot_covariances) reconstruction_builder = sfm.ReconstructionBuilder(options,database) sfm.AddColmapMatchesToReconstructionBuilder(dataset_path+"/two_views.txt",dataset_path+"/images/*.JPG",reconstruction_builder) reconstruction_builder.CheckView() view_graph = reconstruction_builder.get_view_graph() reconstruction = reconstruction_builder.get_reconstruction() reconstruction_estimator = sfm.GlobalReconstructionEstimator(options.reconstruction_estimator_options) # Step 1. Filter the initial view graph and remove any bad two view geometries. # Step 2. Calibrate any uncalibrated cameras. reconstruction_estimator.FilterInitialViewGraphAndCalibrateCameras(view_graph,reconstruction) # Step 3. Estimate global rotations. assert(reconstruction_estimator.EstimateGlobalRotationsUncertainty(loss_func_rotation, rot_covariances, rotation_error_type)) if(onlyRotationAvg): # Set the poses in the reconstruction object. sfm.SetOrientations( reconstruction_estimator.orientations, reconstruction ) return reconstruction # Step 4. Filter bad rotations. reconstruction_estimator.FilterRotations() # Step 5. Optimize relative translations. reconstruction_estimator.OptimizePairwiseTranslations() # Step 6. Filter bad relative translations. reconstruction_estimator.FilterRelativeTranslation() # Step 7. Estimate global positions. assert(reconstruction_estimator.EstimatePosition(loss_func_position,position_error_type)) # Set the poses in the reconstruction object. sfm.SetReconstructionFromEstimatedPoses( reconstruction_estimator.orientations, reconstruction_estimator.positions, reconstruction ) # Always triangulate once, then retriangulate and remove outliers depending # on the reconstruciton estimator options. for i in range(1+reconstruction_estimator.options.num_retriangulation_iterations): # Step 8. Triangulate features. reconstruction_estimator.EstimateStructure() sfm.SetUnderconstrainedAsUnestimated(reconstruction) # Do a single step of bundle adjustment where only the camera positions and # 3D points are refined. This is only done for the very first bundle # adjustment iteration. if i == 0 and reconstruction_estimator.options.refine_camera_positions_and_points_after_position_estimation: reconstruction_estimator.BundleAdjustCameraPositionsAndPoints() # Step 9. Bundle Adjustment. reconstruction_estimator.BundleAdjustmentAndRemoveOutlierPoints() if (not use1DSfM) and os.path.exists(dataset_path+"/database"): shutil.rmtree(dataset_path+"/database") return reconstruction if __name__ == '__main__': if len(sys.argv) < 2: print('Usage: sfm_with_1dsfm_dataset.py <flagfile>') sys.exit(1) flagfile = sys.argv[1] f = open(flagfile,"r") config = yaml.safe_load(f) dataset_path = config['1dsfm_dataset_directory'] output_reconstruction = config['output_reconstruction'] glog_directory = config['glog_directory'] glog_verbose = config['v'] log_to_stderr = config['log_to_stderr'] sfm.InitGlog(glog_verbose,log_to_stderr,glog_directory) position_error_type = sfm.PositionErrorType.BASELINE rotation_error_type = sfm.RotationErrorType.ANGLE_AXIS_COVARIANCE if not os.path.exists(dataset_path+"/covariance_rot.txt"): sfm.CalcCovariance(dataset_path) loss_func = MAGSACWeightBasedLoss(0.02) reconstruction = sfm_with_1dsfm_dataset(flagfile,dataset_path, loss_func,HuberLoss(0.1), rotation_error_type,position_error_type, onlyRotationAvg=False) if os.path.exists(output_reconstruction): os.remove(output_reconstruction) sfm.WriteReconstruction(reconstruction, output_reconstruction) sfm.StopGlog()
6,221
41.040541
134
py
GlobalSfMpy
GlobalSfMpy-main/scripts/loss_functions.py
from math import sqrt,log,atan2,exp,cosh from pickle import FALSE from re import S import sys sys.path.append('../build') import GlobalSfMpy as sfm # For a residual vector with squared 2-norm 'sq_norm', this method # is required to fill in the value and derivatives of the loss # function (rho in this example): # # out[0] = rho(sq_norm), # out[1] = rho'(sq_norm), # out[2] = rho''(sq_norm), # # Here the convention is that the contribution of a term to the # cost function is given by 1/2 rho(s), where # # s = ||residuals||^2. # # Calling the method with a negative value of 's' is an error and # the implementations are not required to handle that case. # # Most sane choices of rho() satisfy: # # rho(0) = 0, # rho'(0) = 1, # rho'(s) < 1 in outlier region, # rho''(s) < 0 in outlier region, # # so that they mimic the least squares cost for small residuals. # Scaling # ------- # Given one robustifier # s -> rho(s) # one can change the length scale at which robustification takes # place, by adding a scale factor 'a' as follows: # # s -> a^2 rho(s / a^2). # # The first and second derivatives are: # # s -> rho'(s / a^2), # s -> (1 / a^2) rho''(s / a^2), class TrivialLoss(sfm.LossFunction): # rho(s)=s def __init__(self): super().__init__() def Evaluate(self, sq_norm, out): out[0] = sq_norm; out[1] = 1.0 out[2] = 0.0 class HuberLoss(sfm.LossFunction): # rho(s) = s for s <= 1, # rho(s) = 2 sqrt(s) - 1 for s >= 1. def __init__(self,a): self.a = a self.b = a*a super().__init__() def Evaluate(self, sq_norm, out): if sq_norm > self.b: r = sqrt(sq_norm) out[0] = 2*self.a*r - self.b out[1] = max(self.a/r, sys.float_info.min) out[2] = -out[1]/(2.0*sq_norm) else: out[0] = sq_norm out[1] = 1.0 out[2] = 0.0 class SoftLOneLoss(sfm.LossFunction): # rho(s) = 2 (sqrt(1 + s) - 1). def __init__(self,a): self.a = a self.b = a*a self.c = 1.0/self.b super().__init__() def Evaluate(self, sq_norm, out): sum = 1.0 + sq_norm*self.c tmp = sqrt(sum) out[0] = 2.0 * self.b * (tmp - 1.0) out[1] = max(1.0/tmp, sys.float_info.min) out[2] = -(self.c*out[1])/(2.0*sum) class CauchyLoss(sfm.LossFunction): # rho(s) = log(1 + s). def __init__(self,a): self.b = a*a self.c = 1.0/self.b super().__init__() def Evaluate(self, sq_norm, out): sum = 1.0 + sq_norm*self.c inv = 1.0/sum out[0] = self.b * log(sum) out[1] = max(inv, sys.float_info.min) out[2] = -self.c * (inv * inv) class ArctanLoss(sfm.LossFunction): # rho(s) = a arctan(s / a). def __init__(self,a): self.a = a self.b = 1 / (a*a) super().__init__() def Evaluate(self, sq_norm, out): sum = 1 + sq_norm*sq_norm*self.b inv = 1.0/sum out[0] = self.a * atan2(sq_norm, self.a) out[1] = max(inv, sys.float_info.min) out[2] = -2.0 * sq_norm * self.b * (inv * inv) class TolerantLoss(sfm.LossFunction): # Loss function that maps to approximately zero cost in a range around the # origin, and reverts to linear in error (quadratic in cost) beyond this range. # The tolerance parameter 'a' sets the nominal point at which the # transition occurs, and the transition size parameter 'b' sets the nominal # distance over which most of the transition occurs. Both a and b must be # greater than zero, and typically b will be set to a fraction of a. # The slope rho'[s] varies smoothly from about 0 at s <= a - b to # about 1 at s >= a + b. # # The term is computed as: # # rho(s) = b log(1 + exp((s - a) / b)) - c0. # # where c0 is chosen so that rho(0) == 0 # # c0 = b log(1 + exp(-a / b) # # This has the following useful properties: # # rho(s) == 0 for s = 0 # rho'(s) ~= 0 for s << a - b # rho'(s) ~= 1 for s >> a + b # rho''(s) > 0 for all s # # In addition, all derivatives are continuous, and the curvature is # concentrated in the range a - b to a + b. # # At s = 0: rho = [0, ~0, ~0]. def __init__(self,a,b): self.a = a self.b = b self.c = b * log(1 + exp(-a / b)) assert(a >= 0) assert(b > 0) super().__init__() def Evaluate(self, sq_norm, out): x = (sq_norm - self.a) / self.b; # The basic equation is rho[0] = b ln(1 + e^x). However, if e^x is too # large, it will overflow. Since numerically 1 + e^x == e^x when the # x is greater than about ln(2^53) for doubles, beyond this threshold # we substitute x for ln(1 + e^x) as a numerically equivalent approximation. kLog2Pow53 = 36.7; # ln(MathLimits<double>::kEpsilon). if (x > kLog2Pow53): out[0] = sq_norm - self.a - self.c; out[1] = 1.0; out[2] = 0.0; else: e_x = exp(x); out[0] = self.b * log(1.0 + e_x) - self.c; out[1] = max(e_x / (1.0 + e_x), sys.float_info.min); out[2] = 0.5 / (self.b * (1.0 + cosh(x))); class TukeyLoss(sfm.LossFunction): # rho(s) = a^2 / 6 * (1 - (1 - s / a^2)^3 ) for s <= a^2, # rho(s) = a^2 / 6 for s > a^2. def __init__(self,a): self.a_squared = a*a super().__init__() def Evaluate(self, sq_norm, out): if sq_norm <= self.a_squared: # Inlier region. value = 1.0 - sq_norm / self.a_squared; value_sq = value * value; out[0] = self.a_squared / 6.0 * (1.0 - value_sq * value); out[1] = 0.5 * value_sq; out[2] = -1.0 / self.a_squared * value; else: # Outlier region. out[0] = self.a_squared / 6.0; out[1] = 0.0; out[2] = 0.0; class LOneHalfLoss(sfm.LossFunction): # Scaling # ------- # Given one robustifier # s -> rho(s) # one can change the length scale at which robustification takes # place, by adding a scale factor 'a' as follows: # # s -> a^2 rho(s / a^2). # # The first and second derivatives are: # # s -> rho'(s / a^2), # s -> (1 / a^2) rho''(s / a^2), # rho(s) = 2*|s|^0.5 # rho'(s) = sign(s)|s|^(-0.5) # rho''(s) = |s|^(-1.5) def __init__(self,a): self.a = a self.sqrt_a = sqrt(a) super().__init__() def Evaluate(self, sq_norm, out): out[0] = 2.0*self.a*self.sqrt_a*pow(sq_norm,0.25) if sq_norm < 0.01: sq_norm = 0.01 out[1] = 0.5*pow(self.a,-1.5)*pow(sq_norm,-0.75) out[2] = -0.375*self.a*self.sqrt_a*pow(sq_norm,-1.75) class LTwoLoss(sfm.LossFunction): # Scaling # ------- # Given one robustifier # s -> rho(s) # one can change the length scale at which robustification takes # place, by adding a scale factor 'a' as follows: # # s -> a^2 rho(s / a^2). # # The first and second derivatives are: # # s -> rho'(s / a^2), # s -> (1 / a^2) rho''(s / a^2), def __init__(self,a,sigma2): self.a_sq = a*a super().__init__() def Evaluate(self, sq_norm, out): out[0] = sq_norm*sq_norm/(self.a_sq*2.0) out[1] = sq_norm/self.a_sq out[2] = 1/self.a_sq class GemanMcClureLoss(sfm.LossFunction): def __init__(self, a, sigma2): self.a_sq = a*a self.sigma2 = sigma2 super().__init__() def Evaluate(self, sq_norm, out): out[0] = self.a_sq * self.sigma2*sq_norm / (2.0*(sq_norm+self.a_sq*self.sigma2)) out[1] = (self.sigma2**2) / (2.0*(sq_norm/self.a_sq + self.sigma2)**2) out[2] = -(self.sigma2**2) / (self.a_sq*(sq_norm/self.a_sq+self.sigma2)**3) class ComposedLoss(sfm.LossFunction): # rho(s) = f(g(s)) def __init__(self,f:sfm.LossFunction, g:sfm.LossFunction): self.f = f self.g = g super().__init__() def Evaluate(self, sq_norm, out): out_f = [0,0,0] out_g = [0,0,0] self.g.Evaluate(sq_norm, out_g) self.f.Evaluate(out_g[0], out_f) out[0] = out_f[0] # f'(g(s)) * g'(s). out[1] = out_f[1] * out_g[1]; # f''(g(s)) * g'(s) * g'(s) + f'(g(s)) * g''(s). out[2] = out_f[2] * out_g[1] * out_g[1] + out_f[1] * out_g[2] class ScaledLoss(sfm.LossFunction): # If rho is the wrapped robustifier, then this simply outputs # s -> a * rho(s) # The first and second derivatives are, not surprisingly # s -> a * rho'(s) # s -> a * rho''(s) def __init__(self, rho:sfm.LossFunction, a): self.rho = rho self.a = a super().__init__() def Evaluate(self, sq_norm, out): self.rho.Evaluate(sq_norm, out) out[0] *= self.a out[1] *= self.a out[2] *= self.a class MAGSACWeightBasedLoss(sfm.LossFunction): def __init__(self,sigma,inverse = False): self.sigma_max = sigma self.nu = sfm.nu3 self.squared_sigma = self.sigma_max * self.sigma_max self.squared_sigma_max_2 = 2.0 * self.squared_sigma self.cubed_sigma_max = self.squared_sigma*self.sigma_max self.dof_minus_one_per_two = (self.nu - 1.0) / 2.0 self.C_times_two_ad_dof = sfm.C3 * (2**self.dof_minus_one_per_two) self.one_over_sigma = self.C_times_two_ad_dof / self.sigma_max self.gamma_value = sfm.tgamma(self.dof_minus_one_per_two) self.gamma_difference = self.gamma_value - sfm.upper_incomplete_gamma_of_k3 self.weight_zero = self.one_over_sigma * self.gamma_difference self.use_weight_inverse = inverse super().__init__() def Evaluate(self, squared_residual, rho): zero_derivative = False; if (squared_residual>sfm.sigma_quantile3*sfm.sigma_quantile3*self.squared_sigma): squared_residual = sfm.sigma_quantile3*sfm.sigma_quantile3*self.squared_sigma zero_derivative = True x = round(sfm.precision_of_stored_gamma3 * squared_residual / self.squared_sigma_max_2) if sfm.stored_gamma_number3 < x: x = sfm.stored_gamma_number3 s = x * self.squared_sigma_max_2 / sfm.precision_of_stored_gamma3 weight = self.one_over_sigma * (sfm.stored_gamma_values3[x] - sfm.upper_incomplete_gamma_of_k3) weight_derivative = -self.C_times_two_ad_dof\ * ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5))\ * exp(-s/self.squared_sigma_max_2) / (2*self.cubed_sigma_max) if(s < 1e-7): s = 1e-7 weight_second_derivative = 2.0 * self.C_times_two_ad_dof *\ ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5)) *\ (1.0/self.squared_sigma - (self.nu-3)/s) *\ exp(-s/self.squared_sigma_max_2) / (8*self.cubed_sigma_max); if self.use_weight_inverse: rho[0] = 1.0/weight; rho[1] = -1.0/(weight*weight)*weight_derivative; rho[2] = 2.0/(weight*weight*weight)*weight_derivative*weight_derivative\ - weight_second_derivative/(weight*weight); if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0; else: rho[0] = self.weight_zero-weight rho[1] = -weight_derivative rho[2] = -weight_second_derivative if rho[1] == 0: rho[1] = 0.00001; if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0; class MAGSACWeightBasedLoss4(sfm.LossFunction): def __init__(self,sigma,inverse = True): self.sigma_max = sigma self.nu = sfm.nu4 self.squared_sigma = self.sigma_max * self.sigma_max self.squared_sigma_max_2 = 2.0 * self.squared_sigma self.cubed_sigma_max = self.squared_sigma*self.sigma_max self.dof_minus_one_per_two = (self.nu - 1.0) / 2.0 self.C_times_two_ad_dof = sfm.C4 * (2**self.dof_minus_one_per_two) self.one_over_sigma = self.C_times_two_ad_dof / self.sigma_max self.gamma_value = sfm.tgamma(self.dof_minus_one_per_two) self.gamma_difference = self.gamma_value - sfm.upper_incomplete_gamma_of_k4 self.weight_zero = self.one_over_sigma * self.gamma_difference self.use_weight_inverse = inverse super().__init__() def Evaluate(self, squared_residual, rho): zero_derivative = False; if (squared_residual>sfm.sigma_quantile4*sfm.sigma_quantile4*self.squared_sigma): squared_residual = sfm.sigma_quantile4*sfm.sigma_quantile4*self.squared_sigma zero_derivative = True x = round(sfm.precision_of_stored_gamma4 * squared_residual / self.squared_sigma_max_2) if sfm.stored_gamma_number4 < x: x = sfm.stored_gamma_number4 s = x * self.squared_sigma_max_2 / sfm.precision_of_stored_gamma4 weight = self.one_over_sigma * (sfm.stored_gamma_values4[x] - sfm.upper_incomplete_gamma_of_k4) weight_derivative = -self.C_times_two_ad_dof\ * ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5))\ * exp(-s/self.squared_sigma_max_2) / (2*self.cubed_sigma_max) if(s < 1e-7): s = 1e-7 weight_second_derivative = 2.0 * self.C_times_two_ad_dof *\ ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5)) *\ (1.0/self.squared_sigma - (self.nu-3)/s) *\ exp(-s/self.squared_sigma_max_2) / (8*self.cubed_sigma_max); if self.use_weight_inverse: rho[0] = 1.0/weight; rho[1] = -1.0/(weight*weight)*weight_derivative; rho[2] = 2.0/(weight*weight*weight)*weight_derivative*weight_derivative\ - weight_second_derivative/(weight*weight); if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0; else: rho[0] = self.weight_zero-weight rho[1] = -weight_derivative rho[2] = -weight_second_derivative if rho[1] == 0: rho[1] = 0.00001; if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0; class MAGSACWeightBasedLoss9(sfm.LossFunction): def __init__(self,sigma,inverse = False): self.sigma_max = sigma self.nu = sfm.nu9 self.squared_sigma = self.sigma_max * self.sigma_max self.squared_sigma_max_2 = 2.0 * self.squared_sigma self.cubed_sigma_max = self.squared_sigma*self.sigma_max self.dof_minus_one_per_two = (self.nu - 1.0) / 2.0 self.C_times_two_ad_dof = sfm.C9 * (2**self.dof_minus_one_per_two) self.one_over_sigma = self.C_times_two_ad_dof / self.sigma_max self.gamma_value = sfm.tgamma(self.dof_minus_one_per_two) self.gamma_difference = self.gamma_value - sfm.upper_incomplete_gamma_of_k9 self.weight_zero = self.one_over_sigma * self.gamma_difference self.use_weight_inverse = inverse super().__init__() def Evaluate(self, squared_residual, rho): zero_derivative = False; if (squared_residual>sfm.sigma_quantile9*sfm.sigma_quantile9*self.squared_sigma): squared_residual = sfm.sigma_quantile9*sfm.sigma_quantile9*self.squared_sigma zero_derivative = True x = round(sfm.precision_of_stored_gamma9 * squared_residual / self.squared_sigma_max_2) if sfm.stored_gamma_number9 < x: x = sfm.stored_gamma_number9 s = x * self.squared_sigma_max_2 / sfm.precision_of_stored_gamma9 weight = self.one_over_sigma * (sfm.stored_gamma_values9[x] - sfm.upper_incomplete_gamma_of_k9) weight_derivative = -self.C_times_two_ad_dof\ * ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5))\ * exp(-s/self.squared_sigma_max_2) / (2*self.cubed_sigma_max) if(s < 1e-7): s = 1e-7 weight_second_derivative = 2.0 * self.C_times_two_ad_dof *\ ((s/self.squared_sigma_max_2)**(self.nu/2 - 1.5)) *\ (1.0/self.squared_sigma - (self.nu-3)/s) *\ exp(-s/self.squared_sigma_max_2) / (8*self.cubed_sigma_max); if self.use_weight_inverse: rho[0] = 1.0/weight; rho[1] = -1.0/(weight*weight)*weight_derivative; rho[2] = 2.0/(weight*weight*weight)*weight_derivative*weight_derivative\ - weight_second_derivative/(weight*weight); if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0; else: rho[0] = self.weight_zero-weight rho[1] = -weight_derivative rho[2] = -weight_second_derivative if rho[1] == 0: rho[1] = 0.00001; if zero_derivative: rho[1] = 0.00001; rho[2] = 0.0;
17,372
36.849673
103
py
GlobalSfMpy
GlobalSfMpy-main/scripts/get_covariance_from_colmap.py
import sys import yaml import os sys.path.append('../build') import GlobalSfMpy as sfm dataset_name = "facade" flagfile = "../flags_1dsfm.yaml" dataset_path = "../datasets/"+dataset_name if os.path.exists(dataset_path+"/covariance_rot.txt"): print("Covariance already exists!") exit() f = open(flagfile,"r") config = yaml.safe_load(f) glog_directory = config['glog_directory'] glog_verbose = config['v'] log_to_stderr = False sfm.InitGlog(glog_verbose,log_to_stderr,glog_directory) database = sfm.FeaturesAndMatchesDatabase( dataset_path+"/database") options = sfm.ReconstructionBuilderOptions() sfm.load_1DSFM_config(flagfile,options) reconstruction_builder = sfm.ReconstructionBuilder(options,database) sfm.AddColmapMatchesToReconstructionBuilder(dataset_path+"/two_views.txt",reconstruction_builder,database) reconstruction_builder.CheckView() view_graph = reconstruction_builder.get_view_graph() reconstruction = reconstruction_builder.get_reconstruction() sfm.store_covariance_rot(dataset_path,reconstruction,view_graph) sfm.StopGlog()
1,071
26.487179
106
py
GlobalSfMpy
GlobalSfMpy-main/scripts/read_colmap_database.py
from colmap_database import * import os import argparse from scipy.spatial.transform import Rotation import cv2 as cv import poselib parser = argparse.ArgumentParser() parser.add_argument("--dataset_path", default="../datasets/facade") args = parser.parse_args() dataset_path = args.dataset_path class Pose: def __init__(self,qw,qx,qy,qz,tx,ty,tz): self.qw = qw self.qx = qx self.qy = qy self.qz = qz self.tx = tx self.ty = ty self.tz = tz def rotation_matrix(self): R = Rotation.from_quat([self.qx,self.qy,self.qz,self.qw]) return R.as_matrix() def translation(self): return np.array([[self.tx],[self.ty],[self.tz]]) f_camera = open(dataset_path+"/cameras.txt") cameras = {} lines = f_camera.readlines() for line in lines[3:]: words = line.split() camera_id = np.uint32(words[0]) camera_width = np.int32(words[2]) camera_height = np.int32(words[3]) fx = np.float64(words[4]) fy = np.float64(words[5]) ux = np.float64(words[6]) uy = np.float64(words[7]) cameras[camera_id] = [camera_width,camera_height,fx,fy,ux,uy] f_camera.close() f_images = open(dataset_path+"/images.txt") images = {} lines = f_images.readlines()[4:] for i in range(len(lines)): if i%2 == 1: continue words = lines[i].split() image_name = words[-1].split("/")[-1] camera_id = np.uint32(words[-2]) qw = np.float64(words[1]) qx = np.float64(words[2]) qy = np.float64(words[3]) qz = np.float64(words[4]) tx = np.float64(words[5]) ty = np.float64(words[6]) tz = np.float64(words[7]) images[image_name]=[camera_id,Pose(qw,qx,qy,qz,tx,ty,tz)] f_images.close() # Open the database. database_path = dataset_path+"/colmap/database.db" db = COLMAPDatabase.connect(database_path) two_view_geometries = db.execute("SELECT * FROM two_view_geometries") write_path = dataset_path + "/two_views.txt" f_write = open(write_path,"w") f_write.write("# img_name1 image_name2 f1 f2 num_inlier rot[0] rot[1] rot[2] trans[0] trans[1] trans[2]\n" +"# features1 [p0x p0y p1x p1y ...]\n"+ "# features2 [p0x p0y p1x p1y ...]\n") for two_view_geometry in two_view_geometries: pair_id = two_view_geometry[0] if two_view_geometry[3]== None: continue matches = blob_to_array(two_view_geometry[3],np.uint32,(-1,2)) image_id1,image_id2 = pair_id_to_image_ids(pair_id) image_id1 = np.uint32(image_id1) image1 = db.execute("SELECT name FROM images WHERE image_id={0}".format(image_id1)) image2 = db.execute("SELECT name FROM images WHERE image_id={0}".format(image_id2)) image1_name = next(image1)[0] image2_name = next(image2)[0] image1_name = image1_name.split("/")[-1] image2_name = image2_name.split("/")[-1] F = blob_to_array(two_view_geometry[5],np.float64,(3,3)) feature1 = db.execute("SELECT data FROM keypoints WHERE image_id={0}".format(image_id1)) feature1_cols = db.execute("SELECT cols FROM keypoints WHERE image_id={0}".format(image_id1)) feature1_cols = next(feature1_cols)[0] feature1 = next(feature1)[0] feature1 = blob_to_array(feature1,np.float32,(-1,feature1_cols)) feature2 = db.execute("SELECT data FROM keypoints WHERE image_id={0}".format(image_id2)) feature2_cols = db.execute("SELECT cols FROM keypoints WHERE image_id={0}".format(image_id2)) feature2_cols = next(feature2_cols)[0] feature2 = next(feature2)[0] feature2 = blob_to_array(feature2,np.float32,(-1,feature2_cols)) feature1 = feature1[:,0:2] feature2 = feature2[:,0:2] corrresponding1 = feature1[matches[:,0],:] corrresponding2 = feature2[matches[:,1],:] camera1 = cameras[images[image1_name][0]] camera2 = cameras[images[image2_name][0]] pose1 = images[image1_name][1] pose2 = images[image2_name][1] camera1 = {'model': 'PINHOLE', 'width': camera1[0], 'height': camera1[1], 'params': [camera1[2],camera1[3],camera1[4],camera1[5]]} camera2 = {'model': 'PINHOLE', 'width': camera2[0], 'height': camera2[1], 'params': [camera2[2],camera2[3],camera2[4],camera2[5]]} K1 = np.array([[camera1['params'][0],0.0,camera1['params'][2]], [0.0,camera1['params'][1],camera1['params'][3]], [0,0,1]]) K2 = np.array([[camera2['params'][0],0.0,camera2['params'][2]], [0.0,camera2['params'][1],camera2['params'][3]], [0,0,1]]) result = poselib.estimate_relative_pose(corrresponding1,corrresponding2,camera1,camera2) relative_pose = result[0] num_inliers = result[1]["num_inliers"] inliers_mask = result[1]["inliers"] point1 = corrresponding1[inliers_mask,:] point2 = corrresponding2[inliers_mask,:] rotation = Rotation.from_quat([relative_pose.q[1],relative_pose.q[2],relative_pose.q[3],relative_pose.q[0]]).as_rotvec() translation = relative_pose.t f1 = (camera1['params'][0]+camera1['params'][1])/2.0 f2 = (camera2['params'][0]+camera2['params'][1])/2.0 f_write.write("{} {} {} {} {} {} {} {} {} {} {}\n".format(image1_name,image2_name,f1,f2,num_inliers, rotation[0],rotation[1],rotation[2], translation[0],translation[1],translation[2])) point1_string = "" point2_string = "" for i in range(num_inliers): point1_string += "{} {} ".format(point1[i,0],point1[i,1]) point2_string += "{} {} ".format(point2[i,0],point2[i,1]) point1_string+="\n" point2_string+="\n" f_write.write(point1_string) f_write.write(point2_string) f_write.close() db.close()
5,742
35.814103
134
py
GlobalSfMpy
GlobalSfMpy-main/scripts/sfm_with_colmap_feature.py
import sys import yaml import os sys.path.append('../build') import GlobalSfMpy as sfm from loss_functions import * from sfm_pipeline import * flagfile = "../flags_1dsfm.yaml" f = open(flagfile,"r") config = yaml.safe_load(f) glog_directory = config['glog_directory'] glog_verbose = config['v'] log_to_stderr = config['log_to_stderr'] sfm.InitGlog(glog_verbose,log_to_stderr,glog_directory) dataset_name = "facade" # dataset_path = "../datasets/"+dataset_name dataset_path = "/home/zhangganlin/Desktop/CVG/final_version_globalsfmpy/datasets/"+dataset_name colmap_path = dataset_path+"/colmap/images.txt" output_reconstruction = "../output/"+dataset_name rotation_error_type = sfm.RotationErrorType.ANGLE_AXIS_COVARIANCE position_error_type = sfm.PositionErrorType.BASELINE reconstruction = sfm_pipeline(flagfile,dataset_path, MAGSACWeightBasedLoss(0.02),HuberLoss(0.1), rotation_error_type,position_error_type, onlyRotationAvg=False) if os.path.exists(output_reconstruction): os.remove(output_reconstruction) sfm.WriteReconstruction(reconstruction, output_reconstruction) sfm.StopGlog()
1,233
29.85
95
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/main.py
import argparse from os import path, makedirs from experiments import select_experiment import torch import yaml import os def create_dir_structure(config): subdirs = ["ckpt", "config", "generated", "log"] structure = {subdir: path.join(config["base_dir"],config["experiment"],subdir,config["project_name"]) for subdir in subdirs} if "DATAPATH" in os.environ: structure = {subdir: path.join(os.environ["DATAPATH"],structure[subdir]) for subdir in structure} return structure def load_parameters(config_name, restart,debug,project_name): with open(config_name,"r") as f: cdict = yaml.load(f,Loader=yaml.FullLoader) if debug: cdict['general']['project_name'] = 'debug' else: cdict['general']['project_name'] = project_name dir_structure = create_dir_structure(cdict["general"]) saved_config = path.join(dir_structure["config"], "config.yaml") if restart: if path.isfile(saved_config): with open(saved_config,"r") as f: cdict = yaml.load(f, Loader=yaml.FullLoader) else: raise FileNotFoundError("No saved config file found but model is intended to be restarted. Aborting....") else: [makedirs(dir_structure[d],exist_ok=True) for d in dir_structure] if path.isfile(saved_config) and not debug: print(f"\033[93m" + "WARNING: Model has been started somewhen earlier: Resume training (y/n)?" + "\033[0m") while True: answer = input() if answer == "y" or answer == "yes": with open(saved_config,"r") as f: cdict = yaml.load(f, Loader=yaml.FullLoader) restart = True break elif answer == "n" or answer == "no": with open(saved_config, "w") as f: yaml.dump(cdict, f, default_flow_style=False) break else: print(f"\033[93m" + "Invalid answer! Try again!(y/n)" + "\033[0m") else: with open(saved_config, "w") as f: yaml.dump(cdict,f,default_flow_style=False) cdict['general']['debug'] = debug return cdict, dir_structure, restart if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument("-c", "--config", type=str, default="config/latent_flow_net.yaml", help="Define config file") parser.add_argument('-p','--project_name',type=str,default='ii2v',help='unique name for the training run to be (re-)started.') parser.add_argument("-r","--restart", default=False,action="store_true",help="Whether training should be resumed.") parser.add_argument("-d", "--debug", default=False, action="store_true", help="Whether training should be resumed.") parser.add_argument("--gpu",default=[0], type=int, nargs="+",help="GPU to use.") parser.add_argument("-m","--mode",default="train",type=str,choices=["train","test"],help="Whether to start in train or infer mode?") parser.add_argument("--test_mode",default="metrics",type=str, choices=["noise_test","metrics","fvd",'diversity','render'], help="The mode in which the test-method should be executed.") parser.add_argument("--metrics_on_patches", default=False,action="store_true",help="Whether to run evaluation on patches (if available or not).") parser.add_argument("--best_ckpt", default=False, action="store_true",help="Whether to use the best ckpt as measured by LPIPS (otherwise, latest_ckpt is used)") args = parser.parse_args() config, structure, restart = load_parameters(args.config, args.restart or args.mode == "test",args.debug,args.project_name) config["general"]["restart"] = restart config["general"]["mode"] = args.mode # config["general"]["first_stage"] = args.first_stage if len(args.gpu) == 1: gpus = torch.device( f"cuda:{int(args.gpu[0])}" if torch.cuda.is_available() and int(args.gpu[0]) >= 0 else "cpu" ) torch.cuda.set_device(gpus) else: gpus = [int(id) for id in args.gpu] mode = config["general"]["mode"] config["testing"].update({"best_ckpt": args.best_ckpt}) if mode == "test" and "testing" in config and "metrics_on_patches" in config["testing"]: config["testing"]["metrics_on_patches"] = args.metrics_on_patches experiment = select_experiment(config, structure, gpus) # start selected experiment if mode == "train": experiment.train() elif mode == "test": config["testing"].update({"mode": args.test_mode}) experiment.test() else: raise ValueError(f"\"mode\"-parameter should be either \"train\" or \"infer\" but is actually {mode}")
4,862
44.448598
188
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/models/discriminator.py
import torch from torch import nn from torch.optim import Adam import functools from torch.nn.utils import spectral_norm import math import numpy as np from utils.general import get_member from models.blocks import SPADE class GANTrainer(object): def __init__(self, config, load_fn,logger,spatial_size=128, parallel=False, devices=None, debug=False, temporal=False, sequence_length = None): self.config = config self.logger = logger # disc self.logger.info("Load discriminator model") self.temporal = temporal if self.temporal: assert sequence_length is not None self.key = "gan_temp" self.disc = resnet(config=config[self.key],spatial_size=spatial_size,sequence_length=sequence_length) self.load_key = "disc_temp" self.postfix = "temp" if self.disc.cond: self.logger.info(f"Using Conditional temporal discriminator.") else: self.key = "gan" self.disc = PatchDiscriminator(self.config[self.key]) self.load_key = "disc_patch" self.postfix = "patch" self.cond = self.config[self.key]["conditional"] if self.temporal and "conditional" in self.config[self.key] else False self.logger.info(f"Number of parameters in discriminator_{self.postfix} is {sum(p.numel() for p in self.disc.parameters())}.") self.parallel = parallel self.devices = devices if self.parallel: assert self.devices is not None # load checkpoint if there's any and it is required disc_ckpt = disc_op_ckpt = None if self.config["general"]["restart"] and not debug: disc_ckpt, disc_op_ckpt = load_fn(key=self.load_key) if disc_ckpt is not None: self.logger.info(f"Resuming training of discriminator...loading weights.") self.disc.load_state_dict(disc_ckpt) if self.parallel: self.disc = nn.DataParallel(self.disc,device_ids=self.devices) self.disc.cuda(self.devices[0]) else: self.disc.cuda() self.logger.info("Discriminator on gpu!") # disc optimizer self.disc_opt = Adam(self.disc.parameters(), lr=self.config["training"]["lr"]) if self.config["general"]["restart"] and disc_op_ckpt is not None: self.disc_opt.load_state_dict(disc_op_ckpt) # scheduler for disc optimizer milestones = [int(self.config["training"]["n_epochs"] * t) for t in self.config["training"]["tau"]] self.disc_scheduler = torch.optim.lr_scheduler.MultiStepLR(self.disc_opt, milestones=milestones, gamma=self.config["training"]["lr_reduce"]) def train_step(self, x_in_true, x_in_fake, cond=None): # predict cond = cond if self.cond else None self.disc.train() # if self.parallel: # x_in_true = x_in_true.cuda(self.devices[0]) # x_in_fake = x_in_fake.cuda(self.devices[0]) # set gradient to zero self.disc_opt.zero_grad() # real examples x_in_true.requires_grad_() pred_true, _ = self.disc(x_in_true, cond) loss_real = get_member(self.disc,"loss")(pred_true, real=True) if self.config[self.key]["gp_weight"] > 0: loss_real.backward(retain_graph=True) # gradient penalty loss_gp = get_member(self.disc,"gp")(pred_true, x_in_true).mean() gp_weighted = self.config[self.key]["gp_weight"] * loss_gp gp_weighted.backward() else: loss_real.backward() # fake examples pred_fake, _ = self.disc(x_in_fake.detach(),cond) loss_fake = get_member(self.disc,"loss")(pred_fake, real=False) loss_fake.backward() # optmize parameters self.disc_opt.step() loss_disc = ((loss_real + loss_fake) / 2.).item() out_dict = {f"loss_disc_{self.postfix}": loss_disc, f"p_true_{self.postfix}": torch.sigmoid(pred_true).mean().item(), f"p_fake_{self.postfix}": torch.sigmoid(pred_fake).mean().item(), f"loss_gp_{self.postfix}": loss_gp.item() if self.config[self.key]["gp_weight"] > 0 else 0 } # train generator pred_fake, fmap_fake = self.disc(x_in_fake,cond) _, fmap_true = self.disc(x_in_true,cond) if get_member(self.disc,"bce_loss"): loss_gen = get_member(self.disc,"bce")(pred_fake, torch.ones_like(pred_fake)) else: loss_gen = -torch.mean(pred_fake) loss_fmap = get_member(self.disc,"fmap_loss")(fmap_fake, fmap_true) # if self.parallel: # loss_fmap = loss_fmap.cuda(self.devices[0]) # loss_gen = loss_gen.cuda(self.devices[0]) return out_dict, loss_gen, loss_fmap # code taken from https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/models/networks.py class PatchDiscriminator(nn.Module): """Defines a PatchGAN discriminator""" def __init__(self, config, norm_layer=nn.InstanceNorm2d): super().__init__() if type(norm_layer) == functools.partial: # no need to use bias as BatchNorm2d has affine parameters use_bias = norm_layer.func == nn.InstanceNorm2d else: use_bias = norm_layer == nn.InstanceNorm2d deep_disc = config["deep_disc"] if "deep_disc" in config else False input_nc = 6 if config["pixel_dynamics"] else 3 n_deep_layers = config["deep_layers"] ndf = 64 n_layers = config["n_layers"] self.bce_loss = config["bce_loss"] if self.bce_loss: self.bce = nn.BCEWithLogitsLoss() kw = 4 padw = 1 self.layers = nn.ModuleList() self.norms = nn.ModuleList() self.in_conv = nn.Conv2d(input_nc, ndf, kernel_size=kw, stride=2, padding=padw) nf_mult = 1 nf_mult_prev = 1 self.act_fn = nn.LeakyReLU(0.2, True) for n in range(1, n_layers): # gradually increase the number of filters nf_mult_prev = nf_mult nf_mult = min(2 ** n, 8) self.layers.append(nn.Conv2d(ndf * nf_mult_prev, ndf * nf_mult, kernel_size=kw, stride=2, padding=padw, bias=use_bias)) self.norms.append(norm_layer(ndf * nf_mult)) nf_mult_prev = nf_mult nf_mult = min(2 ** n_layers, 8) self.layers.append(nn.Conv2d(ndf * nf_mult_prev, ndf * nf_mult, kernel_size=kw, stride=1, padding=padw, bias=use_bias)) self.norms.append(norm_layer(ndf * nf_mult)) n_d = ndf * nf_mult if deep_disc: n_max = 1024 for i in range(n_deep_layers): # add one layer to the original patch discrminator to make it more powerful self.layers.append(nn.Conv2d(n_d, min(n_max, n_d*2), kernel_size=kw, stride=1, padding=padw, bias=use_bias)) self.norms.append(norm_layer(min(n_max, n_d*2))) n_d = min(n_max, n_d*2) self.out_conv = nn.Conv2d(n_d, 1, kernel_size=kw, stride=1, padding=padw) # output 1 channel prediction map def forward(self, input,cond=None): """Standard forward.""" x = self.act_fn(self.in_conv(input)) fmap = [] for i in range(len(self.layers)): x = self.layers[i](x) x = self.act_fn(self.norms[i](x)) fmap.append(x) x = self.out_conv(x) return x, fmap def loss(self, pred, real): if self.bce_loss: # vanilla gan loss return self.bce(pred, torch.ones_like(pred) if real else torch.zeros_like(pred)) else: # hinge loss if real: l = torch.mean(torch.nn.ReLU()(1.0 - pred)) else: l = torch.mean(torch.nn.ReLU()(1.0 + pred)) return l def gp(self, pred_fake, x_fake): batch_size = x_fake.size(0) grad_dout = torch.autograd.grad( outputs=pred_fake.sum(), inputs=x_fake, create_graph=True, retain_graph=True, only_inputs=True )[0] grad_dout2 = grad_dout.pow(2) assert (grad_dout2.size() == x_fake.size()) reg = grad_dout2.view(batch_size, -1).sum(1) return reg def fmap_loss(self, fmap1, fmap2, loss="l1"): recp_loss = 0 for idx in range(len(fmap1)): if loss == "l1": recp_loss += torch.mean(torch.abs((fmap1[idx] - fmap2[idx]))) if loss == "l2": recp_loss += torch.mean((fmap1[idx] - fmap2[idx]) ** 2) return recp_loss / len(fmap1) ###################################################################################################### ###3D-ConvNet Implementation from https://github.com/tomrunia/PyTorchConv3D ########################## def resnet10(**kwargs): """Constructs a ResNet-10 model. """ model = ResNet(BasicBlock, [1, 1, 1, 1], **kwargs) return model def resnet(**kwargs): """Constructs a ResNet-18 model. """ model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs) return model def resnet34(**kwargs): """Constructs a ResNet-34 model. """ model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs) return model def conv3x3x3(in_planes, out_planes, stride=1, stride_t=1): # 3x3x3 convolution with padding return spectral_norm(nn.Conv3d( in_planes, out_planes, kernel_size=(3, 3, 3), stride=[stride_t, stride, stride], padding=[1, 1, 1], bias=False)) class BasicBlock(nn.Module): expansion = 1 def __init__(self, inplanes, planes, stride=1, stride_t=1, downsample=None): super(BasicBlock, self).__init__() self.conv1 = conv3x3x3(inplanes, planes, stride, stride_t) self.bn1 = nn.GroupNorm(num_groups=16, num_channels=planes) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3x3(planes, planes) self.bn2 = nn.GroupNorm(num_groups=16, num_channels=planes) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, spatial_size, sequence_length, config): super(ResNet, self).__init__() # spatial_size = config["spatial_size"] self.inplanes = 64 self.bce_loss = config["bce_loss"] min_spatial_size = int(spatial_size / 8) #sample_duration = dic.Network['sequence_length']-1 self.max_channels = config["max_channels"] if "max_channels" in config else 256 self.conv1 = spectral_norm(nn.Conv3d( 3, 64, kernel_size=(3, 7, 7), stride=(1, 2, 2), padding=(1, 3, 3), bias=False)) self.gn1 = nn.GroupNorm(num_groups=16, num_channels=64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool3d(kernel_size=(3, 3, 3), stride=(1, 2, 2), padding=1) self.layers = nn.ModuleList() self.patch_temp = config["patch_temp_disc"] self.spatio_temporal = config["spatio_temporal"] if"spatio_temporal" in config else False if self.patch_temp: self.layer1 = self._make_layer(block, 64, layers[0], stride=1) self.layers.append(self._make_layer(block, 128, layers[1], stride=1, stride_t=1)) self.layers.append(self._make_layer(block, 128, layers[2], stride=2, stride_t=1)) self.layers.append(self._make_layer(block, 256, layers[3], stride=2, stride_t=1)) last_size = int(math.ceil(spatial_size / 16)) last_duration = 1 self.avgpool = nn.AvgPool3d((last_duration, last_size, last_size), stride=1) self.cond = config["conditional"] if "conditional" in config else False if self.cond: self.spade_emb = SPADE(norm_nc=block.expansion * 256, label_nc=2, config=config) self.fc = nn.Linear(256 * block.expansion, config["num_classes"], bias=False) else: spatial_size /= 2 self.layer1 = self._make_layer(block, 32, layers[0], stride=1) n_channels = 64 if "conditional" in config and config["conditional"]: raise ValueError("If non-patch-gan temporal discriminator is used, conditional must not be True!") self.cond = False n = 0 while sequence_length > 1: blocks = layers[n] if n<sequence_length-1 else layers[-1] n_channels = min(2*n_channels,self.max_channels) stride = 1 if spatial_size <= min_spatial_size else 2 spatial_size = int(spatial_size / stride) stride_t = 1 if self.spatio_temporal else (2 if sequence_length > 1 else 1) self.layers.append(self._make_layer(block,n_channels,blocks,stride=stride,stride_t=stride_t)) sequence_length = int(math.ceil(sequence_length / 2)) n += 1 self.final = nn.Conv2d(n_channels,1,3,padding=1) print(f"Temporal discriminator has {len(self.layers)} layers") for m in self.modules(): if isinstance(m, nn.Conv3d): m.weight = nn.init.orthogonal_(m.weight) def _make_layer(self, block, planes, blocks, stride=1, stride_t=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion or stride_t != 1: downsample = nn.Sequential( spectral_norm(nn.Conv3d( self.inplanes, planes * block.expansion, kernel_size=[3, 3, 3], stride=[stride_t, stride, stride], padding=[1, 1, 1], bias=False)), nn.GroupNorm(num_channels=planes * block.expansion, num_groups=16)) layers = [] layers.append(block(self.inplanes, planes, stride, stride_t, downsample)) self.inplanes = planes * block.expansion for _ in range(1, blocks): layers.append(block(self.inplanes, planes)) return nn.Sequential(*layers) def forward(self, x, cond=None): out = [] x = self.conv1(x) x = self.gn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) out.append(x) for n in range(len(self.layers)): x = self.layers[n](x) out.append(x) if self.patch_temp: if self.cond: x_norm = [] for i in range(x.size(2)): x_norm.append(self.spade_emb(x[:,:,i],cond)) x_norm = torch.stack(x_norm,2) else: x_norm = x x1 = self.avgpool(x_norm) output = [] for i in range(x1.size(2)): output.append(self.fc(x1[:,:,i].reshape(x1.size(0), -1))) return torch.cat(output, dim=1), out else: output = self.final(x.squeeze(2)) return output, out def loss(self, pred, real): if self.bce_loss: # vanilla gan loss return self.bce(pred, torch.ones_like(pred) if real else torch.zeros_like(pred)) else: # hinge loss if real: l = torch.mean(torch.nn.ReLU()(1.0 - pred)) else: l = torch.mean(torch.nn.ReLU()(1.0 + pred)) return l def gp(self, pred_fake, x_fake): batch_size = x_fake.size(0) grad_dout = torch.autograd.grad( outputs=pred_fake.sum(), inputs=x_fake, create_graph=True, retain_graph=True, only_inputs=True )[0] grad_dout2 = grad_dout.pow(2) assert (grad_dout2.size() == x_fake.size()) reg = grad_dout2.view(batch_size, -1).sum(1) return reg def fmap_loss(self, fmap1, fmap2, loss="l1"): recp_loss = 0 for idx in range(len(fmap1)): if loss == "l1": recp_loss += torch.mean(torch.abs((fmap1[idx] - fmap2[idx]))) if loss == "l2": recp_loss += torch.mean((fmap1[idx] - fmap2[idx]) ** 2) return recp_loss / len(fmap1) # return output, out, mu if __name__ == '__main__': ## Test 3dconvnet with dummy input import os os.environ["CUDA_VISIBLE_DEVICES"] = '7' config = {"num_classes": 1, "patch_temp_disc": True,"spatial_size": 128, "bce_loss": False, "conditional": True} dummy = torch.rand((2, 3, 6, 128, 128)).cuda() dummy_cond = torch.rand((2, 2, 128, 128)).cuda() model = resnet(config=config,spatial_size=128, sequence_length=dummy.shape[2]).cuda() print("Number of parameters in generator", sum(p.numel() for p in model.parameters())) if config["conditional"]: out, out2 = model(dummy,dummy_cond) else: out, out2,= model(dummy) test = 1
17,349
37.988764
191
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/models/latent_flow_net.py
import torch from torch import nn from torch.nn import functional as F import numpy as np import math from models.blocks import Conv2dBlock, ResBlock, AdaINLinear, NormConv2d,ConvGRU class OscillatorModel(nn.Module): def __init__(self,spatial_size,config,n_no_motion=2, logger=None): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.reparamterize = config["reparameterize_poke"] if "reparameterize_poke" in config else False self.norm_layer = config["norm_layer"] if "norm_layer" in config else "in" self.layers = config["layers"] self.n_gru_layers = config["n_gru_layers"] if "n_gru_layers" in config else 3 self.n_no_motion = n_no_motion self.n_stages = len(self.layers) assert self.n_no_motion is not None nf_first_shape_enc = int(max(32, config["nf_deep"] / (2 ** self.n_stages))) self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=self.layers[-1], n_stages=self.n_stages, n_skip_stages=self.n_stages ,nf_first=nf_first_shape_enc, norm_layer=self.norm_layer,layers=self.layers) self.dynamics_enc = Encoder(2, nf_max=self.layers[-1], n_stages=self.n_stages, variational=self.reparamterize, norm_layer=self.norm_layer, layers=self.layers) ups = [False] * self.n_gru_layers #input_sizes = [self.layers[-1]*2] + (len(self.layers)-1) * [self.layers[-1]] self.fusion_block = ConvGRU(input_size=self.layers[-1] * 2, hidden_sizes=self.layers[-1], kernel_sizes=3, n_layers=self.n_gru_layers, upsampling=ups,) self.dec = SkipConnectionDecoder(nf_in=self.layers[-1], in_channels=self.shape_enc.depths, n_skip_stages=len(self.layers), disentanglement=False, norm_layer=self.norm_layer, layers=self.layers) if logger is not None: logger.info("Constructed OscillatorModel") logger.info(f"Layers of OscillatorModel is {self.layers}") logger.info(f"Encoder channels of oscillator model is {self.layers}") def forward(self,input_img, poke, len,n_ref,target_img=None): imgs = [] sigmas_hat_out = [] if target_img==None: target_img = input_img if self.reparamterize: delta, mu, _ = self.dynamics_enc(poke) else: delta = self.dynamics_enc(poke)[0] # only first time shape encoding # if self.poke_scale_mode and not poke_linear: sigmas = self.shape_enc(input_img) sigma_tgt = self.shape_enc(target_img)[-1] sigma_dyn = sigmas.pop() pred = [sigma_dyn] * self.n_gru_layers pred_out = pred[-1] for n in range(len): # apply fusion block: input is delta, hidden states are the sigma_n # get inputs for network delta_in = delta * (1. - float(n)/(n_ref-1)) if n <= n_ref else torch.zeros_like(delta) if self.training: sigma_diff = pred_out - sigma_tgt if n < len - self.n_no_motion else torch.zeros_like(pred_out) else: sigma_diff = pred_out - sigma_tgt delta_in = torch.cat([delta_in,sigma_diff],1) # predict object encoding at next time step pred = self.fusion_block(delta_in, pred) pred_out = pred[-1] sigmas.append(pred_out) # decode x = self.dec(sigmas, [], del_shape=False) imgs.append(x) sigmas_hat_out.append(pred_out) #remove pred sigmas.pop() imgs = torch.stack(imgs, dim=1) #sigmas_hat_out[-1].reverse() return imgs, sigmas_hat_out class SAVPArchModel(nn.Module): def __init__(self, spatial_size, config): super().__init__() self.n_stages = int(np.log2(spatial_size[0] // 16)) self.poke_every_t = config["poke_every_t"] if "poke_every_t" in config else True self.dynamics_enc = Encoder(nf_in=2, nf_max=64, n_stages=self.n_stages) self.gen = SAVPGenerator(self.poke_every_t) def forward(self,img,poke,len): # encode dynamics delta = self.dynamics_enc(poke)[0] out = self.gen(img,delta,len) return out class SAVPGenerator(nn.Module): def __init__(self, poke_every): super().__init__() self.poke_every_t = poke_every # encoder stuff self.conv_e1 = Conv2dBlock(3, 32, 3, 2, norm="in", padding=1, activation="relu") # ssize 32 self.rnn_e1 = ConvGRU(32, 32, 3, 1) self.conv_e2 = Conv2dBlock(32, 64, 3, 2, norm="in", padding=1, activation="relu") # ssize 16 self.rnn_e2 = ConvGRU(64, 64, 3, 1) # bottleneck self.conv_bn = Conv2dBlock(128, 128, 3, 2, norm="in", padding=1, activation="relu") # ssize 8 self.rnn_bn = ConvGRU(128, 64, 3, 1) # decoder stuff self.up1 = nn.Upsample((16, 16), mode="bilinear") # ssize 16 self.conv_d1 = Conv2dBlock(128, 64, 3, 1, norm="in", padding=1, activation="relu") self.rnn_d1 = ConvGRU(64, 32, 3, 1) self.up2 = nn.Upsample((32, 32), mode="bilinear") # ssize 32 self.conv_d2 = Conv2dBlock(64, 32, 3, 1, norm="in", padding=1, activation="relu") self.rnn_d2 = ConvGRU(32, 32, 3, 1) self.up3 = nn.Upsample((64, 64), mode="bilinear") self.conv_out = Conv2dBlock(32, 3, 3, 1, 1, norm="none", activation="tanh") def forward(self,img,delta,len): x = img out_imgs = [] for t in range(len): x1e = self.conv_e1(x) x1er = self.rnn_e1(x1e,[x1er] if t>0 else None)[0] x2e = self.conv_e2(x1er) x2er = self.rnn_e2(x2e,[x2er] if t>0 else None)[0] if t > 0 and not self.poke_every_t: delta = np.zeros_like(delta) xbn = torch.cat([x2er,delta],dim=1) xbn = self.conv_bn(xbn) xbnr = self.rnn_bn(xbn,[xbnr] if t>0 else None)[0] x1d = self.up1(xbnr) x1d = self.conv_d1(torch.cat([x1d,x2er],1)) x1dr = self.rnn_d1(x1d,[x1dr] if t>0 else None)[0] x2d = self.up2(x1dr) x2d = self.conv_d2(torch.cat([x2d,x1er],1)) x2dr = self.rnn_d2(x2d,[x2dr] if t > 0 else None)[0] x = self.conv_out(self.up3(x2dr)) out_imgs.append(x) out_imgs = torch.stack(out_imgs,1) return out_imgs class ForegroundBackgroundModel(nn.Module): def __init__(self, spatial_size,config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.cat_poke_img = config["poke_and_img"] self.zeroflow_baseline = config["zeroflow_baseline"] self.variational = config["variational"] if "variational" in config else False foreground_background_div = config["foreground_background_div"] assert foreground_background_div >= 1. nf_first_shape_enc = int(max(32, config["nf_deep"] / (2 ** self.n_stages))) if self.variational: self.shape_enc = VariationalSkipConnectionEncoderFGBG(nf_in=3,nf_max=config["nf_deep"],n_stages=self.n_stages, n_skip_stages=self.n_stages, nf_first=nf_first_shape_enc) else: self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, n_skip_stages=self.n_stages, nf_first=nf_first_shape_enc, fg_bg=True, div=foreground_background_div) self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages) hidden_sizes = [int(config["nf_deep"]/foreground_background_div)] + [int(d/foreground_background_div) for d in self.shape_enc.depths] ups = [False] + self.shape_enc.downs self.fusion_block = ConvGRU(input_size=config["nf_deep"], hidden_sizes=hidden_sizes, kernel_sizes=3, n_layers=self.n_stages + 1, upsampling=ups) self.dec = SkipConnectionDecoder(nf_in=config["nf_deep"], in_channels=self.shape_enc.depths, n_skip_stages=self.n_stages, disentanglement=False) def forward(self,fg_img,bg_img,poke,len): x = fg_img mus = logstds = None if len > 0: if self.zeroflow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke, x], dim=1) imgs = [] sigmas_fg = [] sigmas_bg = [] # infer dynamics input delta = self.dynamics_enc(poke)[0] # only first time shape encoding sigma_n = self.shape_enc(x)[0] sigma_bg = self.shape_enc(bg_img)[1] for n in range(len): # apply fusion block: input is delta, hidden states are the sigma_n sigma_n.reverse() sigma_n = self.fusion_block(delta, sigma_n) sigma_n.reverse() sigma_cat = [torch.cat([sfg,sbg],dim=1) for sfg,sbg in zip(sigma_n,sigma_bg)] x = self.dec(sigma_cat, None, del_shape=True) imgs.append(x) # output foreground representation sigmas_fg.append(sigma_n) # out sigmas_bg.append(sigma_bg) imgs = torch.stack(imgs, dim=1) sigmas_fg[-1].reverse() else: if self.variational: sigmas_fg, sigmas_bg1, mus, logstds = self.shape_enc(x) _, sigmas_bg2, *_ = self.shape_enc(bg_img) else: sigmas_fg, sigmas_bg1 = self.shape_enc(x) _, sigmas_bg2 = self.shape_enc(bg_img) sigmas_bg = (sigmas_bg1,sigmas_bg2) sigmas = [torch.cat([sfg,sbg],dim=1) for sfg,sbg in zip(sigmas_fg,sigmas_bg2)] imgs = self.dec(sigmas, None, del_shape=True) sigmas_fg.reverse() return imgs, sigmas_fg, sigmas_bg, mus, logstds class SkipSequenceModel(nn.Module): def __init__(self,spatial_size,config,n_no_motion=None): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.min_spatial_size = config["min_spatial_size"] if "min_spatial_size" in config else 16 self.n_stages = int(np.log2(spatial_size[0] // self.min_spatial_size)) print(f"number of stages in model is {self.n_stages}") self.disentanglement = config["disentanglement"] if "disentanglement" in config else False self.cat_poke_img = config["poke_and_img"] self.zeroflow_baseline = config["zeroflow_baseline"] self.poke_every_t = config["poke_every_t"] if "poke_every_t" in config else True use_spectral_norm = config["spectnorm_decoder"] if "spectnorm_decoder" in config else False self.reparamterize = config["reparameterize_poke"] if "reparameterize_poke" in config else False self.norm_layer = config["norm_layer"] if "norm_layer" in config else "in" self.layers = config["layers"] if "layers" in config and len(config["layers"])>0 else None self.poke_scale_mode = config["poke_scale"] if "poke_scale" in config else False # self.n_no_motion = n_no_motion # if self.poke_scale_mode: # assert self.n_no_motion is not None # self.multiscale_fusion_block = config["multiscale_dynamics"] # default is dynamics model # if self.disentanglement: # self.appearance_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=True, # resnet_down=config["resnet_down"] if "resnet_down" in config else False) # n_skip_stages = min(config["n_skip_stages"], self.n_stages) if "n_skip_stages" in config else self.n_stages nf_first_shape_enc = int(max(32, config["nf_deep"] / (2 ** self.n_stages))) self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, n_skip_stages=self.n_stages ,nf_first=nf_first_shape_enc, norm_layer=self.norm_layer,layers=self.layers) self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages, variational=self.reparamterize, norm_layer=self.norm_layer, layers=self.layers) hidden_sizes = [config["nf_deep"]]+self.shape_enc.depths ups = [False] + self.shape_enc.downs self.fusion_block = ConvGRU(input_size=config["nf_deep"], hidden_sizes=hidden_sizes, kernel_sizes=3, n_layers=self.n_stages+1 if self.layers is None else len(self.layers)+1, upsampling=ups,) self.dec = SkipConnectionDecoder(nf_in=config["nf_deep"], in_channels=self.shape_enc.depths, n_skip_stages=self.n_stages if self.layers is None else len(self.layers), disentanglement=False, spectral_norm=use_spectral_norm, norm_layer=self.norm_layer, layers=self.layers) def forward(self, app_img, shape_img, poke, len, poke_linear=False,delta_scaling = None, n_zero_frames=0, invert_poke=False, poke_jump=False): # if self.disentanglement: # alpha, *_ = self.appearance_enc(app_img) # else: # alpha = None # sigma = self.shape_enc(shape_img) x = shape_img if len > 0: if self.zeroflow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke, app_img], dim=1) imgs = [] sigmas_hat_out = [] sigmas_out = [] # infer dynamics input if self.reparamterize: delta, mu, _ = self.dynamics_enc(poke) else: delta = self.dynamics_enc(poke)[0] sigmas_out = delta # only first time shape encoding #if self.poke_scale_mode and not poke_linear: sigma_n = self.shape_enc(x) for n in range(len): # apply fusion block: input is delta, hidden states are the sigma_n sigma_n.reverse() if self.poke_scale_mode: if poke_linear: if invert_poke: delta_in = delta * (1 - float(n) / int(len/2)) if n < int(len/2) else delta * (float(n - int(len/2)) / int(math.ceil(float(len)/2)) - 1) else: delta_in = (1 - float(n) / (len-n_zero_frames)) * delta if n <= len - n_zero_frames else torch.zeros_like(delta) else: delta_in = delta_scaling[n] * delta_in else: if poke_jump: delta_in = delta if n < len -n_zero_frames else torch.zeros_like(delta) else: delta_in = delta if self.poke_every_t else (delta if n == 0 else torch.zeros_like(delta)) sigma_n = self.fusion_block(delta_in, sigma_n) sigma_n.reverse() x = self.dec(sigma_n, [], del_shape=False) imgs.append(x) sigmas_hat_out.append(sigma_n) imgs = torch.stack(imgs, dim=1) sigmas_hat_out[-1].reverse() else: sigmas = self.shape_enc(x) sigmas_out = sigmas sigmas_hat_out = None imgs = self.dec(sigmas, [], del_shape=False) sigmas_out.reverse() return imgs, sigmas_out, sigmas_hat_out, [] class SingleScaleBaseline(nn.Module): def __init__(self, spatial_size, config, n_no_motion=None): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.disentanglement = config["disentanglement"] if "disentanglement" in config else False self.cat_poke_img = config["poke_and_img"] self.zeroflow_baseline = config["zeroflow_baseline"] self.poke_scale_mode = config["poke_scale"] if "poke_scale" in config else False self.poke_every_t = config["poke_every_t"] if "poke_every_t" in config else True self.n_no_motion = n_no_motion if self.poke_scale_mode: assert self.n_no_motion is not None print("Initialize SingleScaleBaseline") # self.multiscale_fusion_block = config["multiscale_dynamics"] # default is dynamics model if self.disentanglement: self.appearance_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=True, resnet_down=config["resnet_down"] if "resnet_down" in config else False) self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, n_skip_stages=0) self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages) self.n_gru_layers = 3 self.fusion_block = ConvGRU(input_size=config["nf_deep"], hidden_sizes=config["nf_deep"], kernel_sizes=3, n_layers=config["n_gru_layers"]) self.dec = SkipConnectionDecoder(nf_in=config["nf_deep"], in_channels=self.shape_enc.depths, n_skip_stages=0, disentanglement=self.disentanglement) def forward(self, app_img, shape_img, poke, len, poke_linear=False,delta_scaling = None, n_zero_frames=0, invert_poke=False,poke_jump=False): if self.disentanglement: alpha, *_ = self.appearance_enc(app_img) else: alpha = None # sigma = self.shape_enc(shape_img) if self.zeroflow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke, app_img], dim=1) x = shape_img if len > 0: imgs = [] sigmas_hat_out = [] sigmas_out = [] # infer dynamics input delta = self.dynamics_enc(poke)[0] sigma_n = self.shape_enc(x)[0] sigma_n = torch.stack([sigma_n] * self.n_gru_layers) for n in range(len): # delta scaling delta_in = delta if self.poke_every_t else (delta if n == 0 else torch.zeros_like(delta)) if self.poke_scale_mode: if poke_linear: if invert_poke: delta_in = delta * (1 - float(n) / int(len / 2)) if n < int(len / 2) else delta * (float(n - int(len / 2)) / int(math.ceil(float(len) / 2)) - 1) else: delta_in = (1 - float(n) / (len - n_zero_frames)) * delta if n <= len - n_zero_frames else torch.zeros_like(delta) else: delta_in = delta_scaling[n] * delta_in # apply fusion block sigma_n = self.fusion_block(delta_in, sigma_n) # residual connection sigma_n1 = sigma_n[-1] x = self.dec([sigma_n1], alpha, del_shape=False) imgs.append(x) sigmas_hat_out.append(sigma_n1) #sigmas_hat_out = torch.stack(sigmas_hat_out) imgs = torch.stack(imgs, dim=1) else: sigmas = self.shape_enc(x) sigmas_out = sigmas[-1] sigmas_hat_out = None imgs = self.dec(sigmas, alpha, del_shape=False) return imgs, sigmas_out, sigmas_hat_out, alpha class ResidualSequenceBaseline(nn.Module): def __init__(self,spatial_size,config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.disentanglement = config["disentanglement"] if "disentanglement" in config else False self.cat_poke_img = config["poke_and_img"] self.zeroflow_baseline = config["zeroflow_baseline"] #self.multiscale_fusion_block = config["multiscale_dynamics"] # default is dynamics model if self.disentanglement: self.appearance_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=True, resnet_down=config["resnet_down"] if "resnet_down" in config else False) self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, n_skip_stages=0) self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages) self.n_gru_layers = 3 self.fusion_block = ConvGRU(input_size=config["nf_deep"], hidden_sizes=config["nf_deep"],kernel_sizes=3, n_layers=3) self.dec = SkipConnectionDecoder(nf_in=config["nf_deep"], in_channels=self.shape_enc.depths, n_skip_stages=0, disentanglement=self.disentanglement) def forward(self,app_img,shape_img,poke,len): if self.disentanglement: alpha, *_ = self.appearance_enc(app_img) else: alpha = None #sigma = self.shape_enc(shape_img) if self.zeroflow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke,app_img],dim=1) x = shape_img if len>0: imgs = [] sigmas_hat_out = [] sigmas_out = [] # infer dynamics input delta = self.dynamics_enc(poke)[0] delta = torch.stack([delta]*self.n_gru_layers) for n in range(len): # shape encoding sigma_n = self.shape_enc(x)[0] # apply fusion block delta = self.fusion_block(sigma_n,delta) # residual connection sigma_n1 = sigma_n + delta[-1] x = self.dec([sigma_n1],alpha) imgs.append(x) sigmas_hat_out.append(sigma_n1) sigmas_hat_out = torch.stack(sigmas_hat_out,) imgs = torch.stack(imgs,dim=1) else: sigmas = self.shape_enc(x) sigmas_out = sigmas[-1] sigmas_hat_out = None imgs = self.dec(sigmas,alpha,del_shape=False) return imgs, sigmas_out, sigmas_hat_out, alpha class DynamicSkipModel(nn.Module): def __init__(self, spatial_size,config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.disentanglement = config["disentanglement"] if "disentanglement" in config else False self.cat_poke_img = config["poke_and_img"] self.zeroflow_baseline = config["zeroflow_baseline"] self.multiscale_fusion_block = config["multiscale_dynamics"] # default is dynamics model if self.disentanglement: self.appearance_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=True, resnet_down=config["resnet_down"] if "resnet_down" in config else False) n_skip_stages = min(config["n_skip_stages"],self.n_stages) if "n_skip_stages" in config else self.n_stages self.shape_enc = SkipConnectionEncoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages,n_skip_stages=n_skip_stages) if config["multiscale_dynamics"]: self.dynamics_enc = SkipConnectionEncoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages, n_skip_stages=n_skip_stages) self.fusion_block = FusionBlockMultiscale(nf_in=config["nf_deep"],nfs=self.shape_enc.depths,n_blocks=config["n_blocks"]) else: self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages) self.fusion_block = LearnedFusionBlock(nf=config["nf_deep"], n_blocks=config["n_blocks"]) self.dec = SkipConnectionDecoder(nf_in=config["nf_deep"],in_channels=self.shape_enc.depths, n_skip_stages=n_skip_stages,disentanglement=self.disentanglement) def forward(self,app_img,shape_img,poke, apply_dynamics = False): if self.disentanglement: alpha, *_ = self.appearance_enc(app_img) else: alpha = None sigma = self.shape_enc(shape_img) if self.zeroflow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke,app_img],dim=1) delta = self.dynamics_enc(poke) if self.multiscale_fusion_block else self.dynamics_enc(poke)[0] if apply_dynamics: sigma_hat = self.fusion_block(sigma if self.multiscale_fusion_block else sigma.pop(), delta) if self.multiscale_fusion_block: sigma_in_dec = sigma_hat else: sigma_in_dec = sigma + [sigma_hat] sigma_out = sigma_hat else: sigma_out = sigma if self.multiscale_fusion_block else sigma[-1] img = self.dec(sigma_in_dec if apply_dynamics else sigma,alpha,del_shape=False) return img, sigma_out, alpha class DisentangledModelWithoutDynamics(nn.Module): def __init__(self, spatial_size,config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.adain = config["adain"] # default is dynamics model self.latent_fusion = config["latent_fusion"] if "latent_fusion" in config else None self.appearance_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=self.adain, resnet_down=config["resnet_down"] if "resnet_down" in config else False) self.shape_enc = Encoder(nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages, variational=config["ib_shape"]) if self.adain: self.dec = AdaINDecoderDisentangled(nf_in=config["nf_deep"], n_stages=self.n_stages, latent_fusion=self.latent_fusion, nf_in_bn=self.appearance_enc.nf_in_bn) else: self.dec = DecoderEntangled(nf_in=2 * config["nf_deep"],n_stages=self.n_stages) def forward(self,app_img,shape_img): # appearance representation alpha, alpha_spatial, *_ = self.appearance_enc(app_img) # shape representation sigma, shape_mean, shape_logstd = self.shape_enc(shape_img) # decode img = self.dec(alpha, sigma, alpha_spatial) return img, alpha, sigma, shape_mean,shape_logstd class BasicDisentangledModel(nn.Module): def __init__(self, spatial_size,config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.zero_flow_baseline = config["zero_flow_baseline"] self.adain = config["adain"] self.cat_poke_img = config["poke_and_img"] # default is dynamics model self.latent_fusion = config["latent_fusion"] if "latent_fusion" in config else None self.appearance_enc = Encoder(nf_in=3,nf_max = config["nf_deep"], n_stages=self.n_stages, prepare_adain=self.adain, resnet_down=config["resnet_down"] if "resnet_down" in config else False) self.shape_enc = Encoder(nf_in=3, nf_max = config["nf_deep"],n_stages=self.n_stages, variational=config["ib_shape"]) self.dynamics_enc = Encoder(nf_in=5 if self.cat_poke_img else 2, nf_max=config["nf_deep"], n_stages=self.n_stages) self.fusion_block = LearnedFusionBlock(nf=config["nf_deep"],n_blocks=config["n_blocks"]) if self.adain: self.dec = AdaINDecoderDisentangled(nf_in=config["nf_deep"],n_stages=self.n_stages,latent_fusion=self.latent_fusion,nf_in_bn=self.appearance_enc.nf_in_bn) else: self.dec = DecoderEntangled(nf_in=2 * config["nf_deep"],n_stages=self.n_stages) def forward(self, app_img, shape_img, poke, apply_dynamics = False): # appearance representation alpha, alpha_spatial, *_ = self.appearance_enc(app_img) # shape representation sigma, shape_mean, shape_logstd = self.shape_enc(shape_img) # dynamics representation if self.zero_flow_baseline: poke = torch.zeros_like(poke) if self.cat_poke_img: poke = torch.cat([poke,app_img],dim=1) delta, *_ = self.dynamics_enc(poke) if self.zero_flow_baseline: delta = torch.zeros_like(delta) # apply dynamics to shape represenation sigma_hat = self.fusion_block(sigma,delta) # decode if apply_dynamics: img = self.dec(alpha,sigma_hat,alpha_spatial) else: img = self.dec(alpha,sigma,alpha_spatial) return img, sigma, sigma_hat, alpha, delta, shape_mean, shape_logstd class LearnedFusionBlock(nn.Module): def __init__(self,nf,n_blocks): super().__init__() assert n_blocks >= 1 blocks = [ResBlock(2*nf,nf)] for i in range(1,n_blocks): blocks.append(ResBlock(nf,nf)) self.model = nn.Sequential(*blocks) def forward(self,sigma,delta): x = torch.cat([sigma,delta],dim=1) x = self.model(x) return x class BasicModel(nn.Module): def __init__(self, spatial_size, config): super().__init__() # number of downsampling layers; always such that spatial bottleneck size is 16x16 self.n_stages = int(np.log2(spatial_size[0] // 16)) self.zero_flow_baseline = config["zero_flow_baseline"] self.adain = config["adain"] self.obj_enc = Encoder( nf_in=3, nf_max=config["nf_deep"], n_stages=self.n_stages,variational=config["variational"]) self.flow_enc = Encoder( nf_in=2, nf_max=config["nf_deep"], n_stages=self.n_stages, prepare_adain=self.adain ) if self.adain: self.dec = AdaINDecoderEntangled( nf_in=config["nf_deep"], n_stages=self.n_stages,latent_fusion=config["latent_fusion"] ) else: self.dec = DecoderEntangled(nf_in=2*config["nf_deep"],n_stages=self.n_stages) def forward(self, image, flow,sample_prior=False): # get object code and variational paramezers if model is variational object_code, mean, logstd = self.obj_enc(image,sample_prior) # get dynamics codes dynamics_code1, dynamics_code2, _ = self.flow_enc(flow) if self.zero_flow_baseline: # this is without flow usage, to measure the impacts of the flow as adain input dynamics_code1 = torch.zeros_like(dynamics_code1) dynamics_code2 = torch.zeros_like(dynamics_code2) # decode if self.adain: img = self.dec(object_code, dynamics_code1,dynamics_code2) else: img = self.dec(object_code,dynamics_code1) return img, object_code, dynamics_code1, mean, logstd class VariationalSkipConnectionEncoderFGBG(nn.Module): def __init__(self,nf_in,nf_max, n_stages, n_skip_stages, act = "relu", nf_first=None): super().__init__() self.blocks = nn.ModuleList() self.n_stages = n_stages self.depths = [] self.downs = [] if nf_first is None: nf = 64 else: nf = nf_first # required self.blocks.append( NormConv2d( nf_in, int(1.5 * nf), 3, 2, padding=1 ) ) self.n_skip_stages = n_skip_stages self.depths.append(nf) for n in range(self.n_stages - 1): self.blocks.append( NormConv2d( nf, min(nf * 3, int(1.5*nf_max)), 3, 2, padding=1, ) ) nf = min(nf * 2, nf_max) self.depths.insert(0, nf) self.downs.insert(0, True) self.bottleneck = ResBlock(nf, int(1.5 * nf_max), activation=act, stride=1) self.downs.insert(0, False) self.squash = nn.Sigmoid() def _reparameterize(self,codes): mu = codes[:,:int(codes.shape[1]/2)] logstd = codes[:,int(codes.shape[1]/2):] logstd = self.squash(logstd) std = torch.exp(logstd) eps = torch.randn_like(std) return eps.mul(std) + mu, mu, logstd def forward(self,x): out_fg = [] out_bg = [] out_logstd = [] out_mu = [] for i in range(self.n_stages): x = self.blocks[i](x) if i >= self.n_stages - self.n_skip_stages: act_div = int(x.shape[1] * 2. / 3.) sample, mu, logstd = self._reparameterize(x[:,:act_div]) out_fg.append(sample) bg = x[:,act_div:] out_bg.append(bg) out_mu.append(mu) out_logstd.append(logstd) x = torch.cat([mu,bg], dim=1) x = self.bottleneck(x) act_div = int(x.shape[1] * 2. / 3.) sample, mu, logstd = self._reparameterize(x[:, :act_div]) out_fg.append(sample) bg = x[:, act_div:] out_bg.append(bg) out_mu.append(mu) out_logstd.append(logstd) return out_fg, out_bg, out_mu, out_logstd class SkipConnectionEncoder(nn.Module): def __init__(self,nf_in,nf_max, n_stages, n_skip_stages, act = "elu", nf_first=None, fg_bg = False, div= None, norm_layer="in", layers=None): super().__init__() self.blocks = nn.ModuleList() self.n_stages = n_stages if layers is None else len(layers) self.depths = [] self.downs = [] if nf_first is None: nf = 32 else: nf = nf_first if layers is not None: nf = layers[0] self.fg_bg = fg_bg if self.fg_bg: assert div is not None self.div = div self.blocks.append( Conv2dBlock( nf_in, nf, 3, 2, norm=norm_layer, activation=act, padding=1 ) ) self.n_skip_stages = n_skip_stages if layers is None else len(layers) self.depths.append(nf) for n in range(self.n_stages - 1): self.blocks.append( Conv2dBlock( nf, min(nf * 2, nf_max) if layers is None else layers[n+1], 3, 2, norm=norm_layer, activation=act, padding=1, ) ) nf = min(nf * 2, nf_max) if layers is None else layers[n+1] self.depths.insert(0,nf) self.downs.insert(0,True) self.bottleneck = ResBlock(nf, nf_max, activation=act, stride=1,norm=norm_layer) self.downs.insert(0,False) def forward(self,x): if self.fg_bg: out_fg = [] out_bg = [] else: out = [] for i in range(self.n_stages): x = self.blocks[i](x) if i >= self.n_stages - self.n_skip_stages: if self.fg_bg: act_div = int(x.shape[1] / self.div) out_fg.append(x[:,:act_div]) out_bg.append(x[:,act_div:]) else: out.append(x) x = self.bottleneck(x) if self.fg_bg: act_div = int(x.shape[1] / self.div) out_fg.append(x[:,:act_div]) out_bg.append(x[:,act_div:]) return out_fg, out_bg else: out.append(x) return out class Encoder(nn.Module): def __init__(self, nf_in, nf_max, n_stages, prepare_adain=False, variational=False, resnet_down=False, norm_layer = "in", layers=None): super().__init__() self.prepare_adain = prepare_adain self.variational = variational if self.prepare_adain: assert not self.variational, "Encoder should not be variational if adain is prepared" if self.prepare_adain: self.final_linear = nn.Linear(nf_max, nf_max) act = "elu" #if self.variational else "relu" blocks = [] bottleneck = [] nf = 32 if layers is None else layers[0] blocks.append( Conv2dBlock( nf_in, nf, 3, 2, norm=norm_layer, activation=act, padding=1 ) ) n_stages = n_stages if layers is None else len(layers) for n in range(n_stages - 1): blocks.append( Conv2dBlock( nf, min(nf * 2, nf_max) if layers is None else layers[n+1], 3, 2, norm=norm_layer, activation=act, padding=1, ) ) nf = min(nf * 2, nf_max) if layers is None else layers[n+1] self.resnet_down = resnet_down and self.prepare_adain self.nf_in_bn = nf bottleneck.append(ResBlock(nf, nf_max,activation=act, stride=2 if self.resnet_down else 1, norm=norm_layer)) if layers is None: bottleneck.append(ResBlock(nf_max, nf_max,activation=act, stride=2 if self.resnet_down else 1, norm=norm_layer)) if self.resnet_down: self.make_vector = Conv2dBlock(nf_max,nf_max,4,1,0) if self.variational: self.make_mu = NormConv2d(nf_max,nf_max,3, padding=1) self.make_sigma = NormConv2d(nf_max,nf_max,3, padding=1) self.squash = nn.Sigmoid() self.model = nn.Sequential(*blocks) self.bottleneck = nn.Sequential(*bottleneck) def forward(self, input, sample_prior=False): out = self.model(input) mean = out out = self.bottleneck(out) logstd = None if self.prepare_adain: # mean is a false name here, this is the raw channels of the conv model # mean = out if self.resnet_down: # in this case, mean has spatial_size 4x4 out = self.make_vector(out).squeeze(-1).squeeze(-1) else: out = F.avg_pool2d(out, out.size(2), padding=0) out = out.squeeze(-1).squeeze(-1) # no activation for the first trial, as relu would not allow for values < 0 out = self.final_linear(out) elif self.variational: mean = self.make_mu(out) # normalize sigma in between logstd = self.squash(self.make_sigma(out)) if sample_prior: out = torch.randn_like(mean) else: out = self.reparametrize(mean,logstd) return out, mean, logstd def reparametrize(self,mean,logstd): std = torch.exp(logstd) eps = torch.randn_like(std) return eps.mul(std) + mean class AdaINDecoderEntangled(nn.Module): """ We sample up from spatial resolution 16x16, given quadratic images """ def __init__(self, nf_in, n_stages,latent_fusion): super().__init__() self.blocks = nn.ModuleList() self.affines = nn.ModuleList() self.n_stages = n_stages self.latent_fusion = latent_fusion # results latent fusion results in a deeper model nf = nf_in * 2 if self.latent_fusion else nf_in self.in_block = ResBlock(nf, nf) for n in range(self.n_stages): self.affines.append(AdaINLinear(nf_in, int(nf // 2))) # upsampling adain layers self.blocks.append( ResBlock(nf, int(nf // 2), norm="adain", upsampling=True) ) nf = int(nf // 2) self.out_conv = Conv2dBlock( nf, 3, 3, 1, padding=1, norm="none", activation="tanh" ) def forward(self, object_code, dynamics_linear,dynamics_spatial): if self.latent_fusion: in_code = torch.cat([object_code,dynamics_spatial],dim=1) else: in_code = object_code x = self.in_block(in_code) for n in range(self.n_stages): adain_params = self.affines[n](dynamics_linear) x = self.blocks[n](x, adain_params) x = self.out_conv(x) return x class AdaINDecoderDisentangled(nn.Module): def __init__(self,nf_in, n_stages, latent_fusion = None, nf_in_bn = 0): super().__init__() self.blocks = nn.ModuleList() self.affines = nn.ModuleList() self.n_stages = n_stages self.latent_fusion = False if latent_fusion is None else latent_fusion if self.latent_fusion: assert nf_in_bn > 0 # results latent fusion results in a deeper model nf = nf_in + nf_in_bn if self.latent_fusion else nf_in # self.bottleneck_adain = bottleneck_adain self.in_block = ResBlock(nf, nf,) for n in range(self.n_stages): self.affines.append(AdaINLinear(nf_in, int(nf // 2))) # upsampling adain layers self.blocks.append( ResBlock(nf, int(nf // 2), norm="adain", upsampling=True) ) nf = int(nf // 2) self.out_conv = Conv2dBlock( nf, 3, 3, 1, padding=1, norm="none", activation="tanh" ) def forward(self,alpha,sigma,alpha_spatial=None): if self.latent_fusion: assert alpha_spatial is not None in_code = torch.cat([sigma, alpha_spatial],dim=1) else: in_code = sigma x = self.in_block(in_code) for n in range(self.n_stages): adain_params = self.affines[n](alpha) x = self.blocks[n](x, adain_params) x = self.out_conv(x) return x class SkipConnectionDecoder(nn.Module): def __init__(self,nf_in, in_channels, n_skip_stages, disentanglement=False, spectral_norm=False, norm_layer="in",layers=None): super().__init__() self.n_stages = len(in_channels) self.disentanglement = disentanglement self.n_skip_stages = n_skip_stages self.blocks = nn.ModuleList() if self.disentanglement: self.affines = nn.ModuleList() nf = nf_in self.in_block = ResBlock(nf,in_channels[0], snorm=spectral_norm, norm=norm_layer) for i,nf in enumerate(in_channels): if layers is None: n_out = int(nf // 2) if i < len(in_channels) - 1 else nf if self.disentanglement: self.affines.append(AdaINLinear(nf_in,n_out)) nf_in_dec = 2 * nf if i < self.n_skip_stages else nf if layers is not None: nf_in_dec = 2 * nf n_out = in_channels[i+1] if i < len(in_channels) -1 else nf self.blocks.append(ResBlock(nf_in_dec, n_out , norm="adain" if self.disentanglement else norm_layer, upsampling=True,snorm=spectral_norm)) self.out_conv = Conv2dBlock(nf,3,3,1,1,norm="none",activation="tanh") def forward(self,shape, appearance = None, del_shape=True): x = self.in_block(shape.pop() if del_shape else shape[-1]) for n in range(self.n_stages): if n < self.n_skip_stages: x = torch.cat([x,shape.pop() if del_shape else shape[self.n_skip_stages-1-n]],1) if self.disentanglement: adain_params = self.affines[n](appearance) x = self.blocks[n](x,adain_params) else: x = self.blocks[n](x) if del_shape: assert not shape out = self.out_conv(x) return out class DecoderEntangled(nn.Module): """ We sample up from spatial resolution 16x16, given quadratic images """ def __init__(self, nf_in, n_stages): super().__init__() self.blocks = nn.ModuleList() self.n_stages = n_stages nf = nf_in self.in_block = ResBlock(nf, nf) for n in range(self.n_stages): self.blocks.append( ResBlock(nf, int(nf // 2), norm="in", upsampling=True) ) nf = int(nf // 2) self.out_conv = Conv2dBlock( nf, 3, 3, 1, padding=1, norm="none", activation="tanh" ) def forward(self, object_code, dynamics_code,*args): in_code = torch.cat([object_code,dynamics_code],dim=1) x = self.in_block(in_code) for n in range(self.n_stages): x = self.blocks[n](x) x = self.out_conv(x) return x class FusionBlockMultiscale(nn.Module): def __init__(self,nf_in,nfs,n_blocks): super().__init__() self.blocks = nn.ModuleList() self.n_stages = len(nfs) + 1 nf = nf_in for n in range(self.n_stages): self.blocks.append(LearnedFusionBlock(nf,n_blocks)) if n < len(nfs): nf = nfs[n] def forward(self,sigmas,deltas): out = [] for i,n in enumerate(range(len(sigmas)-1,-1,-1)) : out.insert(0,self.blocks[i](sigmas[n],deltas[n])) return out
45,992
39.274081
181
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/models/blocks.py
import torch from torch import nn from torch.nn import functional as F from torch.nn.utils import weight_norm, spectral_norm from torch.nn import init class ResBlock(nn.Module): def __init__( self, dim_in, dim_out, norm="in", activation="elu", pad_type="zero", upsampling=False, stride = 1, snorm=False ): super(ResBlock, self).__init__() self.norm = norm self.model = nn.ModuleList() if upsampling: self.conv1 = Conv2dTransposeBlock( dim_in, dim_out, 3, 2, 1, norm=self.norm, activation=activation, snorm= snorm ) self.conv2 = Conv2dBlock( dim_out, dim_out, 3, 1, 1, norm=self.norm, activation="none", pad_type=pad_type, snorm=snorm ) else: self.conv1 = Conv2dBlock( dim_in, dim_out, 3, stride, 1, norm=self.norm, activation=activation, pad_type=pad_type, snorm=snorm ) self.conv2 = Conv2dBlock( dim_out, dim_out, 3, 1, 1, norm=self.norm, activation="none", pad_type=pad_type, snorm=snorm ) self.convolve_res = dim_in != dim_out or upsampling or stride != 1 if self.convolve_res: if not upsampling: self.res_conv = Conv2dBlock(dim_in,dim_out,3,stride,1, norm="in", activation=activation, pad_type=pad_type, snorm=snorm) else: self.res_conv = Conv2dTransposeBlock(dim_in,dim_out,3,2,1, norm="in", activation=activation, snorm=snorm) def forward(self, x,adain_params=None): residual = x if self.convolve_res: residual = self.res_conv(residual) out = self.conv1(x,adain_params) out = self.conv2(out,adain_params) out += residual return out class Conv2dBlock(nn.Module): def __init__( self, in_dim, out_dim, ks, st, padding=0, norm="none", activation="elu", pad_type="zero", use_bias=True, activation_first=False, snorm=False ): super().__init__() self.use_bias = use_bias self.activation_first = activation_first # initialize padding if pad_type == "reflect": self.pad = nn.ReflectionPad2d(padding) elif pad_type == "replicate": self.pad = nn.ReplicationPad2d(padding) elif pad_type == "zero": self.pad = nn.ZeroPad2d(padding) else: assert 0, "Unsupported padding type: {}".format(pad_type) # initialize normalization norm_dim = out_dim if norm == "bn": self.norm = nn.BatchNorm2d(norm_dim) elif norm == "in": self.norm = nn.InstanceNorm2d(norm_dim) elif norm == "group": self.norm = nn.GroupNorm(num_channels=norm_dim,num_groups=16) elif norm == "adain": self.norm = AdaptiveInstanceNorm2d(norm_dim) elif norm == "none": self.norm = None else: assert 0, "Unsupported normalization: {}".format(norm) # initialize activation if activation == "relu": self.activation = nn.ReLU(inplace=True) elif activation == "lrelu": self.activation = nn.LeakyReLU(0.2, inplace=True) elif activation == "tanh": self.activation = nn.Tanh() elif activation == "elu": self.activation = nn.ELU() elif activation == "none": self.activation = None else: assert 0, "Unsupported activation: {}".format(activation) if snorm: self.conv = spectral_norm(nn.Conv2d(in_dim, out_dim, ks, st, bias=self.use_bias)) else: self.conv = nn.Conv2d(in_dim, out_dim, ks, st, bias=self.use_bias) def forward(self, x, adain_params=None): if self.activation_first: if self.activation: x = self.activation(x) x = self.conv(self.pad(x)) if self.norm and not isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x) elif isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x, adain_params) else: x = self.conv(self.pad(x)) if self.norm and not isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x) elif isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x, adain_params) if self.activation: x = self.activation(x) return x class NormConv2d(nn.Module): """ Convolutional layer with l2 weight normalization and learned scaling parameters """ def __init__( self, in_channels, out_channels, kernel_size, stride=1, padding=0 ): super().__init__() self.beta = nn.Parameter( torch.zeros([1, out_channels, 1, 1], dtype=torch.float32) ) self.gamma = nn.Parameter( torch.ones([1, out_channels, 1, 1], dtype=torch.float32) ) self.conv = weight_norm( nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding), name="weight", ) def forward(self, x): # weight normalization # self.conv.weight = normalize(self.conv.weight., dim=[0, 2, 3]) out = self.conv(x) out = self.gamma * out + self.beta return out class Conv2dTransposeBlock(nn.Module): def __init__( self, in_dim, out_dim, ks, st, padding=0, norm="none", activation="elu", use_bias=True, activation_first=False, snorm=False ): super().__init__() self.use_bias = use_bias self.activation_first = activation_first # initialize normalization norm_dim = out_dim if norm == "bn": self.norm = nn.BatchNorm2d(norm_dim) elif norm == "in": self.norm = nn.InstanceNorm2d(norm_dim) elif norm == "group": self.norm = nn.GroupNorm(num_channels=norm_dim,num_groups=16) elif norm == "adain": self.norm = AdaptiveInstanceNorm2d(norm_dim) elif norm == "none": self.norm = None else: assert 0, "Unsupported normalization: {}".format(norm) # initialize activation if activation == "elu": self.activation = nn.ReLU(inplace=True) elif activation == "lrelu": self.activation = nn.LeakyReLU(0.2, inplace=True) elif activation == "tanh": self.activation = nn.Tanh() elif activation == "none": self.activation = None else: assert 0, "Unsupported activation: {}".format(activation) if snorm: self.conv = spectral_norm(nn.ConvTranspose2d(in_dim, out_dim, ks, st, bias=self.use_bias, padding=padding, output_padding=padding)) else: self.conv = nn.ConvTranspose2d(in_dim, out_dim, ks, st, bias=self.use_bias, padding=padding,output_padding=padding) def forward(self, x, adain_params=None): if self.activation_first: if self.activation: x = self.activation(x) x = self.conv(x) if self.norm and not isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x) elif isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x, adain_params) else: x = self.conv(x) if self.norm and not isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x) elif isinstance(self.norm,AdaptiveInstanceNorm2d): x = self.norm(x, adain_params) if self.activation: x = self.activation(x) return x class AdaptiveInstanceNorm2d(nn.Module): def __init__(self, num_features, eps=1e-5, momentum=0.1): super().__init__() self.num_features = num_features self.eps = eps self.momentum = momentum self.register_buffer("running_mean", torch.zeros(num_features)) self.register_buffer("running_var", torch.ones(num_features)) def forward(self, x, adain_params): b, c = x.size(0), x.size(1) running_mean = self.running_mean.repeat(b) running_var = self.running_var.repeat(b) x_reshaped = x.contiguous().view(1, b * c, *x.size()[2:]) out = F.batch_norm( x_reshaped, running_mean, running_var, adain_params["weight"], adain_params["bias"], True, self.momentum, self.eps, ) return out.view(b, c, *x.size()[2:]) def __repr__(self): return self.__class__.__name__ + "(" + str(self.num_features) + ")" class AdaINLinear(nn.Module): def __init__(self, in_units, target_units, use_bias=True, actfn=nn.ReLU): super().__init__() self.linear = nn.Linear(in_units, 2 * target_units, bias=use_bias) self.act_fn = actfn() def forward(self, x): out = self.act_fn(self.linear(x)) out = { "weight": out[:, : out.size(1) // 2], "bias": out[:, out.size(1) // 2 :], } return out class ConvGRUCell(nn.Module): """ Generate a convolutional GRU cell """ def __init__(self, input_size, hidden_size, kernel_size,upsample=False): super().__init__() padding = kernel_size // 2 self.input_size = input_size self.upsample = upsample self.hidden_size = hidden_size self.reset_gate = nn.Conv2d(input_size + hidden_size, hidden_size, kernel_size, padding=padding) self.update_gate = nn.Conv2d(input_size + hidden_size, hidden_size, kernel_size, padding=padding) self.out_gate = nn.Conv2d(input_size + hidden_size, hidden_size, kernel_size, padding=padding) if self.upsample: self.up_gate = nn.ConvTranspose2d(input_size,input_size,kernel_size,2,padding=padding, output_padding=padding) init.orthogonal_(self.reset_gate.weight) init.orthogonal_(self.update_gate.weight) init.orthogonal_(self.out_gate.weight) init.constant_(self.reset_gate.bias, 0.) init.constant_(self.update_gate.bias, 0.) init.constant_(self.out_gate.bias, 0.) if self.upsample: init.orthogonal_(self.up_gate.weight) init.constant_(self.up_gate.bias, 0.) def forward(self, input_, prev_state): if self.upsample: input_ = self.up_gate(input_) # get batch and spatial sizes batch_size = input_.data.size()[0] spatial_size = input_.data.size()[2:] # generate empty prev_state, if None is provided if prev_state is None: state_size = [batch_size, self.hidden_size] + list(spatial_size) if torch.cuda.is_available(): prev_state = torch.zeros(state_size).cuda() else: prev_state = torch.zeros(state_size) # data size is [batch, channel, height, width] stacked_inputs = torch.cat([input_, prev_state], dim=1) update = torch.sigmoid(self.update_gate(stacked_inputs)) reset = torch.sigmoid(self.reset_gate(stacked_inputs)) out_inputs = torch.tanh(self.out_gate(torch.cat([input_, prev_state * reset], dim=1))) new_state = prev_state * (1 - update) + out_inputs * update return new_state class ConvGRU(nn.Module): def __init__(self, input_size, hidden_sizes, kernel_sizes, n_layers, upsampling:list=None): ''' Generates a multi-layer convolutional GRU. Preserves spatial dimensions across cells, only altering depth. Parameters ---------- input_size : integer. depth dimension of input tensors. hidden_sizes : integer or list. depth dimensions of hidden state. if integer, the same hidden size is used for all cells. kernel_sizes : integer or list. sizes of Conv2d gate kernels. if integer, the same kernel size is used for all cells. n_layers : integer. number of chained `ConvGRUCell`. ''' super(ConvGRU, self).__init__() if upsampling is None: upsampling = [False]*n_layers self.input_size = input_size if type(hidden_sizes) != list: self.hidden_sizes = [hidden_sizes]*n_layers else: assert len(hidden_sizes) == n_layers, '`hidden_sizes` must have the same length as n_layers' self.hidden_sizes = hidden_sizes if type(kernel_sizes) != list: self.kernel_sizes = [kernel_sizes]*n_layers else: assert len(kernel_sizes) == n_layers, '`kernel_sizes` must have the same length as n_layers' self.kernel_sizes = kernel_sizes self.n_layers = n_layers self.cells = [] for i in range(self.n_layers): if i == 0: input_dim = self.input_size else: input_dim = self.hidden_sizes[i - 1] self.cells.append(ConvGRUCell(input_dim, self.hidden_sizes[i], self.kernel_sizes[i],upsample=upsampling[i])) self.cells = nn.Sequential(*self.cells) def forward(self, x, hidden=None): ''' Parameters ---------- x : 4D input tensor. (batch, channels, height, width). hidden : list of 4D hidden state representations. (layer, batch, channels, height, width). Returns ------- upd_hidden : 5D hidden representation. (layer, batch, channels, height, width). ''' if hidden is None: hidden = [None]*self.n_layers input_ = x upd_hidden = [] for layer_idx in range(self.n_layers): cell = self.cells[layer_idx] cell_hidden = hidden[layer_idx] # pass through layer upd_cell_hidden = cell(input_, cell_hidden) upd_hidden.append(upd_cell_hidden) # update input_ to the last updated hidden layer for next pass input_ = upd_cell_hidden # retain tensors in list to allow different hidden sizes return upd_hidden # taken from official NVLabs implementation # Creates SPADE normalization layer based on the given configuration # SPADE consists of two steps. First, it normalizes the activations using # your favorite normalization method, such as Batch Norm or Instance Norm. # Second, it applies scale and bias to the normalized output, conditioned on # the segmentation map. # The format of |config_text| is spade(norm)(ks), where # (norm) specifies the type of parameter-free normalization. # (e.g. syncbatch, batch, instance) # (ks) specifies the size of kernel in the SPADE module (e.g. 3x3) # Example |config_text| will be spadesyncbatch3x3, or spadeinstance5x5. # Also, the other arguments are # |norm_nc|: the #channels of the normalized activations, hence the output dim of SPADE # |label_nc|: the #channels of the input semantic map, hence the input dim of SPADE class SPADE(nn.Module): def __init__(self, norm_nc, label_nc, config): super().__init__() param_free_norm_type = config["base_norm_spade"] if "base_norm_spade" in config else "instance" ks = 3 if param_free_norm_type == 'instance': self.param_free_norm = nn.InstanceNorm2d(norm_nc, affine=False) elif param_free_norm_type == 'batch': self.param_free_norm = nn.BatchNorm2d(norm_nc, affine=False) else: raise ValueError('%s is not a recognized param-free norm type in SPADE' % param_free_norm_type) # The dimension of the intermediate embedding space. Yes, hardcoded. nhidden = 128 pw = ks // 2 self.mlp_shared = nn.Sequential( nn.Conv2d(label_nc, nhidden, kernel_size=ks, padding=pw), nn.ReLU() ) self.mlp_gamma = nn.Conv2d(nhidden, norm_nc, kernel_size=ks, padding=pw) self.mlp_beta = nn.Conv2d(nhidden, norm_nc, kernel_size=ks, padding=pw) def forward(self, x, segmap): # Part 1. generate parameter-free normalized activations normalized = self.param_free_norm(x) # Part 2. produce scaling and bias conditioned on semantic map segmap = F.interpolate(segmap, size=x.size()[2:], mode='nearest') actv = self.mlp_shared(segmap) gamma = self.mlp_gamma(actv) beta = self.mlp_beta(actv) # apply scale and bias out = normalized * (1 + gamma) + beta return out
17,622
33.690945
143
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/experiments/experiment.py
from abc import abstractmethod import torch import wandb import os from os import path from glob import glob import numpy as np from utils.general import get_logger WANDB_DISABLE_CODE = True class Experiment: def __init__(self, config:dict, dirs: dict, device): self.parallel = isinstance(device, list) self.config = config self.logger = get_logger(self.config["general"]["project_name"]) self.is_debug = self.config["general"]["debug"] if self.is_debug: self.logger.info("Running in debug mode") if self.parallel: self.device = torch.device( f"cuda:{device[0]}" if torch.cuda.is_available() else "cpu" ) self.all_devices = device self.logger.info("Running experiment on multiple gpus!") else: self.device = device self.all_devices = [device] self.dirs = dirs if torch.cuda.is_available(): os.environ["CUDA_VISIBLE_DEVICES"] = ",".join([str(dev.index if self.parallel else dev) for dev in self.all_devices]) if self.config["general"]["restart"]: self.logger.info(f'Resume training run with name "{self.config["general"]["project_name"]}" on device(s) {self.all_devices}') else: self.logger.info(f'Start new training run with name "{self.config["general"]["project_name"]}" on device(s) {self.all_devices}') ########## seed setting ########## torch.manual_seed(self.config["general"]["seed"]) torch.cuda.manual_seed(self.config["general"]["seed"]) np.random.seed(self.config["general"]["seed"]) # random.seed(opt.seed) torch.backends.cudnn.deterministic = True torch.manual_seed(self.config["general"]["seed"]) rng = np.random.RandomState(self.config["general"]["seed"]) if self.config["general"]["mode"] == "train": project = "visual_poking_unsupervised" wandb.init( dir=self.dirs["log"], project=project, name=self.config["general"]["project_name"], group=self.config["general"]["experiment"], ) # log paramaters self.logger.info("Training parameters:") for key in self.config: if key != "testing": self.logger.info(f"{key}: {self.config[key]}") # print to console wandb.config.update({key: self.config[key]}) # update wandb config def _load_ckpt(self, key, dir=None,name=None, single_opt = True, use_best=False, load_name = "model"): if dir is None: dir = self.dirs["ckpt"] if name is None: if len(os.listdir(dir)) > 0: ckpts = glob(path.join(dir,"*.pt")) # load latest stored checkpoint ckpts = [ckpt for ckpt in ckpts if key in ckpt.split("/")[-1]] if len(ckpts) == 0: self.logger.info(f"*************No ckpt found****************") op_ckpt = mod_ckpt = None return mod_ckpt, op_ckpt if use_best: ckpts = [x for x in glob(path.join(dir,"*.pt")) if "=" in x.split("/")[-1]] ckpts = {float(x.split("=")[-1].split(".")[0]): x for x in ckpts} ckpt = torch.load( ckpts[max(list(ckpts.keys()))], map_location="cpu" ) else: ckpts = {float(x.split("_")[-1].split(".")[0]): x for x in ckpts} ckpt = torch.load( ckpts[max(list(ckpts.keys()))], map_location="cpu" ) mod_ckpt = ckpt[load_name] if load_name in ckpt else None if single_opt: key = [key for key in ckpt if key.startswith("optimizer")] assert len(key) == 1 key = key[0] op_ckpt = ckpt[key] else: op_ckpt = {key: ckpt[key] for key in ckpt if "optimizer" in key} msg = "best model" if use_best else "model" if mod_ckpt is not None: self.logger.info(f"*************Restored {msg} with key {key} from checkpoint****************") else: self.logger.info(f"*************No ckpt for {msg} with key {key} found, not restoring...****************") if op_ckpt is not None: self.logger.info(f"*************Restored optimizer with key {key} from checkpoint****************") else: self.logger.info(f"*************No ckpt for optimizer with key {key} found, not restoring...****************") else: mod_ckpt = op_ckpt = None return mod_ckpt, op_ckpt else: # fixme add checkpoint loading for best performing models ckpt_path = path.join(dir,name) if not path.isfile(ckpt_path): self.logger.info(f"*************No ckpt for model and optimizer found under {ckpt_path}, not restoring...****************") mod_ckpt = op_ckpt = None else: if "epoch_ckpts" in ckpt_path: mod_ckpt = torch.load( ckpt_path, map_location="cpu" ) op_path = ckpt_path.replace("model@","opt@") op_ckpt = torch.load(op_path,map_location="cpu") return mod_ckpt,op_ckpt ckpt = torch.load(ckpt_path, map_location="cpu") mod_ckpt = ckpt[load_name] if load_name in ckpt else None op_ckpt = ckpt["optimizer"] if "optimizer" in ckpt else None if mod_ckpt is not None: self.logger.info(f"*************Restored model under {ckpt_path} ****************") else: self.logger.info(f"*************No ckpt for model found under {ckpt_path}, not restoring...****************") if op_ckpt is not None: self.logger.info(f"*************Restored optimizer under {ckpt_path}****************") else: self.logger.info(f"*************No ckpt for optimizer found under {ckpt_path}, not restoring...****************") return mod_ckpt,op_ckpt @abstractmethod def train(self): """ Here, the experiment shall be run :return: """ pass @abstractmethod def test(self): """ Here the prediction shall be run :param ckpt_path: The path where the checkpoint file to load can be found :return: """ pass
6,865
40.361446
140
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/experiments/fixed_length_model.py
import torch from torch.utils.data import DataLoader, RandomSampler, SequentialSampler from torch.optim import Adam from ignite.engine import Engine, Events from ignite.handlers import ModelCheckpoint from ignite.contrib.handlers import ProgressBar from ignite.metrics import Average, MetricUsage import numpy as np import wandb from functools import partial from lpips import LPIPS from tqdm import tqdm from experiments.experiment import Experiment from data import get_dataset from data.samplers import FixedLengthSampler from models.latent_flow_net import SingleScaleBaseline,SkipSequenceModel from models.discriminator import GANTrainer from utils.losses import PerceptualVGG,vgg_loss_agg,DynamicsLoss, style_loss from utils.testing import make_flow_grid, make_img_grid, make_video, make_plot from utils.metrics import metric_fid, FIDInceptionModel, metric_lpips, psnr_lightning, ssim_lightning from utils.general import linear_var, get_member, get_patches class FixedLengthModel(Experiment): def __init__(self, config, dirs, device): super().__init__(config, dirs, device) self.datakeys = ["images","poke"] if self.config["architecture"]["disentanglement"]: self.datakeys.append("img_aT") self.datakeys.append("app_img_random") # used for efficient metrics computation self.fid_feats_real_per_frame = {} self.fid_feats_fake_per_frame = {} self.psnrs = {"t": [], "tk": [], "pl" : []} self.ssims = {"t": [], "tk": [], "pl" : []} self.lpips = {"t": [], "tk": []} self.use_gan = self.config["gan"]["use"] self.use_temp_disc = self.config["gan_temp"]["use"] if self.use_temp_disc: if not self.config["gan_temp"]["patch_temp_disc"]: assert not self.config["gan_temp"]["conditional"] #self.pixel_decoder_loss = self.config["training"]["pixel_dynamics_weight"] > 0 self.lr_dec_t = 0 self.target_dev = None # metrics for each frame self.ssims_per_frame = {} self.lpips_per_frame = {} self.psnrs_per_frame = {} # self.ssims_per_frame_pl = {} # self.psnrs_per_frame_pl = {} self.lpips_avg = None self.custom_sampler = self.config["training"]["custom_sampler"] if "custom_sampler" in self.config["training"] else False self.poke_jump = self.config["training"]["poke_jump"] if "poke_jump" in self.config["training"] else False self.poke_scale_mode = self.config["architecture"]["poke_scale"] if "poke_scale" in self.config["architecture"] else False if self.poke_jump: assert not self.poke_scale_mode def __clear_metric_arrs(self): [self.psnrs[key].clear() for key in self.psnrs] [self.ssims[key].clear() for key in self.ssims] [self.lpips[key].clear() for key in self.lpips] self.lpips_per_frame = {} self.psnrs_per_frame = {} self.ssims_per_frame = {} self.fid_feats_real_per_frame = {} self.fid_feats_fake_per_frame = {} # self.ssims_per_frame_pl = {} # self.psnrs_per_frame_pl = {} def train(self): ########## checkpoints ########## if self.config["general"]["restart"] and not self.is_debug: mod_ckpt, op_ckpts = self._load_ckpt("reg_ckpt", single_opt=False) op_ckpt_dis = op_ckpts["optimizer_dis"] op_ckpt_dyn = op_ckpts["optimizer_dyn"] else: mod_ckpt = op_ckpt_dis = op_ckpt_dyn = None # get datasets for training and testing def w_init_fn(worker_id): return np.random.seed(np.random.get_state()[1][0] + worker_id) if not self.poke_scale_mode: del self.config["data"]["n_ref_frames"] dataset, transforms = get_dataset(config=self.config["data"]) train_dataset = dataset(transforms, self.datakeys, self.config["data"], train=True) test_datakeys = self.datakeys + ["app_img_random"] if self.config["testing"]["eval_app_transfer"] and "app_img_random" not in self.datakeys else self.datakeys test_datakeys.append("flow") test_dataset = dataset(transforms, test_datakeys, self.config["data"], train=False) if self.custom_sampler: train_sampler = FixedLengthSampler(train_dataset, self.config["training"]["batch_size"],shuffle=True, weighting=train_dataset.obj_weighting,drop_last=True,zero_poke=True, zero_poke_amount=self.config["training"]["zeropoke_amount"]) train_loader = DataLoader(train_dataset, batch_sampler=train_sampler,num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn,) test_sampler = FixedLengthSampler(test_dataset, batch_size=self.config["training"]["batch_size"], shuffle=True, drop_last=True, weighting=test_dataset.obj_weighting,zero_poke=True,zero_poke_amount=self.config["training"]["zeropoke_amount"]) test_loader = DataLoader( test_dataset, batch_sampler=test_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], # worker_init_fn=w_init_fn, ) eval_sampler = FixedLengthSampler(test_dataset,batch_size=self.config["testing"]["test_batch_size"],shuffle=True, drop_last=True,weighting=test_dataset.obj_weighting,zero_poke=False) eval_loader = DataLoader(test_dataset, batch_sampler=eval_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn,) self.logger.info("Using custom fixed length sampler.") else: self.logger.info("Using standard pytorch random sampler") train_sampler = RandomSampler(train_dataset) train_loader = DataLoader( train_dataset, sampler=train_sampler, batch_size=self.config["training"]["batch_size"], num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn, drop_last=True ) test_sampler = RandomSampler(test_dataset,) test_loader = DataLoader( test_dataset, batch_size=self.config["training"]["batch_size"], sampler=test_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn, drop_last=True ) # no zeropoke for evaluation as zeropoke is only to ensure no reaction when poking outside eval_sampler = SequentialSampler(test_dataset,) eval_loader = DataLoader(test_dataset, sampler=eval_sampler, batch_size=self.config["testing"]["test_batch_size"], num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn, drop_last=True) # define model self.logger.info(f"Load model...") #net_model = SkipSequenceModel if self.config["architecture"]["use_skip_model"] else ResidualSequenceBaseline net = SkipSequenceModel(spatial_size=self.config["data"]["spatial_size"],config=self.config["architecture"]) if self.config["architecture"]["use_skip_model"] else \ SingleScaleBaseline(spatial_size=self.config["data"]["spatial_size"], config=self.config["architecture"], ) self.logger.info( f"Number of trainable parameters in model is {sum(p.numel() for p in net.parameters())}" ) if self.config["general"]["restart"] and mod_ckpt is not None: self.logger.info("Load pretrained paramaters and resume training.") net.load_state_dict(mod_ckpt) if self.parallel: net = torch.nn.DataParallel(net, device_ids=self.all_devices) net.cuda(self.all_devices[0]) self.logger.info("Model on gpu!") # log weights and gradients wandb.watch(net, log="all") # define optimizers # appearance and shape disentanglement dis_params = [{"params": get_member(net,"shape_enc").parameters(), "name": "shape_encoder"}, {"params": get_member(net,"dec").parameters(), "name": "decoder"} ] optimizer_dis = Adam(dis_params, lr=self.config["training"]["lr"]) if self.config["general"]["restart"] and op_ckpt_dis is not None: self.logger.info("Load state_dict of optimizer.") optimizer_dis.load_state_dict(op_ckpt_dis) milestones = [int(self.config["training"]["n_epochs"] * t) for t in self.config["training"]["tau"]] scheduler_dis = torch.optim.lr_scheduler.MultiStepLR(optimizer_dis, milestones=milestones, gamma=self.config["training"]["lr_reduce"]) # dynamics dyn_params = [{"params": get_member(net,"dynamics_enc").parameters(), "name": "dynamics_encoder", }, {"params": get_member(net,"fusion_block").parameters(), "name": "fusion_block",},] if self.config["training"]["decoder_update_tk"]: dyn_params.append({"params": get_member(net,"dec").parameters(), "name": "decoder"}) if "singlestage" in self.config["training"] and self.config["training"]["singlestage"]: dyn_params.append({"params": get_member(net, "shape_enc").parameters(), "name": "shape_encoder"}) optimizer_dyn = Adam(dyn_params, lr = self.config["training"]["lr"]) if self.config["general"]["restart"] and op_ckpt_dyn is not None: self.logger.info("Load state_dict of optimizer.") optimizer_dyn.load_state_dict(op_ckpt_dyn) milestones = [int(self.config["training"]["n_epochs"] * t) for t in self.config["training"]["tau"]] scheduler_dyn = torch.optim.lr_scheduler.MultiStepLR(optimizer_dyn, milestones=milestones, gamma=self.config["training"]["lr_reduce"]) # initialize disc if gan mode is enabled if self.use_gan: gan_trainer = GANTrainer(self.config, self._load_ckpt, self.logger,spatial_size=self.config["data"]["spatial_size"][0] , parallel=self.parallel, devices=self.all_devices, debug=self.is_debug) if self.use_temp_disc: gan_trainer_temp = GANTrainer(self.config, self._load_ckpt,self.logger,spatial_size=self.config["data"]["spatial_size"][0], parallel=self.parallel,devices=self.all_devices, debug=self.is_debug,temporal=True, sequence_length=train_dataset.max_frames) # set start iteration and epoch in case model training is resumed start_it = 0 start_epoch = 0 n_epoch_train = self.config["training"]["n_epochs"] if self.config["general"]["restart"] and op_ckpts is not None: start_it = list(optimizer_dis.state_dict()["state"].values())[-1]["step"] start_epoch = int(np.floor(start_it / len(train_loader))) assert self.config["training"]["n_epochs"] > start_epoch n_epoch_train = self.config["training"]["n_epochs"] - start_epoch # lr_dec_rec = partial(linear_var,start_it=0, end_it=self.config["training"]["lr_dec_end_it"], start_val=self.config["training"]["lr"], end_val=self.config["training"]["lr_dec_end_val"], clip_min=0, clip_max=self.config["training"]["lr"],) self.lr_dec_t = lr_dec_rec(start_it) # losses self.logger.info("Load VGG") self.vgg = PerceptualVGG() if self.parallel: self.vgg = torch.nn.DataParallel(self.vgg,device_ids=self.all_devices) self.vgg.cuda(self.all_devices[0]) self.logger.info("VGG on gpu") # from torchsummary import summary # summary(vgg.vgg,(3,224,224)) self.logger.info("Initialize persistent losses") latent_dynamics_loss = DynamicsLoss(config=self.config["training"]) self.logger.info("Finished initializing persistent losses.") def train_step(engine,batch): net.train() # prepare data weights=None loss_dis = 0 out_dict = {} if train_dataset.flow_weights: poke = batch["poke"][0].cuda(self.all_devices[0]) weights = batch["poke"][1].cuda(self.all_devices[0]) else: poke = batch["poke"].cuda(self.all_devices[0]) x_t = batch["images"][:, 0].cuda(self.all_devices[0]) x_seq = batch["images"][:, 1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: shape_img = batch["img_aT"].cuda(self.all_devices[0]) # apply style loss app_img_tr = batch["app_img_random"].cuda(self.all_devices[0]) x_trans, *_ = net(app_img_tr,x_t,poke,len=0) loss_style = style_loss(self.vgg,app_img_tr,x_trans) loss_dis = self.config["training"]["style_loss_weight"] * loss_style out_dict.update({"style_loss": loss_style.item()}) else: shape_img = x_t x_t_hat_i, sigma_t, _ , alpha = net(x_seq[:,-1],shape_img,poke,len=0) n_ref_frames = self.config["data"]["n_ref_frames"] - 1 if self.poke_scale_mode else train_dataset.max_frames -1 # static loss to obtain fixed image state space if "singlestage" not in self.config["training"] or not self.config["training"]["singlestage"]: loss_dis = loss_dis + vgg_loss_agg(self.vgg, x_t, x_t_hat_i) #optimize parameter of appearance, shape encoders and decoder optimizer_dis.zero_grad() loss_dis.backward() optimizer_dis.step() out_dict.update({"loss_dis" : loss_dis.item()}) #optimize in alternating gradient descent as this results in equal results than training the static/dynamic model in two completely seperate stages # however, it performs significantly better than training both models jointly with a single optimizer step (see ablations or run the model with 'singlestage' set to true) # forward pass for training of dynamics part of the model # dynamics losses seq_len = x_seq.shape[1] seq_rec, mu_delta, sigmas_hat, logstd_delta = net(x_t,shape_img,poke,len=seq_len, poke_linear=self.poke_scale_mode, n_zero_frames=seq_len-n_ref_frames-1, poke_jump=self.poke_jump) sigmas_gt = [] ll_loss_dyn = [] rec_imgs = [] if weights is not None: seq_rec = get_patches(seq_rec,weights,self.config["data"],train_dataset.weight_value_flow, logger=self.logger) x_seq = get_patches(x_seq,weights,self.config["data"],train_dataset.weight_value_flow, logger=self.logger) for n in range(seq_len): x_hat_tn,s_tn,*_ = net(x_seq[:,n],x_seq[:,n],poke,len=0) sigmas_gt.append(s_tn) rec_imgs.append(x_hat_tn) w = 1. if n != n_ref_frames else self.config["training"]["target_weight"] ll_dyn_n =w * vgg_loss_agg(self.vgg,x_seq[:,n],seq_rec[:,n]) ll_loss_dyn.append(ll_dyn_n) ll_loss_dyn = torch.stack(ll_loss_dyn,dim=0).mean() rec_imgs = torch.stack(rec_imgs,1) #latent dynamics dyn_losses = [] for s_tk,s_hat_tk in zip(sigmas_gt,sigmas_hat): dyn_losses.append(latent_dynamics_loss(s_hat_tk,s_tk,[])) latent_loss_dyn = torch.stack(dyn_losses).mean() loss_dyn = self.config["training"]["vgg_dyn_weight"] * ll_loss_dyn + self.config["training"]["latent_dynamics_weight"] * latent_loss_dyn if self.use_gan and engine.state.iteration >= self.config["gan"]["start_iteration"]: if self.config["gan"]["pixel_dynamics"]: offsets = np.random.choice(np.arange(max(1,x_seq.shape[1]-train_dataset.max_frames)),size=x_seq.shape[0]) true_exmpls = torch.stack([seq[o:o+train_dataset.max_frames] for seq, o in zip(x_seq,offsets)],dim=0) fake_exmpls = torch.stack([seq[o:o+train_dataset.max_frames] for seq, o in zip(seq_rec, offsets)], dim=0) x_true = torch.cat([true_exmpls[:,1:],true_exmpls[:,:-1]],dim=2).reshape(-1,2*true_exmpls.shape[2],*true_exmpls.shape[3:]) x_fake = torch.cat([fake_exmpls[:, 1:], true_exmpls[:, :-1]], dim=2).reshape(-1, 2 * fake_exmpls.shape[2], *fake_exmpls.shape[3:]) else: true_exmpls = np.random.choice(np.arange(x_seq.shape[0]*x_seq.shape[1]),self.config["gan"]["n_examples"]) fake_exmpls = np.random.choice(np.arange(seq_rec.shape[0]*seq_rec.shape[1]), self.config["gan"]["n_examples"]) x_true = x_seq.view(-1,*x_seq.shape[2:])[true_exmpls] x_fake = seq_rec.view(-1,*seq_rec.shape[2:])[fake_exmpls] disc_dict, loss_gen, loss_fmap = gan_trainer.train_step(x_true, x_fake) loss_dyn = loss_dyn + self.config["gan"]["gen_weight"] * loss_gen + self.config["gan"]["fmap_weight"] * loss_fmap if self.use_temp_disc and engine.state.iteration >= self.config["gan_temp"]["start_iteration"]: seq_len_act = x_seq.shape[1] offset = int(np.random.choice(np.arange(max(1,seq_len_act-train_dataset.max_frames)),1)) # offset_fake = int(np.random.choice(np.arange(max(1,seq_len_act-seq_len_temp_disc)), 1)) x_fake_tmp = seq_rec[:,offset:offset+train_dataset.max_frames].permute(0,2,1,3,4) x_true_tmp = x_seq[:, offset:offset+train_dataset.max_frames].permute(0,2,1,3,4) if self.config["gan_temp"]["conditional"]: cond = get_patches(poke,weights,self.config["data"],test_dataset.weight_value_flow,self.logger) if test_dataset.flow_weights else poke else: cond = None disc_dict_temp, loss_gen_temp, loss_fmap_temp = gan_trainer_temp.train_step(x_true_tmp,x_fake_tmp,cond) loss_dyn = loss_dyn + self.config["gan_temp"]["gen_weight"] * loss_gen_temp + self.config["gan_temp"]["fmap_weight"] * loss_fmap_temp # optimize parameters of dynamics part optimizer_dyn.zero_grad() loss_dyn.backward() optimizer_dyn.step() out_dict.update({"loss_dyn":loss_dyn.item() ,"vgg_loss_dyn" : ll_loss_dyn.item(), "latent_loss_dyn": latent_loss_dyn.item(), "lr_dec_t": self.lr_dec_t}) if self.use_gan and engine.state.iteration >= self.config["gan"]["start_iteration"]: out_dict.update(disc_dict) out_dict.update({"loss_gen_patch" :loss_gen.item(), "loss_fmap_patch": loss_fmap.item()}) if self.use_temp_disc and engine.state.iteration >= self.config["gan_temp"]["start_iteration"]: out_dict.update(disc_dict_temp) out_dict.update({"loss_gen_temp" :loss_gen_temp.item(), "loss_fmap_temp": loss_fmap_temp.item()}) return out_dict self.logger.info("Initialize inception model...") self.inception_model = FIDInceptionModel() self.logger.info("Finished initialization of inception model...") # note that lpips is exactly vgg-cosine similarity as proposed in the google papers and savp self.lpips_fn = LPIPS(net="vgg") def eval_step(engine, eval_batch): net.eval() out_dict = {} with torch.no_grad(): # prepare data weights = None if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:,0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:,1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: app_img_tr = eval_batch["app_img_random"].cuda(self.all_devices[0]) x_trans, *_ = net(app_img_tr, x_t, poke,len=0) loss_style = style_loss(self.vgg, app_img_tr, x_trans) out_dict.update({"style_loss_eval": loss_style.item()}) n_ref_frames = self.config["data"]["n_ref_frames"] - 1 if self.poke_scale_mode else train_dataset.max_frames -1 # eval forward passes seq_len = x_seq_gt.shape[1] x_t_hat, sigma_t, _, alpha = net(x_t,x_t,poke,len=0) x_seq_hat, _, sigmas_hat,_ = net(x_t, x_t, poke,len=seq_len,poke_linear=self.poke_scale_mode, n_zero_frames=seq_len-n_ref_frames-1,poke_jump=self.poke_jump) if weights is not None and self.config["testing"]["metrics_on_patches"]: x_seq_hat = get_patches(x_seq_hat,weights,self.config["data"],test_dataset.weight_value_flow, logger=self.logger) x_seq_gt = get_patches(x_seq_gt, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) sigmas_gt = [] ll_loss_dyn = [] rec_imgs = [] for n in range(seq_len): x_hat_tn, s_tn, *_ = net(x_seq_gt[:, n], x_seq_gt[:, n], poke, len=0) sigmas_gt.append(s_tn) rec_imgs.append(x_hat_tn) ll_dyn_n = vgg_loss_agg(self.vgg, x_seq_gt[:, n], x_seq_hat[:, n]) ll_loss_dyn.append(ll_dyn_n) ll_loss_tk_eval = torch.stack(ll_loss_dyn,dim=0).mean() rec_imgs = torch.stack(rec_imgs,1) if weights is not None and self.config["testing"]["metrics_on_patches"]: rec_imgs = get_patches(rec_imgs, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) # apply inception model for fid calculation at all timesteps for t in range(x_seq_gt.shape[1]): real_features_t = self.inception_model(x_seq_gt[:, t]).cpu().numpy() fake_features_t = self.inception_model(x_seq_hat[:, t]).cpu().numpy() if t not in self.fid_feats_fake_per_frame: self.fid_feats_fake_per_frame.update({t: fake_features_t}) self.fid_feats_real_per_frame.update({t: real_features_t}) else: self.fid_feats_fake_per_frame[t] = np.concatenate([self.fid_feats_fake_per_frame[t], fake_features_t], axis=0) self.fid_feats_real_per_frame[t] = np.concatenate([self.fid_feats_real_per_frame[t], real_features_t], axis=0) # evaluate training losses # ll_loss_tk_eval = vgg_loss_agg(self.vgg, x_tk, x_tk_hat) ll_loss_t_i_eval = vgg_loss_agg(self.vgg, x_t, x_t_hat) dyn_losses = [] for s_tk, s_hat_tk in zip(sigmas_gt, sigmas_hat): dyn_losses.append(latent_dynamics_loss(s_hat_tk, s_tk, [])) latent_loss_dyn_eval = torch.stack(dyn_losses).mean() out_dict.update({"vgg_loss_dyn_eval": ll_loss_tk_eval.item(), "loss_dis_i_eval": ll_loss_t_i_eval.item(), "latent_loss_dyn_eval": latent_loss_dyn_eval.item()}) # compute metrics ssim_t = ssim_lightning(x_t, x_t_hat) psnr_t = psnr_lightning(x_t, x_t_hat) lpips_t = metric_lpips(x_t,x_t_hat, self.lpips_fn, reduce=False) ssim_tk, ssim_per_frame = ssim_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) psnr_tk, psnr_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) lpips_avg, lpips_per_frame = metric_lpips(x_seq_gt, x_seq_hat,self.lpips_fn,reduce=False,return_per_frame=True) # ssim_pl, ssim_pl_per_frame = ssim_lightning(x_seq_gt,x_seq_hat,return_per_frame=True) # psnr_pl, psnr_pl_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) # append to arrays self.lpips["t"].append(lpips_t) self.psnrs["t"].append(psnr_t) self.ssims["t"].append(ssim_t) self.psnrs["tk"].append(psnr_tk) self.ssims["tk"].append(ssim_tk) self.lpips["tk"].append(lpips_avg) #self.ssims["pl"].append(ssim_pl) #self.psnrs["pl"].append(psnr_pl) # append the values of the respective sequence length [self.ssims_per_frame[key].append(ssim_per_frame[key]) if key in self.ssims_per_frame else self.ssims_per_frame.update({key:[ssim_per_frame[key]]}) for key in ssim_per_frame] [self.psnrs_per_frame[key].append(psnr_per_frame[key]) if key in self.psnrs_per_frame else self.psnrs_per_frame.update({key:[psnr_per_frame[key]]}) for key in psnr_per_frame] [self.lpips_per_frame[key].append(lpips_per_frame[key]) if key in self.lpips_per_frame else self.lpips_per_frame.update({key:[lpips_per_frame[key]]}) for key in lpips_per_frame] #[self.ssims_per_frame_pl[key].append(ssim_pl_per_frame[key]) if key in self.ssims_per_frame_pl else self.ssims_per_frame_pl.update({key: [ssim_pl_per_frame[key]]}) for key in ssim_pl_per_frame] #[self.psnrs_per_frame_pl[key].append(psnr_pl_per_frame[key]) if key in self.psnrs_per_frame_pl else self.psnrs_per_frame_pl.update({key: [psnr_pl_per_frame[key]]}) for key in psnr_pl_per_frame] return out_dict # test_it steps are performed while generating test_imgs, there n_test_img is overall number divided by number of test iterations n_test_img = int(self.config["testing"]["n_test_img"] // self.config["testing"]["test_it"]) def eval_visual(engine, eval_batch): net.eval() with torch.no_grad(): # prepare data if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1] else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) flow = eval_batch["flow"] if self.config["architecture"]["disentanglement"]: shape_img = eval_batch["img_aT"].cuda(self.all_devices[0]) else: shape_img = x_t n_ref_frames = self.config["data"]["n_ref_frames"] - 1 if self.poke_scale_mode else train_dataset.max_frames -1 seq_len = x_seq_gt.shape[1] x_seq_hat, *_ = net(x_t,x_t, poke, len=seq_len,poke_linear=self.poke_scale_mode,n_zero_frames=seq_len-n_ref_frames-1, poke_jump = self.poke_jump) x_t_hat , *_ = net(x_seq_gt[:,-1],shape_img,poke,len=0) grid_dis = make_img_grid(x_seq_gt[:,-1],shape_img, x_t_hat,x_t, n_logged=n_test_img) grid_dyn = make_flow_grid(x_t, poke, x_seq_hat[:,-1], x_seq_gt[:,-1], n_logged=n_test_img, flow=flow) seq_vis_hat = torch.cat([x_t.unsqueeze(1), x_seq_hat], 1) seq_vis_gt = torch.cat([x_t.unsqueeze(1), x_seq_gt], 1) grid_anim = make_video(x_t,poke,seq_vis_hat,seq_vis_gt,n_logged=n_test_img,flow=flow, display_frame_nr=True) it = engine.state.iteration log_dict = {"Last Frame Comparison Test data": wandb.Image(grid_dyn, caption=f"Last frames test grid #{it}."), "Disentanglement Grid Test Data": wandb.Image(grid_dis, caption=f"Test grid disentanglement #{it}."), "Video Grid Test Data": wandb.Video(grid_anim,caption=f"Test Video Grid #{it}.",fps=5)} if self.config["testing"]["eval_app_transfer"]: app_img_unrelated = eval_batch["app_img_random"].cuda(self.all_devices[0]) x_transferred, *_ = net(app_img_unrelated,x_t, poke,len=0) transfer_grid = make_img_grid(app_img_unrelated,x_t,x_transferred) log_dict.update({"Appearance transfer grid Test Data": wandb.Image(transfer_grid, caption=f"Test_grid appearance transfer #{it}")}) wandb.log(log_dict) return None self.logger.info("Initialize engines...") trainer = Engine(train_step) evaluator = Engine(eval_step) test_img_generator = Engine(eval_visual) self.logger.info("Finish engine initialization...") # checkpointing self.logger.info("Add checkpointing and pbar...") n_saved = 10 self.logger.info(f"Checkpoint saving window is {n_saved}") ckpt_handler = ModelCheckpoint(self.dirs["ckpt"], "reg_ckpt", n_saved=n_saved, require_empty=False) save_dict = {"model": net, "optimizer_dis": optimizer_dis, "optimizer_dyn": optimizer_dyn} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler, save_dict) if self.use_gan: ckpt_handler_disc = ModelCheckpoint(self.dirs["ckpt"], gan_trainer.load_key, n_saved=10, require_empty=False) save_dict_disc = {"model": gan_trainer.disc, "optimizer": gan_trainer.disc_opt} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler_disc, save_dict_disc) if self.use_temp_disc: ckpt_handler_disc_temp = ModelCheckpoint(self.dirs["ckpt"], gan_trainer_temp.load_key, n_saved=10, require_empty=False) save_dict_disc_temp = {"model": gan_trainer_temp.disc, "optimizer": gan_trainer_temp.disc_opt} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler_disc_temp, save_dict_disc_temp) pbar = ProgressBar(ascii=True) pbar.attach(trainer, output_transform=lambda x: x) pbar.attach(evaluator, output_transform=lambda x: x) #reduce the learning rate of the decoder for the image reconstruction task, such that the model focusses more on t --> tk @trainer.on(Events.ITERATION_COMPLETED) def update_lr(engine): self.lr_dec_t = lr_dec_rec(engine.state.iteration) for g in optimizer_dis.param_groups: if g["name"] == "decoder": g["lr"] = self.lr_dec_t @trainer.on(Events.ITERATION_COMPLETED(every=self.config["testing"]["log_intervall"])) def log(engine): it = engine.state.iteration wandb.log({"iteration": it}) # log losses for key in engine.state.output: wandb.log({key: engine.state.output[key]}) data = engine.state.batch if test_dataset.flow_weights: poke = data["poke"][0].cuda(self.all_devices[0]) else: poke = data["poke"].cuda(self.all_devices[0]) x_t = data["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = data["images"][:, 1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: shape_img = data["img_aT"].cuda(self.all_devices[0]) else: shape_img = x_t n_ref_frames = self.config["data"]["n_ref_frames"] - 1 if self.poke_scale_mode else train_dataset.max_frames -1 net.eval() seq_len = x_seq_gt.shape[1] with torch.no_grad(): x_seq_hat, *_ = net(x_t, x_t, poke, len=seq_len, poke_linear=self.poke_scale_mode, n_zero_frames=seq_len-n_ref_frames-1, poke_jump=self.poke_jump) x_t_hat, *_ = net(x_seq_gt[:,-1], shape_img, poke,len=0) #x_t_hat_e, *_ = net(img_aT, img_sT, poke) grid_dis_i = make_img_grid(x_seq_gt[:,-1], shape_img, x_t_hat, x_t, n_logged=n_test_img) grid_dyn = make_flow_grid(x_t, poke, x_seq_hat[:,-1], x_seq_gt[:,-1], n_logged=n_test_img) seq_vis_hat = torch.cat([x_t.unsqueeze(1),x_seq_hat],1) seq_vis_gt = torch.cat([x_t.unsqueeze(1), x_seq_gt], 1) grid_anim = make_video(x_t,poke,seq_vis_hat,seq_vis_gt,n_logged=n_test_img, display_frame_nr=True) wandb.log({"Last Frame Comparison Train Data": wandb.Image(grid_dyn, caption=f"Last frames train grid after {it} train steps."), "Disentanglement Grid Invariance Train Data": wandb.Image(grid_dis_i, caption=f"Invariance Disentanglement Grid on train set after {it} train steps."), "Video Grid Train Data": wandb.Video(grid_anim, caption=f"Train Video Grid after {it} train steps",fps=5)}) #"Disentanglement Grid Equivariance Train Data": wandb.Image(grid_dis_e, caption=f"Eqiuvariance Disentanglement Grid on train set after {it} train steps.") self.logger.info("Initialize metrics...") # compute loss average over epochs # Average(output_transform=lambda x: x["loss_dis"]).attach(trainer, "loss_dis-epoch_avg") if "singlestage" not in self.config["training"] or not self.config["training"]["singlestage"]: Average(output_transform=lambda x: x["loss_dis"]).attach(trainer, "loss_dis-epoch_avg") Average(output_transform=lambda x: x["loss_dis_i_eval"]).attach(evaluator, "loss_dis_i_eval") Average(output_transform=lambda x: x["vgg_loss_dyn"]).attach(trainer, "vgg_loss_dyn-epoch_avg") Average(output_transform=lambda x: x["latent_loss_dyn"]).attach(trainer, "latent_loss_dyn-epoch_avg") if "disentanglement" in self.config["architecture"] and self.config["architecture"]["disentanglement"]: Average(output_transform=lambda x: x["style_loss"]).attach(trainer, "style_loss-epoch_avg") Average(output_transform=lambda x: x["style_loss_eval"]).attach(evaluator, "style_loss_eval") if self.use_temp_disc or self.use_gan: def gan_training_started(engine,epoch, key="gan"): return engine.state.iteration >= self.config[key]["start_iteration"] if self.use_gan: use_patchgan_metrics = MetricUsage(started=Events.EPOCH_STARTED(event_filter=gan_training_started), completed=Events.EPOCH_COMPLETED(event_filter=gan_training_started), iteration_completed=Events.ITERATION_COMPLETED(event_filter=gan_training_started)) # gan losses Average(output_transform=lambda x: x["loss_gen_patch"]).attach(trainer, "loss_gen_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["loss_fmap_patch"]).attach(trainer, "loss_fmap_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["loss_disc_patch"]).attach(trainer, "loss_disc_patch-epoch_avg",usage=use_patchgan_metrics) #if self.config["gan"]["gp_weighflow_video_generatort"] > 0: Average(output_transform=lambda x: x["loss_gp_patch"]).attach(trainer, "loss_gp_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["p_""true_patch"]).attach(trainer, "p_true_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["p_fake_patch"]).attach(trainer, "p_fake_patch-epoch_avg",usage=use_patchgan_metrics) @trainer.on(Events.EPOCH_COMPLETED(event_filter=gan_training_started)) def gan_stuff(engine): gan_trainer.disc_scheduler.step() if self.use_temp_disc: use_tmpgan_metrics = MetricUsage(started=Events.EPOCH_STARTED(event_filter=partial(gan_training_started,key="gan_temp")), completed=Events.EPOCH_COMPLETED(event_filter=partial(gan_training_started,key="gan_temp")), iteration_completed=Events.ITERATION_COMPLETED(event_filter=partial(gan_training_started,key="gan_temp"))) # gan losses Average(output_transform=lambda x: x["loss_gen_temp"]).attach(trainer, "loss_gen_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["loss_fmap_temp"]).attach(trainer, "loss_fmap_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["loss_disc_temp"]).attach(trainer, "loss_disc_temp-epoch_avg",usage=use_tmpgan_metrics) #if self.config["gan"]["gp_weight"] > 0: Average(output_transform=lambda x: x["loss_gp_temp"]).attach(trainer, "loss_gp_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["p_true_temp"]).attach(trainer, "p_true_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["p_fake_temp"]).attach(trainer, "p_fake_temp-epoch_avg",usage=use_tmpgan_metrics) @trainer.on(Events.EPOCH_COMPLETED(event_filter=gan_training_started)) def temp_disc_stuff(engine): gan_trainer_temp.disc_scheduler.step() # evaluation losses Average(output_transform=lambda x: x["vgg_loss_dyn_eval"]).attach(evaluator, "vgg_loss_dyn_eval") Average(output_transform=lambda x: x["latent_loss_dyn_eval"]).attach(evaluator, "latent_loss_dyn_eval") self.logger.info("Finish metric initialization.") @trainer.on(Events.EPOCH_COMPLETED(every=self.config["testing"]["n_epoch_metrics"])) def metrics(engine): # set incpetion model to cpu self.inception_model.eval() self.inception_model.cuda(self.all_devices[0]) self.lpips_fn.cuda(self.all_devices[0]) self.lpips_fn.eval() if self.config["gan_temp"]["use"]: gan_trainer_temp.disc.cpu() if self.config["gan"]["use"]: gan_trainer.disc.cpu() # compute metrics over an epoch self.logger.info(f"Computing metrics after epoch #{engine.state.epoch}") batch_size = eval_sampler.batch_size if self.config["training"]["custom_sampler"] else eval_loader.batch_size bs = 20 if self.is_debug else (int(8000 / batch_size) if len(test_dataset) > 8000 else len(eval_loader)) evaluator.run(eval_loader, max_epochs=1, epoch_length=bs) [wandb.log({key: evaluator.state.metrics[key]}) for key in evaluator.state.metrics] # compute metrics test = np.stack(self.ssims["t"], axis=0) ssim_t = np.mean(np.stack(self.ssims["t"], axis=0)) psnr_t = np.mean(np.stack(self.psnrs["t"], axis=0)) lpips_t = np.mean(np.concatenate(self.lpips["t"], axis=0)) ssim_tk = np.mean(np.stack(self.ssims["tk"], axis=0)) psnr_tk = np.mean(np.stack(self.psnrs["tk"], axis=0)) lpips_avg = np.mean(np.concatenate(self.lpips["tk"], axis=0)) self.lpips_avg = lpips_avg fid_per_frame = {} for key in tqdm(self.fid_feats_real_per_frame, desc="Computing FID per frame"): fid_per_frame[key] = metric_fid(self.fid_feats_real_per_frame[key], self.fid_feats_fake_per_frame[key]) #fid_tk = metric_fid(self.features_real_fid["tk"], self.features_fake_fid["tk"]) fid_avg = np.mean([fid_per_frame[key] for key in fid_per_frame]) log_dict = {"ssim-t": ssim_t, "psnr-t": psnr_t, "lpips-t": lpips_t,"ssim-tk": ssim_tk, "psnr-tk": psnr_tk, "fid-tk": fid_avg, "lpips-avg": lpips_avg} # add histograms for per-frame-metrics self.lpips_per_frame = {key: np.concatenate(self.lpips_per_frame[key], axis=0).mean() for key in self.lpips_per_frame} self.ssims_per_frame = {key: np.stack(self.ssims_per_frame[key], axis=0).mean() for key in self.ssims_per_frame} self.psnrs_per_frame = {key: np.stack(self.psnrs_per_frame[key], axis=0).mean() for key in self.psnrs_per_frame} # self.ssims_per_frame_pl = {key: np.stack(self.ssims_per_frame_pl[key], axis=0).mean() for key in self.ssims_per_frame_pl} # self.psnrs_per_frame_pl = {key: np.stack(self.psnrs_per_frame_pl[key], axis=0).mean() for key in self.psnrs_per_frame_pl} x = [k + 1 for k in self.lpips_per_frame] make_plot(x, list(self.lpips_per_frame.values()), "LPIPS of predicted frames", ylabel="Average LPIPS") make_plot(x, list(self.ssims_per_frame.values()), "SSIM of predicted frames", ylabel="Average SSIM") make_plot(x, list(self.psnrs_per_frame.values()), "PSNR of predicted frames", ylabel="Average PSNR") make_plot(x, list(fid_per_frame.values()), "FIDs of predicted frames", ylabel="FID") wandb.log(log_dict) # clear collection arrays self.__clear_metric_arrs() self.inception_model.cpu() self.lpips_fn.cpu() if self.config["gan_temp"]["use"]: gan_trainer_temp.disc.cuda(self.all_devices[0]) if self.config["gan"]["use"]: gan_trainer.disc.cuda(self.all_devices[0]) # set #toggle_gpu(True) @trainer.on(Events.ITERATION_COMPLETED(every=self.config["testing"]["test_img_intervall"])) def make_test_grid(engine): test_img_generator.run(test_loader, max_epochs=1, epoch_length=self.config["testing"]["test_it"]) @trainer.on(Events.EPOCH_COMPLETED) def log_train_avg(engine): wandb.log({"epoch": engine.state.epoch}) [wandb.log({key: engine.state.metrics[key]}) for key in engine.state.metrics] # also perform scheduler step scheduler_dis.step() scheduler_dyn.step() def score_fn(engine): assert self.lpips_avg is not None return -self.lpips_avg # define best ckpt best_ckpt_handler = ModelCheckpoint(self.dirs["ckpt"],filename_prefix="ckpt_metric" ,score_function=score_fn,score_name="lpips",n_saved=5,require_empty=False) trainer.add_event_handler(Events.EPOCH_COMPLETED(every=self.config["testing"]["n_epoch_metrics"]),best_ckpt_handler,save_dict) @trainer.on(Events.STARTED) def set_start_it(engine): self.logger.info(f'Engine starting from iteration {start_it}, epoch {start_epoch}') engine.state.iteration = start_it engine.state.epoch = start_epoch # run everything n_step_per_epoch = 10 if self.is_debug else len(train_loader) self.logger.info("Start training...") trainer.run(train_loader, max_epochs=n_epoch_train, epoch_length=n_step_per_epoch) self.logger.info("End training.") def test(self): from tqdm import tqdm import cv2 from os import makedirs,path mod_ckpt, _ = self._load_ckpt("reg_ckpt", single_opt=False) dataset, transforms = get_dataset(config=self.config["data"]) test_dataset = dataset(transforms, self.datakeys, self.config["data"], train=False) # get datasets for training and testing def w_init_fn(worker_id): return np.random.seed(np.random.get_state()[1][0] + worker_id) if self.custom_sampler: test_sampler = FixedLengthSampler(test_dataset, batch_size=self.config["testing"]["test_batch_size"], shuffle=True, drop_last=True, weighting=test_dataset.obj_weighting, zero_poke=False) test_loader = DataLoader( test_dataset, batch_sampler=test_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], # worker_init_fn=w_init_fn, ) self.logger.info("Using custom data sampler") else: test_sampler = RandomSampler(test_dataset, ) test_loader = DataLoader(test_dataset, sampler=test_sampler, batch_size=16, num_workers=self.config["data"]["num_workers"], worker_init_fn=w_init_fn, drop_last=True) self.logger.info("Using common torch sampler") # define model self.logger.info(f"Sequence length is {test_dataset.max_frames}") self.logger.info(f"Load model...") net_model = SkipSequenceModel if self.config["architecture"]["use_skip_model"] else SingleScaleBaseline net = net_model(spatial_size=self.config["data"]["spatial_size"], config=self.config["architecture"], ) weights = [5, 5, 0] self.logger.info( f"Number of trainable parameters in model is {sum(p.numel() for p in net.parameters())}" ) net.load_state_dict(mod_ckpt) net.cuda(self.all_devices[0]) self.logger.info("Model on gpu!") net.eval() if self.config["testing"]["mode"] == "metrics": fid_feats_real_per_frame = {} fid_feats_fake_per_frame = {} def metric_step(engine, eval_batch): net.eval() out_dict = {} with torch.no_grad(): # prepare data weights = None if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) n_ref_frames = self.config["data"]["n_ref_frames"] -1 if "n_ref_frames" in self.config["data"] else self.config["data"]["max_frames"] # eval forward passes seq_len = x_seq_gt.shape[1] x_t_hat, sigma_t, _, alpha = net(x_t, x_t, poke, len=0) x_seq_hat, _, sigmas_hat, _ = net(x_t, x_t, poke, len=seq_len, poke_linear=self.poke_scale_mode, n_zero_frames=seq_len-n_ref_frames-1, poke_jump=self.poke_jump) if weights is not None and self.config["testing"]["metrics_on_patches"]: x_seq_hat = get_patches(x_seq_hat, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) x_seq_gt = get_patches(x_seq_gt, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) # apply inception model for fid calculation at time t+k for t in range(x_seq_gt.shape[1]): real_features_t = self.inception_model(x_seq_gt[:, t]).cpu().numpy() fake_features_t = self.inception_model(x_seq_hat[:, t]).cpu().numpy() if t not in fid_feats_fake_per_frame: fid_feats_fake_per_frame.update({t: fake_features_t}) fid_feats_real_per_frame.update({t: real_features_t}) else: fid_feats_fake_per_frame[t] = np.concatenate([fid_feats_fake_per_frame[t], fake_features_t], axis=0) fid_feats_real_per_frame[t] = np.concatenate([fid_feats_real_per_frame[t], real_features_t], axis=0) ssim_tk, ssim_per_frame = ssim_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) psnr_tk, psnr_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) lpips_avg, lpips_per_frame = metric_lpips(x_seq_gt, x_seq_hat, self.lpips_fn, reduce=False, return_per_frame=True) # append to arrays self.psnrs["tk"].append(psnr_tk) self.ssims["tk"].append(ssim_tk) self.lpips["tk"].append(lpips_avg) # append the values of the respective sequence length [self.ssims_per_frame[key].append(ssim_per_frame[key]) if key in self.ssims_per_frame else self.ssims_per_frame.update({key: [ssim_per_frame[key]]}) for key in ssim_per_frame] [self.psnrs_per_frame[key].append(psnr_per_frame[key]) if key in self.psnrs_per_frame else self.psnrs_per_frame.update({key: [psnr_per_frame[key]]}) for key in psnr_per_frame] [self.lpips_per_frame[key].append(lpips_per_frame[key]) if key in self.lpips_per_frame else self.lpips_per_frame.update({key: [lpips_per_frame[key]]}) for key in lpips_per_frame] return out_dict evaluator = Engine(metric_step) self.logger.info("Initialize inception model...") self.inception_model = FIDInceptionModel() self.logger.info("Finished initialization of inception model...") # note that lpips is exactly vgg-cosine similarity as proposed in the google papers and savp self.lpips_fn = LPIPS(net="vgg") pbar = ProgressBar(ascii=True) pbar.attach(evaluator, output_transform=lambda x: x) # set incpetion model to cpu self.inception_model.eval() self.inception_model.cuda(self.all_devices[0]) self.lpips_fn.cuda(self.all_devices[0]) self.lpips_fn.eval() # compute metrics over an epoch self.logger.info(f"Start metrics computation.") batch_size = test_sampler.batch_size if self.custom_sampler else test_loader.batch_size el = (int(8000 / batch_size) if len(test_dataset) > 8000 else len(test_loader)) evaluator.run(test_loader, max_epochs=1, epoch_length=el) # [wandb.log({key: evaluator.state.metrics[key]}) for key in evaluator.state.metrics] # compute metrics ssim_tk = np.mean(np.stack(self.ssims["tk"], axis=0)) psnr_tk = np.mean(np.stack(self.psnrs["tk"], axis=0)) lpips_avg = np.mean(np.concatenate(self.lpips["tk"], axis=0)) assert list(fid_feats_real_per_frame.keys()) == list(fid_feats_fake_per_frame.keys()) fid_per_frame = {} for key in tqdm(fid_feats_real_per_frame, desc="Computing FID per frame"): fid_per_frame[key] = metric_fid(fid_feats_real_per_frame[key], fid_feats_fake_per_frame[key]) # fid_tk = metric_fid(self.features_real_fid["tk"], self.features_fake_fid["tk"]) fid_avg = np.mean([fid_per_frame[key] for key in fid_per_frame]) log_dict = {"ssim-avg-temp": ssim_tk, "psnr-avg_temp": psnr_tk, "fid-avg_temp": fid_avg, "lpips-avg-temp": lpips_avg} # add histograms for per-frame-metrics self.lpips_per_frame = {key: np.concatenate(self.lpips_per_frame[key], axis=0).mean() for key in self.lpips_per_frame} self.ssims_per_frame = {key: np.stack(self.ssims_per_frame[key], axis=0).mean() for key in self.ssims_per_frame} self.psnrs_per_frame = {key: np.stack(self.psnrs_per_frame[key], axis=0).mean() for key in self.psnrs_per_frame} savedir = path.join(self.dirs["generated"], "metric_summaries") makedirs(savedir, exist_ok=True) x = [k + 1 for k in self.lpips_per_frame] make_plot(x, list(self.lpips_per_frame.values()), "LPIPS of predicted frames", ylabel="Average LPIPS", savename=path.join(savedir, "lpips.svg")) make_plot(x, list(self.ssims_per_frame.values()), "SSIM of predicted frames", ylabel="Average SSIM", savename=path.join(savedir, "ssim.svg")) make_plot(x, list(self.psnrs_per_frame.values()), "PSNR of predicted frames", ylabel="Average PSNR", savename=path.join(savedir, "psnr.svg")) make_plot(x, list(fid_per_frame.values()), "FIDs of predicted frames", ylabel="FID", savename=path.join(savedir, "fid.svg")) self.logger.info("Averaged metrics: ") for key in log_dict: self.logger.info(f'{key}: {log_dict[key]}') elif self.config["testing"]["mode"] == "fvd": batch_size = test_sampler.batch_size if self.custom_sampler else test_loader.batch_size el = (int(1000 / batch_size) if len(test_dataset) > 1000 else len(test_loader)) real_samples = [] fake_samples = [] real_samples_out = [] def generate_vids(engine, eval_batch): net.eval() with torch.no_grad(): # prepare data if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) if engine.state.iteration < el: x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) n_ref_frames = self.config["data"]["n_ref_frames"] -1 if "n_ref_frames" in self.config["data"] else self.config["data"]["max_frames"] # eval forward passes seq_len = x_seq_gt.shape[1] x_seq_hat, *_ = net(x_t, x_t, poke, len=seq_len,poke_linear=self.poke_scale_mode, n_zero_frames=seq_len-n_ref_frames-1, poke_jump=self.poke_jump) real_batch = ((x_seq_gt + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy().astype(np.uint8) fake_batch = ((x_seq_hat + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy().astype(np.uint8) real_samples.append(real_batch) fake_samples.append(fake_batch) else: real_batch = ((eval_batch["images"][:, 1:] + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy().astype(np.uint8) real_samples_out.append(real_batch) generator = Engine(generate_vids) pbar = ProgressBar(ascii=True) pbar.attach(generator, output_transform=lambda x: x) self.logger.info(f"Start collecting sequences for fvd computation...") generator.run(test_loader, max_epochs=1, epoch_length=el) savedir = path.join(self.dirs["generated"], "samples_fvd") savedir_exmpls = path.join(savedir,"vid_examples") makedirs(savedir, exist_ok=True) makedirs(savedir_exmpls, exist_ok=True) real_samples = np.stack(real_samples, axis=0) fake_samples = np.stack(fake_samples, axis=0) real_samples_out = np.stack(real_samples_out, axis=0) n_ex = 0 self.logger.info(f"Generating example videos") for i,(r,f) in enumerate(zip(real_samples,fake_samples)): savename = path.join(savedir_exmpls,f"sample{i}.mp4") r = np.concatenate([v for v in r],axis=2) f = np.concatenate([v for v in f],axis=2) all = np.concatenate([r,f],axis=1) writer = cv2.VideoWriter( savename, cv2.VideoWriter_fourcc(*"MP4V"), 5, (all.shape[2], all.shape[1]), ) # writer = vio.FFmpegWriter(savename,inputdict=inputdict,outputdict=outputdict) for frame in all: frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR) writer.write(frame) writer.release() n_ex+=1 if n_ex > 20: break self.logger.info(f"Saving samples to {savedir}") np.save(path.join(savedir, "real_samples.npy"), real_samples) np.save(path.join(savedir, "fake_samples.npy"), fake_samples) self.logger.info(f'Finish generation of vid samples.')
57,839
54.776278
210
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/experiments/__init__.py
from experiments.experiment import Experiment from experiments.sequence_model import SequencePokeModel from experiments.fixed_length_model import FixedLengthModel __experiments__ = { "sequence_poke_model": SequencePokeModel, "fixed_length_model": FixedLengthModel, } def select_experiment(config,dirs, device): experiment = config["general"]["experiment"] project_name = config["general"]["project_name"] if experiment not in __experiments__: raise NotImplementedError(f"No such experiment! {experiment}") if config["general"]["restart"]: print(f"Restarting experiment \"{project_name}\" of type \"{experiment}\". Device: {device}") else: print(f"Running new experiment \"{project_name}\" of type \"{experiment}\". Device: {device}") return __experiments__[experiment](config, dirs, device)
851
37.727273
102
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/experiments/sequence_model.py
import torch from torch.utils.data import DataLoader from torch.optim import Adam from ignite.engine import Engine, Events from ignite.handlers import ModelCheckpoint from ignite.contrib.handlers import ProgressBar from ignite.metrics import Average, MetricUsage import numpy as np import wandb from functools import partial from lpips import LPIPS from tqdm import tqdm from experiments.experiment import Experiment from data import get_dataset from data.samplers import SequenceLengthSampler from models.latent_flow_net import ResidualSequenceBaseline,SkipSequenceModel from models.discriminator import GANTrainer from utils.losses import PerceptualVGG,vgg_loss_agg,DynamicsLoss, pixel_triplet_loss, style_loss, PixelDynamicsLoss, kl_loss from utils.testing import make_flow_grid, make_img_grid, make_video, make_hist, make_plot from utils.metrics import metric_fid, FIDInceptionModel, metric_lpips, psnr_lightning, ssim_lightning from utils.general import linear_var, get_member, get_patches class SequencePokeModel(Experiment): def __init__(self, config, dirs, device): super().__init__(config, dirs, device) self.datakeys = ["images","poke"] if self.config["architecture"]["disentanglement"]: self.datakeys.append("img_aT") self.datakeys.append("app_img_random") if self.config["gan_temp"]["conditional"]: self.datakeys.append("flow") # used for efficient metrics computation self.fid_feats_fake_per_frame = {} self.fid_feats_real_per_frame = {} self.psnrs = {"t": [], "tk": [], "pl" : []} self.ssims = {"t": [], "tk": [], "pl" : []} self.lpips = {"t": [], "tk": []} self.fvd_vids_real = [] self.fvd_vids_fake = [] self.use_gan = self.config["gan"]["use"] self.use_temp_disc = self.config["gan_temp"]["use"] #self.pixel_decoder_loss = self.config["training"]["pixel_dynamics_weight"] > 0 self.lr_dec_t = 0 self.target_dev = None # metrics for each frame self.ssims_per_frame = {} self.lpips_per_frame = {} self.psnrs_per_frame = {} # self.ssims_per_frame_pl = {} # self.psnrs_per_frame_pl = {} self.lpips_avg = None self.use_norm_loss = self.config["training"]["norm_loss_weight"] > 0 if "norm_loss_weight" in self.config["training"] else False if self.use_norm_loss: assert not self.config["architecture"]["dynamics_var"] def __clear_metric_arrs(self): [self.psnrs[key].clear() for key in self.psnrs] [self.ssims[key].clear() for key in self.ssims] [self.lpips[key].clear() for key in self.lpips] self.fvd_vids_real.clear() self.fvd_vids_fake.clear() self.lpips_per_frame = {} self.psnrs_per_frame = {} self.ssims_per_frame = {} self.fid_feats_fake_per_frame = {} self.fid_feats_real_per_frame = {} # self.ssims_per_frame_pl = {} # self.psnrs_per_frame_pl = {} def train(self): ########## checkpoints ########## if self.config["general"]["restart"] and not self.is_debug: mod_ckpt, op_ckpts = self._load_ckpt("reg_ckpt", single_opt=not self.config["training"]["two_stage"]) if self.config["training"]["two_stage"]: op_ckpt_dis = op_ckpts["optimizer_dis"] op_ckpt_dyn = op_ckpts["optimizer_dyn"] else: op_ckpt_dyn = op_ckpts else: mod_ckpt = op_ckpt_dis = op_ckpt_dyn = None # get datasets for training and testing def w_init_fn(worker_id): return np.random.seed(np.random.get_state()[1][0] + worker_id) dataset, transforms = get_dataset(config=self.config["data"]) train_dataset = dataset(transforms, self.datakeys, self.config["data"], train=True) test_datakeys = self.datakeys + ["app_img_random"] if self.config["testing"]["eval_app_transfer"] and "app_img_random" not in self.datakeys else self.datakeys test_datakeys.append("flow") test_dataset = dataset(transforms, test_datakeys, self.config["data"], train=False) train_sampler = SequenceLengthSampler(train_dataset, batch_size=self.config["training"]["batch_size"], shuffle=True, drop_last=True, zero_poke=self.config["data"]["include_zeropoke"]) train_loader = DataLoader( train_dataset, batch_sampler=train_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn ) test_sampler = SequenceLengthSampler(test_dataset, batch_size=self.config["training"]["batch_size"], shuffle=True, drop_last=True, zero_poke=self.config["data"]["include_zeropoke"]) test_loader = DataLoader( test_dataset, batch_sampler=test_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn ) # n_eval_frames = int(test_dataset.max_frames/2) # if int(test_dataset.min_frames + n_eval_frames) % 2 != 0: # n_eval_frames+=1 # no zeropoke for evaluation as zeropoke is only to ensure no reaction when poking outside eval_sampler = SequenceLengthSampler(test_dataset, batch_size=self.config["testing"]["test_batch_size"], shuffle=False, drop_last=True, zero_poke = False ) eval_loader = DataLoader(test_dataset, batch_sampler=eval_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn ) # define model self.logger.info(f"Load model...") net_model = SkipSequenceModel if self.config["architecture"]["use_skip_model"] else ResidualSequenceBaseline net = net_model(spatial_size=self.config["data"]["spatial_size"], config=self.config["architecture"],) self.logger.info( f"Number of trainable parameters in model is {sum(p.numel() for p in net.parameters())}" ) if self.config["general"]["restart"] and mod_ckpt is not None: self.logger.info("Load pretrained paramaters and resume training.") net.load_state_dict(mod_ckpt) if self.parallel: net = torch.nn.DataParallel(net, device_ids=self.all_devices) net.cuda(self.all_devices[0]) self.logger.info("Model on gpu!") # log weights and gradients wandb.watch(net, log="all") # define optimizers # appearance and shape disentanglement if self.config["training"]["two_stage"]: dis_params = [{"params": get_member(net,"shape_enc").parameters(), "name": "shape_encoder"}, {"params": get_member(net,"dec").parameters(), "name": "decoder"} ] if self.config["architecture"]["disentanglement"]: dis_params.append({"params": net.appearance_enc.parameters(), "name": "appearance_encoder", },) optimizer_dis = Adam(dis_params, lr=self.config["training"]["lr"]) if self.config["general"]["restart"] and op_ckpt_dis is not None: self.logger.info("Load state_dict of optimizer.") optimizer_dis.load_state_dict(op_ckpt_dis) milestones = [int(self.config["training"]["n_epochs"] * t) for t in self.config["training"]["tau"]] scheduler_dis = torch.optim.lr_scheduler.MultiStepLR(optimizer_dis, milestones=milestones, gamma=self.config["training"]["lr_reduce"]) # dynamics dyn_params = [{"params": get_member(net,"dynamics_enc").parameters(), "name": "dynamics_encoder", }, {"params": get_member(net,"fusion_block").parameters(), "name": "fusion_block",},] if self.config["training"]["decoder_update_tk"] or not self.config["training"]["two_stage"]: dyn_params.append({"params": get_member(net,"dec").parameters(), "name": "decoder"}) if not self.config["training"]["two_stage"]: dyn_params.append({"params": get_member(net,"shape_enc").parameters(), "name": "shape_encoder"}) optimizer_dyn = Adam(dyn_params, lr = self.config["training"]["lr"]) if self.config["general"]["restart"] and op_ckpt_dyn is not None: self.logger.info("Load state_dict of optimizer.") optimizer_dyn.load_state_dict(op_ckpt_dyn) milestones = [int(self.config["training"]["n_epochs"] * t) for t in self.config["training"]["tau"]] scheduler_dyn = torch.optim.lr_scheduler.MultiStepLR(optimizer_dyn, milestones=milestones, gamma=self.config["training"]["lr_reduce"]) # initialize disc if gan mode is enabled if self.use_gan: gan_trainer = GANTrainer(self.config, self._load_ckpt, self.logger,spatial_size=self.config["data"]["spatial_size"][0] , parallel=self.parallel, devices=self.all_devices, debug=self.is_debug) if self.use_temp_disc: gan_trainer_temp = GANTrainer(self.config, self._load_ckpt,self.logger,spatial_size=self.config["data"]["spatial_size"][0], parallel=self.parallel,devices=self.all_devices, debug=self.is_debug,temporal=True, sequence_length=train_dataset.min_frames) # set start iteration and epoch in case model training is resumed start_it = 0 start_epoch = 0 n_epoch_train = self.config["training"]["n_epochs"] n_epoch_overall = self.config["training"]["n_epochs"] if self.config["general"]["restart"] and op_ckpts is not None: start_it = list(optimizer_dyn.state_dict()["state"].values())[-1]["step"] start_epoch = int(np.floor(start_it / len(train_loader))) assert self.config["training"]["n_epochs"] > start_epoch n_epoch_train = self.config["training"]["n_epochs"] - start_epoch # lr_dec_rec = partial(linear_var,start_it=0, end_it=self.config["training"]["lr_dec_end_it"], start_val=self.config["training"]["lr"], end_val=self.config["training"]["lr_dec_end_val"], clip_min=0, clip_max=self.config["training"]["lr"],) self.lr_dec_t = lr_dec_rec(start_it) # losses self.logger.info("Load VGG") self.vgg = PerceptualVGG() if self.parallel: self.vgg = torch.nn.DataParallel(self.vgg,device_ids=self.all_devices) self.vgg.cuda(self.all_devices[0]) self.logger.info("VGG on gpu") # from torchsummary import summary # summary(vgg.vgg,(3,224,224)) self.logger.info("Initialize persistent losses") latent_dynamics_loss = DynamicsLoss(config=self.config["training"]) pixel_dynamics_loss = partial(pixel_triplet_loss,vgg = self.vgg, diff_pp=self.config["training"]["pixel_dyn_spatial"]) if self.config["training"]["pixel_dynamics_vgg"] else PixelDynamicsLoss() self.logger.info("Finished initializing persistent losses.") def train_step(engine,batch): net.train() # prepare data weights=None loss_dis = 0 out_dict = {} if train_dataset.flow_weights: poke = batch["poke"][0].cuda(self.all_devices[0]) weights = batch["poke"][1].cuda(self.all_devices[0]) else: poke = batch["poke"].cuda(self.all_devices[0]) x_t = batch["images"][:, 0].cuda(self.all_devices[0]) x_seq = batch["images"][:, 1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: shape_img = batch["img_aT"].cuda(self.all_devices[0]) # apply style loss app_img_tr = batch["app_img_random"].cuda(self.all_devices[0]) x_trans, *_ = net(app_img_tr,x_t,poke,len=0) loss_style = style_loss(self.vgg,app_img_tr,x_trans) loss_dis = self.config["training"]["style_loss_weight"] * loss_style out_dict.update({"style_loss": loss_style.item()}) else: shape_img = x_t x_t_hat_i, sigma_t, _, alpha = net(x_seq[:, -1], shape_img, poke, len=0) # disentanglement loss loss_dis = loss_dis + vgg_loss_agg(self.vgg, x_t, x_t_hat_i) if engine.state.epoch <= self.config["training"]["stop_seq_stat"] and self.config["training"]["two_stage"]: #optimize parameter of appearance, shape encoders and decoder optimizer_dis.zero_grad() loss_dis.backward() optimizer_dis.step() out_dict.update({"loss_dis" : loss_dis.item()}) # forward pass for training of dynamics part of the model # dynamics losses seq_len = x_seq.shape[1] seq_rec, mu_delta, sigmas_hat, logstd_delta = net(x_t,shape_img,poke,len=seq_len) sigmas_gt = [] ll_loss_dyn = [] rec_imgs = [] if weights is not None: seq_rec = get_patches(seq_rec,weights,self.config["data"],train_dataset.weight_value_flow, logger=self.logger) x_seq = get_patches(x_seq,weights,self.config["data"],train_dataset.weight_value_flow, logger=self.logger) for n in range(seq_len): x_hat_tn,s_tn,*_ = net(x_seq[:,n],x_seq[:,n],poke,len=0) sigmas_gt.append(s_tn) rec_imgs.append(x_hat_tn) ll_dyn_n = vgg_loss_agg(self.vgg,x_seq[:,n],seq_rec[:,n]) ll_loss_dyn.append(ll_dyn_n) ll_loss_dyn = torch.stack(ll_loss_dyn,dim=0).mean() rec_imgs = torch.stack(rec_imgs,1) if weights is not None: rec_imgs = get_patches(rec_imgs,weights,self.config["data"],train_dataset.weight_value_flow, logger=self.logger) #latent dynamics dyn_losses = [] for s_tk,s_hat_tk in zip(sigmas_gt,sigmas_hat): dyn_losses.append(latent_dynamics_loss(s_hat_tk,s_tk,[])) latent_loss_dyn = torch.stack(dyn_losses).mean() loss_dyn = self.config["training"]["vgg_dyn_weight"] * ll_loss_dyn + self.config["training"]["latent_dynamics_weight"] * latent_loss_dyn if self.use_norm_loss: poke_norms = [] for p in poke: magns = p.norm(dim=0) ids = magns.nonzero(as_tuple=True) if ids[0].shape[0] > 0: poke_norms.append(magns[ids].mean().unsqueeze(0)) else: poke_norms.append(torch.zeros(1).cuda(self.all_devices[0])) poke_norms = torch.cat(poke_norms, 0) norm_loss = ((poke_norms - mu_delta.reshape(poke_norms.shape[0], -1).norm(dim=-1)) ** 2).mean() loss_dyn = loss_dyn + self.config["training"]["norm_loss_weight"] * norm_loss out_dict.update({"norm_loss": norm_loss.item()}) # kl loss for if self.config["architecture"]["dynamics_var"]: kl_dyn = kl_loss(mu_delta,logstd_delta) loss_dyn = self.config["training"]["kl_weight"] * kl_dyn out_dict.update({"kl_dyn": kl_dyn.item()}) # pixel dynamics loss_dec_dyn = [] for n in range(seq_len-1): loss_dec_dyn_tn = pixel_dynamics_loss(x_seq[:,n],x_seq[:,n+1],rec_imgs[:,n],seq_rec[:,n+1]) loss_dec_dyn.append(loss_dec_dyn_tn) loss_dec_dyn = torch.stack(loss_dec_dyn,dim=0).mean() loss_dyn = loss_dyn + self.config["training"]["pixel_dynamics_weight"] * loss_dec_dyn if self.use_gan and engine.state.iteration >= self.config["gan"]["start_iteration"]: if self.config["gan"]["pixel_dynamics"]: offsets = np.random.choice(np.arange(max(1,x_seq.shape[1]-train_dataset.min_frames)),size=x_seq.shape[0]) true_exmpls = torch.stack([seq[o:o+train_dataset.min_frames] for seq, o in zip(x_seq,offsets)],dim=0) fake_exmpls = torch.stack([seq[o:o+train_dataset.min_frames] for seq, o in zip(seq_rec, offsets)], dim=0) x_true = torch.cat([true_exmpls[:,1:],true_exmpls[:,:-1]],dim=2).reshape(-1,2*true_exmpls.shape[2],*true_exmpls.shape[3:]) x_fake = torch.cat([fake_exmpls[:, 1:], true_exmpls[:, :-1]], dim=2).reshape(-1, 2 * fake_exmpls.shape[2], *fake_exmpls.shape[3:]) else: true_exmpls = np.random.choice(np.arange(x_seq.shape[0]*x_seq.shape[1]),self.config["gan"]["n_examples"]) fake_exmpls = np.random.choice(np.arange(seq_rec.shape[0]*seq_rec.shape[1]), self.config["gan"]["n_examples"]) x_true = x_seq.view(-1,*x_seq.shape[2:])[true_exmpls] x_fake = seq_rec.view(-1,*seq_rec.shape[2:])[fake_exmpls] disc_dict, loss_gen, loss_fmap = gan_trainer.train_step(x_true, x_fake) loss_dyn = loss_dyn + self.config["gan"]["gen_weight"] * loss_gen + self.config["gan"]["fmap_weight"] * loss_fmap if self.use_temp_disc and engine.state.iteration >= self.config["gan_temp"]["start_iteration"]: seq_len_act = x_seq.shape[1] offset = int(np.random.choice(np.arange(max(1,seq_len_act-train_dataset.min_frames)),1)) # offset_fake = int(np.random.choice(np.arange(max(1,seq_len_act-seq_len_temp_disc)), 1)) x_fake_tmp = seq_rec[:,offset:offset+train_dataset.min_frames].permute(0,2,1,3,4) x_true_tmp = x_seq[:, offset:offset+train_dataset.min_frames].permute(0,2,1,3,4) if self.config["gan_temp"]["conditional"]: flow = batch["flow"].cuda(self.all_devices[0]) cond = get_patches(flow,weights,self.config["data"],test_dataset.weight_value_flow,self.logger) if test_dataset.flow_weights else flow else: cond = None disc_dict_temp, loss_gen_temp, loss_fmap_temp = gan_trainer_temp.train_step(x_true_tmp,x_fake_tmp,cond) loss_dyn = loss_dyn + self.config["gan_temp"]["gen_weight"] * loss_gen_temp + self.config["gan_temp"]["fmap_weight"] * loss_fmap_temp # optimize parameters of dynamics part optimizer_dyn.zero_grad() loss_dyn.backward() optimizer_dyn.step() out_dict.update({"loss_dyn":loss_dyn.item() ,"vgg_loss_dyn" : ll_loss_dyn.item(), "latent_loss_dyn": latent_loss_dyn.item(), "lr_dec_t": self.lr_dec_t}) if self.use_gan and engine.state.iteration >= self.config["gan"]["start_iteration"]: out_dict.update(disc_dict) out_dict.update({"loss_gen_patch" :loss_gen.item(), "loss_fmap_patch": loss_fmap.item()}) if self.use_temp_disc and engine.state.iteration >= self.config["gan_temp"]["start_iteration"]: out_dict.update(disc_dict_temp) out_dict.update({"loss_gen_temp" :loss_gen_temp.item(), "loss_fmap_temp": loss_fmap_temp.item()}) #if self.pixel_decoder_loss: out_dict.update({"pixel_loss_dec": loss_dec_dyn.item()}) return out_dict self.logger.info("Initialize inception model...") self.inception_model = FIDInceptionModel() self.logger.info("Finished initialization of inception model...") # note that lpips is exactly vgg-cosine similarity as proposed in the google papers and savp self.lpips_fn = LPIPS(net="vgg") def eval_step(engine, eval_batch): net.eval() out_dict = {} with torch.no_grad(): # prepare data weights = None if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:,0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:,1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: app_img_tr = eval_batch["app_img_random"].cuda(self.all_devices[0]) x_trans, *_ = net(app_img_tr, x_t, poke,len=0) loss_style = style_loss(self.vgg, app_img_tr, x_trans) out_dict.update({"style_loss_eval": loss_style.item()}) # eval forward passes seq_len = x_seq_gt.shape[1] x_t_hat, sigma_t, _, alpha = net(x_t,x_t,poke,len=0) x_seq_hat, mu_delta, sigmas_hat,_ = net(x_t, x_t, poke,len=seq_len) if weights is not None and self.config["testing"]["metrics_on_patches"]: x_seq_hat = get_patches(x_seq_hat,weights,self.config["data"],test_dataset.weight_value_flow, logger=self.logger) x_seq_gt = get_patches(x_seq_gt, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) sigmas_gt = [] ll_loss_dyn = [] rec_imgs = [] for n in range(seq_len): x_hat_tn, s_tn, *_ = net(x_seq_gt[:, n], x_seq_gt[:, n], poke, len=0) sigmas_gt.append(s_tn) rec_imgs.append(x_hat_tn) ll_dyn_n = vgg_loss_agg(self.vgg, x_seq_gt[:, n], x_seq_hat[:, n]) ll_loss_dyn.append(ll_dyn_n) ll_loss_tk_eval = torch.stack(ll_loss_dyn,dim=0).mean() rec_imgs = torch.stack(rec_imgs,1) if self.use_norm_loss: poke_norms = [] for p in poke: magns = p.norm(dim=0) ids = magns.nonzero(as_tuple=True) if ids[0].shape[0] > 0: poke_norms.append(magns[ids].mean().unsqueeze(0)) else: poke_norms.append(torch.zeros(1).cuda(self.all_devices[0])) poke_norms = torch.cat(poke_norms, 0) norm_loss = ((poke_norms - mu_delta.reshape(poke_norms.shape[0], -1).norm(dim=-1)) ** 2).mean() out_dict.update({"norm_loss": norm_loss.item()}) if weights is not None and self.config["testing"]["metrics_on_patches"]: rec_imgs = get_patches(rec_imgs, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) # apply inception model for fid calculation at time t+k for t in range(x_seq_gt.shape[1]): real_features_t = self.inception_model(x_seq_gt[:, t]).cpu().numpy() fake_features_t = self.inception_model(x_seq_hat[:, t]).cpu().numpy() if t not in self.fid_feats_fake_per_frame: self.fid_feats_fake_per_frame.update({t: fake_features_t}) self.fid_feats_real_per_frame.update({t: real_features_t}) else: self.fid_feats_fake_per_frame[t] = np.concatenate([self.fid_feats_fake_per_frame[t], fake_features_t], axis=0) self.fid_feats_real_per_frame[t] = np.concatenate([self.fid_feats_real_per_frame[t], real_features_t], axis=0) # evaluate training losses # ll_loss_tk_eval = vgg_loss_agg(self.vgg, x_tk, x_tk_hat) ll_loss_t_i_eval = vgg_loss_agg(self.vgg, x_t, x_t_hat) dyn_losses = [] for s_tk, s_hat_tk in zip(sigmas_gt, sigmas_hat): dyn_losses.append(latent_dynamics_loss(s_hat_tk, s_tk, [])) latent_loss_dyn_eval = torch.stack(dyn_losses).mean() out_dict.update({"vgg_loss_dyn_eval": ll_loss_tk_eval.item(), "loss_dis_i_eval": ll_loss_t_i_eval.item(), "latent_loss_dyn_eval": latent_loss_dyn_eval.item()}) #if self.pixel_decoder_loss: #x_t_hat_dec = net.dec(sigma_t, alpha) #loss_dec_dyn = (vgg_loss_agg(self.vgg, x_t_hat_dec, x_tk_hat) - vgg_loss_agg(self.vgg, x_t, x_tk)) ** 2 loss_dec_dyn = [] for n in range(seq_len - 1): loss_dec_dyn_tn = pixel_dynamics_loss(x_seq_gt[:, n], x_seq_gt[:, n + 1], rec_imgs[:,n], x_seq_hat[:, n + 1]) loss_dec_dyn.append(loss_dec_dyn_tn) loss_dec_dyn = torch.stack(loss_dec_dyn, dim=0).mean() out_dict.update({"pixel_loss_dec_eval": loss_dec_dyn.item()}) # compute metrics ssim_t = ssim_lightning(x_t, x_t_hat) psnr_t = psnr_lightning(x_t, x_t_hat) lpips_t = metric_lpips(x_t,x_t_hat, self.lpips_fn, reduce=False) ssim_tk, ssim_per_frame = ssim_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) psnr_tk, psnr_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) lpips_avg, lpips_per_frame = metric_lpips(x_seq_gt, x_seq_hat,self.lpips_fn,reduce=False,return_per_frame=True) # ssim_pl, ssim_pl_per_frame = ssim_lightning(x_seq_gt,x_seq_hat,return_per_frame=True) # psnr_pl, psnr_pl_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) # append to arrays self.lpips["t"].append(lpips_t) self.psnrs["t"].append(psnr_t) self.ssims["t"].append(ssim_t) self.psnrs["tk"].append(psnr_tk) self.ssims["tk"].append(ssim_tk) self.lpips["tk"].append(lpips_avg) #self.ssims["pl"].append(ssim_pl) #self.psnrs["pl"].append(psnr_pl) # append the values of the respective sequence length [self.ssims_per_frame[key].append(ssim_per_frame[key]) if key in self.ssims_per_frame else self.ssims_per_frame.update({key:[ssim_per_frame[key]]}) for key in ssim_per_frame] [self.psnrs_per_frame[key].append(psnr_per_frame[key]) if key in self.psnrs_per_frame else self.psnrs_per_frame.update({key:[psnr_per_frame[key]]}) for key in psnr_per_frame] [self.lpips_per_frame[key].append(lpips_per_frame[key]) if key in self.lpips_per_frame else self.lpips_per_frame.update({key:[lpips_per_frame[key]]}) for key in lpips_per_frame] #[self.ssims_per_frame_pl[key].append(ssim_pl_per_frame[key]) if key in self.ssims_per_frame_pl else self.ssims_per_frame_pl.update({key: [ssim_pl_per_frame[key]]}) for key in ssim_pl_per_frame] #[self.psnrs_per_frame_pl[key].append(psnr_pl_per_frame[key]) if key in self.psnrs_per_frame_pl else self.psnrs_per_frame_pl.update({key: [psnr_pl_per_frame[key]]}) for key in psnr_pl_per_frame] return out_dict # test_it steps are performed while generating test_imgs, there n_test_img is overall number divided by number of test iterations n_test_img = int(self.config["testing"]["n_test_img"] // self.config["testing"]["test_it"]) def eval_visual(engine, eval_batch): net.eval() with torch.no_grad(): # prepare data if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1] else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) flow = eval_batch["flow"] if self.config["architecture"]["disentanglement"]: shape_img = eval_batch["img_aT"].cuda(self.all_devices[0]) else: shape_img = x_t seq_len = x_seq_gt.shape[1] x_seq_hat, *_ = net(x_t,x_t, poke, len=seq_len) x_t_hat , *_ = net(x_seq_gt[:,-1],shape_img,poke,len=0) grid_dis = make_img_grid(x_seq_gt[:,-1],shape_img, x_t_hat,x_t, n_logged=n_test_img) grid_dyn = make_flow_grid(x_t, poke, x_seq_hat[:,-1], x_seq_gt[:,-1], n_logged=n_test_img, flow=flow) seq_vis_hat = torch.cat([x_t.unsqueeze(1), x_seq_hat], 1) seq_vis_gt = torch.cat([x_t.unsqueeze(1), x_seq_gt], 1) grid_anim = make_video(x_t,poke,seq_vis_hat,seq_vis_gt,n_logged=n_test_img,flow=flow) it = engine.state.iteration log_dict = {"Last Frame Comparison Test data": wandb.Image(grid_dyn, caption=f"Last frames test grid #{it}."), "Disentanglement Grid Test Data": wandb.Image(grid_dis, caption=f"Test grid disentanglement #{it}."), "Video Grid Test Data": wandb.Video(grid_anim,caption=f"Test Video Grid #{it}.",fps=5)} if self.config["testing"]["eval_app_transfer"]: app_img_unrelated = eval_batch["app_img_random"].cuda(self.all_devices[0]) x_transferred, *_ = net(app_img_unrelated,x_t, poke,len=0) transfer_grid = make_img_grid(app_img_unrelated,x_t,x_transferred) log_dict.update({"Appearance transfer grid Test Data": wandb.Image(transfer_grid, caption=f"Test_grid appearance transfer #{it}")}) wandb.log(log_dict) return None self.logger.info("Initialize engines...") trainer = Engine(train_step) evaluator = Engine(eval_step) test_img_generator = Engine(eval_visual) self.logger.info("Finish engine initialization...") # checkpointing ckpt_handler = ModelCheckpoint(self.dirs["ckpt"], "reg_ckpt", n_saved=10, require_empty=False) if self.config["training"]["two_stage"]: save_dict = {"model": net, "optimizer_dis": optimizer_dis, "optimizer_dyn": optimizer_dyn} else: save_dict = {"model": net, "optimizer_dyn": optimizer_dyn} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler, save_dict) if self.use_gan: ckpt_handler_disc = ModelCheckpoint(self.dirs["ckpt"], gan_trainer.load_key, n_saved=10, require_empty=False) save_dict_disc = {"model": gan_trainer.disc, "optimizer": gan_trainer.disc_opt} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler_disc, save_dict_disc) if self.use_temp_disc: ckpt_handler_disc_temp = ModelCheckpoint(self.dirs["ckpt"], gan_trainer_temp.load_key, n_saved=10, require_empty=False) save_dict_disc_temp = {"model": gan_trainer_temp.disc, "optimizer": gan_trainer_temp.disc_opt} trainer.add_event_handler(Events.ITERATION_COMPLETED(every=self.config["testing"]["ckpt_intervall"]), ckpt_handler_disc_temp, save_dict_disc_temp) pbar = ProgressBar(ascii=True) pbar.attach(trainer, output_transform=lambda x: x) pbar.attach(evaluator, output_transform=lambda x: x) #reduce the learning rate of the decoder for the image reconstruction task, such that the model focusses more on t --> tk if self.config["training"]["two_stage"]: @trainer.on(Events.ITERATION_COMPLETED) def update_lr(engine): self.lr_dec_t = lr_dec_rec(engine.state.iteration) for g in optimizer_dis.param_groups: if g["name"] == "decoder": g["lr"] = self.lr_dec_t @trainer.on(Events.ITERATION_COMPLETED(every=self.config["testing"]["log_intervall"])) def log(engine): it = engine.state.iteration wandb.log({"iteration": it}) # log losses for key in engine.state.output: wandb.log({key: engine.state.output[key]}) data = engine.state.batch if test_dataset.flow_weights: poke = data["poke"][0].cuda(self.all_devices[0]) else: poke = data["poke"].cuda(self.all_devices[0]) x_t = data["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = data["images"][:, 1:].cuda(self.all_devices[0]) if self.config["architecture"]["disentanglement"]: shape_img = data["img_aT"].cuda(self.all_devices[0]) else: shape_img = x_t net.eval() seq_len = x_seq_gt.shape[1] with torch.no_grad(): x_seq_hat, *_ = net(x_t, x_t, poke, len=seq_len) x_t_hat, *_ = net(x_seq_gt[:,-1], shape_img, poke,len=0) #x_t_hat_e, *_ = net(img_aT, img_sT, poke) grid_dis_i = make_img_grid(x_seq_gt[:,-1], shape_img, x_t_hat, x_t, n_logged=n_test_img) grid_dyn = make_flow_grid(x_t, poke, x_seq_hat[:,-1], x_seq_gt[:,-1], n_logged=n_test_img) seq_vis_hat = torch.cat([x_t.unsqueeze(1),x_seq_hat],1) seq_vis_gt = torch.cat([x_t.unsqueeze(1), x_seq_gt], 1) grid_anim = make_video(x_t,poke,seq_vis_hat,seq_vis_gt,n_logged=n_test_img) wandb.log({"Last Frame Comparison Train Data": wandb.Image(grid_dyn, caption=f"Last frames train grid after {it} train steps."), "Disentanglement Grid Invariance Train Data": wandb.Image(grid_dis_i, caption=f"Invariance Disentanglement Grid on train set after {it} train steps."), "Video Grid Train Data": wandb.Video(grid_anim, caption=f"Train Video Grid after {it} train steps",fps=5)}) #"Disentanglement Grid Equivariance Train Data": wandb.Image(grid_dis_e, caption=f"Eqiuvariance Disentanglement Grid on train set after {it} train steps.") self.logger.info("Initialize metrics...") # compute loss average over epochs # Average(output_transform=lambda x: x["loss_dis"]).attach(trainer, "loss_dis-epoch_avg") Average(output_transform=lambda x: x["loss_dis"]).attach(trainer, "loss_dis-epoch_avg") Average(output_transform=lambda x: x["vgg_loss_dyn"]).attach(trainer, "vgg_loss_dyn-epoch_avg") Average(output_transform=lambda x: x["latent_loss_dyn"]).attach(trainer, "latent_loss_dyn-epoch_avg") if self.config["architecture"]["disentanglement"]: Average(output_transform=lambda x: x["style_loss"]).attach(trainer, "style_loss-epoch_avg") Average(output_transform=lambda x: x["style_loss_eval"]).attach(evaluator, "style_loss_eval") if self.use_norm_loss: Average(output_transform=lambda x: x["norm_loss"]).attach(trainer, "norm_loss-epoch_avg") Average(output_transform=lambda x: x["norm_loss"]).attach(evaluator, "norm_loss_eval") if self.config["architecture"]["dynamics_var"]: Average(output_transform=lambda x: x["kl_dyn"]).attach(trainer, "kl_dyn_loss-epoch_avg") if self.use_temp_disc or self.use_gan: def gan_training_started(engine,epoch, key="gan"): return engine.state.iteration >= self.config[key]["start_iteration"] if self.use_gan: use_patchgan_metrics = MetricUsage(started=Events.EPOCH_STARTED(event_filter=gan_training_started), completed=Events.EPOCH_COMPLETED(event_filter=gan_training_started), iteration_completed=Events.ITERATION_COMPLETED(event_filter=gan_training_started)) # gan losses Average(output_transform=lambda x: x["loss_gen_patch"]).attach(trainer, "loss_gen_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["loss_fmap_patch"]).attach(trainer, "loss_fmap_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["loss_disc_patch"]).attach(trainer, "loss_disc_patch-epoch_avg",usage=use_patchgan_metrics) #if self.config["gan"]["gp_weighflow_video_generatort"] > 0: Average(output_transform=lambda x: x["loss_gp_patch"]).attach(trainer, "loss_gp_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["p_""true_patch"]).attach(trainer, "p_true_patch-epoch_avg",usage=use_patchgan_metrics) Average(output_transform=lambda x: x["p_fake_patch"]).attach(trainer, "p_fake_patch-epoch_avg",usage=use_patchgan_metrics) @trainer.on(Events.EPOCH_COMPLETED(event_filter=gan_training_started)) def gan_stuff(engine): gan_trainer.disc_scheduler.step() if self.use_temp_disc: use_tmpgan_metrics = MetricUsage(started=Events.EPOCH_STARTED(event_filter=partial(gan_training_started,key="gan_temp")), completed=Events.EPOCH_COMPLETED(event_filter=partial(gan_training_started,key="gan_temp")), iteration_completed=Events.ITERATION_COMPLETED(event_filter=partial(gan_training_started,key="gan_temp"))) # gan losses Average(output_transform=lambda x: x["loss_gen_temp"]).attach(trainer, "loss_gen_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["loss_fmap_temp"]).attach(trainer, "loss_fmap_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["loss_disc_temp"]).attach(trainer, "loss_disc_temp-epoch_avg",usage=use_tmpgan_metrics) #if self.config["gan"]["gp_weight"] > 0: Average(output_transform=lambda x: x["loss_gp_temp"]).attach(trainer, "loss_gp_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["p_true_temp"]).attach(trainer, "p_true_temp-epoch_avg",usage=use_tmpgan_metrics) Average(output_transform=lambda x: x["p_fake_temp"]).attach(trainer, "p_fake_temp-epoch_avg",usage=use_tmpgan_metrics) @trainer.on(Events.EPOCH_COMPLETED(event_filter=gan_training_started)) def temp_disc_stuff(engine): gan_trainer_temp.disc_scheduler.step() # if self.pixel_decoder_loss: Average(output_transform=lambda x: x["pixel_loss_dec"]).attach(trainer, "pixel_loss_dec-epoch_avg") Average(output_transform=lambda x: x["pixel_loss_dec_eval"]).attach(evaluator, "pixel_loss_dec_eval") # evaluation losses Average(output_transform=lambda x: x["vgg_loss_dyn_eval"]).attach(evaluator, "vgg_loss_dyn_eval") Average(output_transform=lambda x: x["loss_dis_i_eval"]).attach(evaluator, "loss_dis_i_eval") Average(output_transform=lambda x: x["latent_loss_dyn_eval"]).attach(evaluator, "latent_loss_dyn_eval") self.logger.info("Finish metric initialization.") @trainer.on(Events.EPOCH_COMPLETED(every=self.config["testing"]["n_epoch_metrics"])) def metrics(engine): # set incpetion model to cpu self.inception_model.eval() self.inception_model.cuda(self.all_devices[0]) self.lpips_fn.cuda(self.all_devices[0]) self.lpips_fn.eval() if self.use_temp_disc: gan_trainer_temp.disc.cpu() if self.use_gan: gan_trainer.disc.cpu() # compute metrics over an epoch self.logger.info(f"Computing metrics after epoch #{engine.state.epoch}") bs = 20 if self.is_debug else (int(8000 / eval_sampler.batch_size) if len(test_dataset) > 8000 else len(eval_loader)) evaluator.run(eval_loader, max_epochs=1, epoch_length=bs) [wandb.log({key: evaluator.state.metrics[key]}) for key in evaluator.state.metrics] # compute metrics ssim_t = np.mean(np.stack(self.ssims["t"], axis=0)) psnr_t = np.mean(np.stack(self.psnrs["t"], axis=0)) lpips_t = np.mean(np.concatenate(self.lpips["t"], axis=0)) ssim_tk = np.mean(np.stack(self.ssims["tk"], axis=0)) psnr_tk = np.mean(np.stack(self.psnrs["tk"], axis=0)) lpips_avg = np.mean(np.concatenate(self.lpips["tk"], axis=0)) self.lpips_avg = lpips_avg fid_per_frame = {} for key in tqdm(self.fid_feats_real_per_frame, desc="Computing FID per frame"): fid_per_frame[key] = metric_fid(self.fid_feats_real_per_frame[key], self.fid_feats_fake_per_frame[key]) # fid_tk = metric_fid(self.features_real_fid["tk"], self.features_fake_fid["tk"]) fid_avg = np.mean([fid_per_frame[key] for key in fid_per_frame]) log_dict = {"ssim-t": ssim_t, "psnr-t": psnr_t, "fid-avg": fid_avg, "lpips-t": lpips_t,"ssim-tk": ssim_tk, "psnr-tk": psnr_tk, "lpips-avg": lpips_avg} # add histograms for per-frame-metrics self.lpips_per_frame = {key: np.concatenate(self.lpips_per_frame[key], axis=0).mean() for key in self.lpips_per_frame} self.ssims_per_frame = {key: np.stack(self.ssims_per_frame[key], axis=0).mean() for key in self.ssims_per_frame} self.psnrs_per_frame = {key: np.stack(self.psnrs_per_frame[key], axis=0).mean() for key in self.psnrs_per_frame} # self.ssims_per_frame_pl = {key: np.stack(self.ssims_per_frame_pl[key], axis=0).mean() for key in self.ssims_per_frame_pl} # self.psnrs_per_frame_pl = {key: np.stack(self.psnrs_per_frame_pl[key], axis=0).mean() for key in self.psnrs_per_frame_pl} x = [k+1 for k in self.lpips_per_frame] make_plot(x,list(self.lpips_per_frame.values()),"LPIPS of predicted frames", ylabel="Average LPIPS",) make_plot(x, list(self.ssims_per_frame.values()), "SSIM of predicted frames", ylabel="Average SSIM",) make_plot(x, list(self.psnrs_per_frame.values()), "PSNR of predicted frames", ylabel="Average PSNR",) make_plot(x, list(fid_per_frame.values()), "FIDs of predicted frames", ylabel="FID") wandb.log(log_dict) # clear collection arrays self.__clear_metric_arrs() self.inception_model.cpu() self.lpips_fn.cpu() if self.use_temp_disc: gan_trainer_temp.disc.cuda(self.all_devices[0]) if self.use_gan: gan_trainer.disc.cuda(self.all_devices[0]) @trainer.on(Events.ITERATION_COMPLETED(every=self.config["testing"]["test_img_intervall"])) def make_test_grid(engine): test_img_generator.run(test_loader, max_epochs=1, epoch_length=self.config["testing"]["test_it"]) @trainer.on(Events.EPOCH_COMPLETED) def log_train_avg(engine): wandb.log({"epoch": engine.state.epoch}) [wandb.log({key: engine.state.metrics[key]}) for key in engine.state.metrics] # also perform scheduler step if self.config["training"]["two_stage"]: scheduler_dis.step() scheduler_dyn.step() def score_fn(engine): assert self.lpips_avg is not None return -self.lpips_avg # define best ckpt best_ckpt_handler = ModelCheckpoint(self.dirs["ckpt"],filename_prefix="ckpt_metric" ,score_function=score_fn,score_name="lpips",n_saved=5,require_empty=False) trainer.add_event_handler(Events.EPOCH_COMPLETED(every=self.config["testing"]["n_epoch_metrics"]),best_ckpt_handler,save_dict) @trainer.on(Events.STARTED) def set_start_it(engine): self.logger.info(f'Engine starting from iteration {start_it}, epoch {start_epoch}') engine.state.iteration = start_it engine.state.epoch = start_epoch # run everything n_step_per_epoch = 10 if self.is_debug else len(train_loader) self.logger.info("Start training...") trainer.run(train_loader, max_epochs=n_epoch_overall, epoch_length=n_step_per_epoch) self.logger.info("End training.") def test(self): from tqdm import tqdm import cv2 from os import makedirs,path # load checkpoint mod_ckpt, _ = self._load_ckpt("reg_ckpt", single_opt=False,use_best=self.config["testing"]["best_ckpt"],) #dir="/export/data/ablattma/visual_poking/final_models/final_models/var_length/iper/",name="baseline_wo_upsampling.pt") dataset, transforms = get_dataset(config=self.config["data"]) test_dataset = dataset(transforms, self.datakeys, self.config["data"], train=False) # get datasets for training and testing def w_init_fn(worker_id): return np.random.seed(np.random.get_state()[1][0] + worker_id) test_sampler = SequenceLengthSampler(test_dataset, batch_size=self.config["testing"]["test_batch_size"] if self.config["testing"]["test_batch_size"] > 16 else 16, shuffle=False, drop_last=True, zero_poke=False) test_loader = DataLoader( test_dataset, batch_sampler=test_sampler, num_workers=0 if self.is_debug else self.config["data"]["num_workers"], worker_init_fn=w_init_fn ) # define model self.logger.info(f"Load model...") net_model = SkipSequenceModel if self.config["architecture"]["use_skip_model"] else ResidualSequenceBaseline net = net_model(spatial_size=self.config["data"]["spatial_size"], config=self.config["architecture"],) weights = [5,5,0] self.logger.info( f"Number of trainable parameters in model is {sum(p.numel() for p in net.parameters())}" ) net.load_state_dict(mod_ckpt) net.cuda(self.all_devices[0]) self.logger.info("Model on gpu!") net.eval() if self.config["testing"]["mode"] == "noise_comp": target_dir = path.join(self.dirs["generated"], "test_decoder_noise") makedirs(target_dir, exist_ok=True) n_gen = 0 for i,batch in enumerate(test_loader): if n_gen > 100: break if test_dataset.flow_weights: poke = batch["poke"][0].cuda(self.all_devices[0]) else: poke = batch["poke"].cuda(self.all_devices[0]) img = batch["images"][:,0].cuda(self.all_devices[0]) with torch.no_grad(): x_rec, sigmas, *_ = net(img,img,poke,len=0) # add noise to sigmas for n in tqdm(range(x_rec.shape[0]),desc=f'Generating noise xs for batch #{i}'): simgs_noise = [weights[k] * torch.randn((self.config["testing"]["n_examples_noise"],*sigm[n].shape)).cuda(self.all_devices[0])+sigm[n].unsqueeze(0) for k,sigm in enumerate(sigmas)] # xs = torch.stack([img[n]]*self.config["testing"]["n_examples"]) simgs_noise.reverse() xs_noise = net.dec(simgs_noise,None,del_shape=True) xs_disp = torch.cat([x_rec[n].unsqueeze(0),xs_noise],dim=0) xs_disp = np.concatenate([((x.permute(1,2,0).cpu().numpy()+ 1.)*127.5).astype(np.uint8) for x in xs_disp],axis=1) n_gen += 1 xs_disp = cv2.cvtColor(xs_disp,cv2.COLOR_RGB2BGR) cv2.imwrite(path.join(target_dir,f"noise_imgs_{n_gen}.png"),xs_disp) elif self.config["testing"]["mode"] == "metrics": fid_feats_real_per_frame = {} fid_feats_fake_per_frame = {} def metric_step(engine, eval_batch): net.eval() out_dict = {} with torch.no_grad(): # prepare data weights = None if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) weights = eval_batch["poke"][1].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) # eval forward passes seq_len = x_seq_gt.shape[1] x_t_hat, sigma_t, _, alpha = net(x_t, x_t, poke, len=0) x_seq_hat, _, sigmas_hat, _ = net(x_t, x_t, poke, len=seq_len) if weights is not None and self.config["testing"]["metrics_on_patches"]: x_seq_hat = get_patches(x_seq_hat, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) x_seq_gt = get_patches(x_seq_gt, weights, self.config["data"], test_dataset.weight_value_flow, logger=self.logger) # apply inception model for fid calculation at time t+k for t in range(x_seq_gt.shape[1]): real_features_t = self.inception_model(x_seq_gt[:, t]).cpu().numpy() fake_features_t = self.inception_model(x_seq_hat[:, t]).cpu().numpy() if t not in fid_feats_fake_per_frame: fid_feats_fake_per_frame.update({t:fake_features_t}) fid_feats_real_per_frame.update({t: real_features_t}) else: fid_feats_fake_per_frame[t] = np.concatenate([fid_feats_fake_per_frame[t],fake_features_t], axis=0) fid_feats_real_per_frame[t] = np.concatenate([fid_feats_real_per_frame[t], real_features_t], axis=0) #self.features_real_fid["tk"].append(real_features_tk) #self.features_fake_fid["tk"].append(fake_features_tk) self.fvd_vids_real.append(x_seq_gt.cpu()) self.fvd_vids_fake.append(x_seq_hat.cpu()) ssim_tk, ssim_per_frame = ssim_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) psnr_tk, psnr_per_frame = psnr_lightning(x_seq_gt, x_seq_hat, return_per_frame=True) lpips_avg, lpips_per_frame = metric_lpips(x_seq_gt, x_seq_hat, self.lpips_fn, reduce=False, return_per_frame=True) # append to arrays self.psnrs["tk"].append(psnr_tk) self.ssims["tk"].append(ssim_tk) self.lpips["tk"].append(lpips_avg) # append the values of the respective sequence length [self.ssims_per_frame[key].append(ssim_per_frame[key]) if key in self.ssims_per_frame else self.ssims_per_frame.update({key: [ssim_per_frame[key]]}) for key in ssim_per_frame] [self.psnrs_per_frame[key].append(psnr_per_frame[key]) if key in self.psnrs_per_frame else self.psnrs_per_frame.update({key: [psnr_per_frame[key]]}) for key in psnr_per_frame] [self.lpips_per_frame[key].append(lpips_per_frame[key]) if key in self.lpips_per_frame else self.lpips_per_frame.update({key: [lpips_per_frame[key]]}) for key in lpips_per_frame] return out_dict evaluator = Engine(metric_step) self.logger.info("Initialize inception model...") self.inception_model = FIDInceptionModel() self.logger.info("Finished initialization of inception model...") # note that lpips is exactly vgg-cosine similarity as proposed in the google papers and savp self.lpips_fn = LPIPS(net="vgg") pbar = ProgressBar(ascii=True) pbar.attach(evaluator, output_transform=lambda x: x) # set incpetion model to cpu self.inception_model.eval() self.inception_model.cuda(self.all_devices[0]) self.lpips_fn.cuda(self.all_devices[0]) self.lpips_fn.eval() # compute metrics over an epoch self.logger.info(f"Start metrics computation.") el = (int(8000 / test_sampler.batch_size) if len(test_dataset) > 8000 else len(test_loader)) evaluator.run(test_loader, max_epochs=1, epoch_length=el) # [wandb.log({key: evaluator.state.metrics[key]}) for key in evaluator.state.metrics] # compute metrics ssim_tk = np.mean(np.stack(self.ssims["tk"], axis=0)) psnr_tk = np.mean(np.stack(self.psnrs["tk"], axis=0)) lpips_avg = np.mean(np.concatenate(self.lpips["tk"], axis=0)) assert list(fid_feats_real_per_frame.keys()) == list(fid_feats_fake_per_frame.keys()) fid_per_frame = {} for key in tqdm(fid_feats_real_per_frame, desc="Computing FID per frame"): fid_per_frame[key] = metric_fid(fid_feats_real_per_frame[key], fid_feats_fake_per_frame[key]) #fid_tk = metric_fid(self.features_real_fid["tk"], self.features_fake_fid["tk"]) fid_avg = np.mean([fid_per_frame[key] for key in fid_per_frame]) log_dict = {"ssim-avg-temp": ssim_tk, "psnr-avg_temp": psnr_tk, "fid-avg_temp": fid_avg, "lpips-avg-temp": lpips_avg} # add histograms for per-frame-metrics self.lpips_per_frame = {key: np.concatenate(self.lpips_per_frame[key], axis=0).mean() for key in self.lpips_per_frame} self.ssims_per_frame = {key: np.stack(self.ssims_per_frame[key], axis=0).mean() for key in self.ssims_per_frame} self.psnrs_per_frame = {key: np.stack(self.psnrs_per_frame[key], axis=0).mean() for key in self.psnrs_per_frame} savedir = path.join(self.dirs["generated"],"metric_summaries") makedirs(savedir, exist_ok=True) x = [k+1 for k in self.lpips_per_frame] make_plot(x,list(self.lpips_per_frame.values()),"LPIPS of predicted frames", ylabel="Average LPIPS",savename=path.join(savedir,"lpips.svg")) make_plot(x, list(self.ssims_per_frame.values()), "SSIM of predicted frames", ylabel="Average SSIM", savename=path.join(savedir, "ssim.svg")) make_plot(x, list(self.psnrs_per_frame.values()), "PSNR of predicted frames", ylabel="Average PSNR", savename=path.join(savedir, "psnr.svg")) make_plot(x,list(fid_per_frame.values()), "FIDs of predicted frames", ylabel="FID", savename=path.join(savedir, "fid.svg")) self.logger.info("Averaged metrics: ") for key in log_dict: self.logger.info(f'{key}: {log_dict[key]}') elif self.config["testing"]["mode"] == "fvd": test_sampler = SequenceLengthSampler(test_dataset,16,shuffle=True, drop_last=True, n_frames=10,zero_poke=False) test_loader = DataLoader(test_dataset,batch_sampler=test_sampler,num_workers=self.config["data"]["num_workers"],worker_init_fn=w_init_fn) real_samples = [] fake_samples = [] def generate_vids(engine,eval_batch): net.eval() with torch.no_grad(): # prepare data if test_dataset.flow_weights: poke = eval_batch["poke"][0].cuda(self.all_devices[0]) else: poke = eval_batch["poke"].cuda(self.all_devices[0]) x_t = eval_batch["images"][:, 0].cuda(self.all_devices[0]) x_seq_gt = eval_batch["images"][:, 1:].cuda(self.all_devices[0]) # eval forward passes seq_len = x_seq_gt.shape[1] x_seq_hat, *_ = net(x_t, x_t, poke, len=seq_len) real_batch = ((x_seq_gt + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy().astype(np.uint8) fake_batch = ((x_seq_hat + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy().astype(np.uint8) real_samples.append(real_batch) fake_samples.append(fake_batch) generator = Engine(generate_vids) pbar = ProgressBar(ascii=True) pbar.attach(generator, output_transform=lambda x: x) self.logger.info(f"Start collecting sequences for fvd computation...") el = (int(1000 / test_sampler.batch_size) if len(test_dataset) > 1000 else len(test_loader)) generator.run(test_loader,max_epochs=1,epoch_length=el) savedir = path.join(self.dirs["generated"],"samples_fvd") makedirs(savedir,exist_ok=True) real_samples = np.stack(real_samples,axis=0) fake_samples = np.stack(fake_samples, axis=0) np.save(path.join(savedir,"real_samples.npy"),real_samples) np.save(path.join(savedir,"fake_samples.npy"),fake_samples) self.logger.info(f'Finish generation of vid samples.') else: raise ValueError(f'Specified testing mode "{self.config["testing"]["mode"]}" does not exist.')
58,565
53.581547
210
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/frechet_video_distance.py
# coding=utf-8 # Copyright 2020 The Google Research Authors. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python2, python3 """Minimal Reference implementation for the Frechet Video Distance (FVD). FVD is a metric for the quality of video generation models. It is inspired by the FID (Frechet Inception Distance) used for images, but uses a different embedding to be better suitable for videos. """ from __future__ import absolute_import from __future__ import division from __future__ import print_function import six import tensorflow.compat.v1 as tf import tensorflow_gan as tfgan import tensorflow_hub as hub def preprocess(videos, target_resolution): """Runs some preprocessing on the videos for I3D model. Args: videos: <T>[batch_size, num_frames, height, width, depth] The videos to be preprocessed. We don't care about the specific dtype of the videos, it can be anything that tf.image.resize_bilinear accepts. Values are expected to be in the range 0-255. target_resolution: (width, height): target video resolution Returns: videos: <float32>[batch_size, num_frames, height, width, depth] """ videos_shape = videos.shape.as_list() all_frames = tf.reshape(videos, [-1] + videos_shape[-3:]) resized_videos = tf.image.resize_bilinear(all_frames, size=target_resolution) target_shape = [videos_shape[0], -1] + list(target_resolution) + [3] output_videos = tf.reshape(resized_videos, target_shape) scaled_videos = 2. * tf.cast(output_videos, tf.float32) / 255. - 1 return scaled_videos def _is_in_graph(tensor_name): """Checks whether a given tensor does exists in the graph.""" try: tf.get_default_graph().get_tensor_by_name(tensor_name) except KeyError: return False return True class Embedder: def __init__(self, videos): module_name = "fvd_kinetics-400_id3_module_" + six.ensure_str( videos.name).replace(":", "_") module_spec = "https://tfhub.dev/deepmind/i3d-kinetics-400/1" self.model = hub.Module(module_spec, name=module_name) def create_id3_embedding(self,videos): """Embeds the given videos using the Inflated 3D Convolution network. Downloads the graph of the I3D from tf.hub and adds it to the graph on the first call. Args: videos: <float32>[batch_size, num_frames, height=224, width=224, depth=3]. Expected range is [-1, 1]. Returns: embedding: <float32>[batch_size, embedding_size]. embedding_size depends on the model used. Raises: ValueError: when a provided embedding_layer is not supported. """ batch_size = 16 module_spec = "https://tfhub.dev/deepmind/i3d-kinetics-400/1" # Making sure that we import the graph separately for # each different input video tensor. module_name = "fvd_kinetics-400_id3_module_" + six.ensure_str( videos.name).replace(":", "_") assert_ops = [ tf.Assert( tf.reduce_max(videos) <= 1.001, ["max value in frame is > 1", videos]), tf.Assert( tf.reduce_min(videos) >= -1.001, ["min value in frame is < -1", videos]), tf.assert_equal( tf.shape(videos)[0], batch_size, ["invalid frame batch size: ", tf.shape(videos)], summarize=6), ] with tf.control_dependencies(assert_ops): videos = tf.identity(videos) module_scope = "%s_apply_default/" % module_name # To check whether the module has already been loaded into the graph, we look # for a given tensor name. If this tensor name exists, we assume the function # has been called before and the graph was imported. Otherwise we import it. # Note: in theory, the tensor could exist, but have wrong shapes. # This will happen if create_id3_embedding is called with a frames_placehoder # of wrong size/batch size, because even though that will throw a tf.Assert # on graph-execution time, it will insert the tensor (with wrong shape) into # the graph. This is why we need the following assert. video_batch_size = int(videos.shape[0]) assert video_batch_size in [batch_size, -1, None], "Invalid batch size" tensor_name = module_scope + "RGB/inception_i3d/Mean:0" if not _is_in_graph(tensor_name): # i3d_model = hub.Module(module_spec, name=module_name) self.model(videos) # gets the kinetics-i3d-400-logits layer tensor_name = module_scope + "RGB/inception_i3d/Mean:0" tensor = tf.get_default_graph().get_tensor_by_name(tensor_name) return tensor def calculate_fvd(real_activations, generated_activations): """Returns a list of ops that compute metrics as funcs of activations. Args: real_activations: <float32>[num_samples, embedding_size] generated_activations: <float32>[num_samples, embedding_size] Returns: A scalar that contains the requested FVD. """ return tfgan.eval.frechet_classifier_distance_from_activations( real_activations, generated_activations)
5,688
35.941558
83
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/losses.py
import torch from torch import nn from torchvision.models import vgg19 from collections import namedtuple from operator import mul from functools import reduce from utils.general import get_member VGGOutput = namedtuple( "VGGOutput", ["input", "relu1_2", "relu2_2", "relu3_2", "relu4_2", "relu5_2"], ) StyleLayers = namedtuple("StyleLayers",["relu1_2","relu2_2","relu3_3", "relu4_3"]) class PerceptualVGG(nn.Module): def __init__(self, weights=None): super().__init__() self.vgg = vgg19(pretrained=True) self.vgg.eval() self.vgg_layers = self.vgg.features self.register_buffer( "mean", torch.tensor([0.485, 0.456, 0.406], dtype=torch.float) .unsqueeze(dim=0) .unsqueeze(dim=-1) .unsqueeze(dim=-1), ) self.register_buffer( "std", torch.tensor([0.229, 0.224, 0.225], dtype=torch.float) .unsqueeze(dim=0) .unsqueeze(dim=-1) .unsqueeze(dim=-1), ) self.target_layers = { "3": "relu1_2", "8": "relu2_2", "13": "relu3_2", "15" : "relu3_3", "22": "relu4_2", "24" : "relu4_3", "31": "relu5_2", } if weights is None: self.loss_weights = {"input":1., "relu1_2": 1.,"relu2_2": 1.,"relu3_2": 1.,"relu3_3": 1.,"relu4_2": 1.,"relu4_3": 1.,"relu5_2": 1. } else: assert isinstance(weights, dict) and list(weights.keys()) == list(self.target_layers.keys()), f"The weights passed to PerceptualVGG have to be a dict with the keys {list(self.target_layers.keys())}" self.loss_weights = weights def forward(self, x): # IMPORTANT: Input is assumed to be in range [-1,1] here. x = (x + 1.0) / 2.0 x = (x - self.mean) / self.std # add also common reconstruction loss in pixel space out = {"input": x} for name, submodule in self.vgg_layers._modules.items(): # x = submodule(x) if name in self.target_layers: x = submodule(x) out[self.target_layers[name]] = x else: x = submodule(x) return out def vgg_loss(custom_vgg:PerceptualVGG, target, pred, weights=None): """ Implements a vgg based perceptual loss, as extensively used for image/video generation tasks :param custom_vgg: The vgg feature extractor for the perceptual loss, definition see above :param target: :param pred: :return: """ target_feats = custom_vgg(target) pred_feats = custom_vgg(pred) target_feats = VGGOutput(**{key: target_feats[key] for key in VGGOutput._fields}) pred_feats = VGGOutput(**{key: pred_feats[key] for key in VGGOutput._fields}) names = list(pred_feats._asdict().keys()) if weights is None: losses = {} for i, (tf, pf) in enumerate(zip(target_feats, pred_feats)): loss = get_member(custom_vgg,"loss_weights")[VGGOutput._fields[i]] * torch.mean( torch.abs(tf - pf) ).unsqueeze(dim=-1) losses.update({names[i]: loss}) else: losses = { names[0]: get_member(custom_vgg,"loss_weights")[VGGOutput._fields[0]] * torch.mean(weights * torch.abs(target_feats[0] - pred_feats[0])) .unsqueeze(dim=-1) .to(torch.float) } for i, (tf, pf) in enumerate(zip(target_feats[1:], pred_feats[1:])): loss = get_member(custom_vgg,"loss_weights")[i + 1] * torch.mean( torch.abs(tf - pf) ).unsqueeze(dim=-1) losses.update({names[i + 1]: loss}) return losses def vgg_loss_agg(vgg, target, pred, weights=None): """ To aggreagate the vgg losses :param vgg: :param target: :param pred: :param weights: :return: """ # basic_device = target.get_device() # net_device = list(vgg.parameters())[0].get_device() # pred = pred.cuda(net_device) # target = target.cuda(net_device) loss_list = vgg_loss(vgg,target,pred,weights) loss_tensor = torch.stack([loss_list[key] for key in loss_list],dim=0,) return loss_tensor.sum()#.cuda(basic_device) class PixelDynamicsLoss(nn.Module): def __init__(self, diff_pp=False): super().__init__() self.diff_pp = diff_pp def forward(self,target_t,target_tk,pred_t,pred_tk): if self.diff_pp: loss = (((target_t-target_tk).abs()-(pred_t.detach()-pred_tk).abs()).mean())**2 else: loss = ((target_t-target_tk).abs().mean()-(pred_t.detach()-pred_tk).abs().mean())**2 return loss def pixel_triplet_loss(target_t,target_tk,pred_t, pred_tk,vgg:PerceptualVGG,layerwise = True, detach=True, diff_pp=False): """ :param vgg: :param target_t: :param target_tk: :param pred_t: :param pred_tk: :param layerwise: :param detach: whether or not to detach the predicted feats at time t :param diff_pp: whether to consider differences for each spatial location in each channel or average over all (default average) :return: """ if layerwise: losses = {} # old_device = target_tk.get_device() # new_device = list(vgg.parameters())[0].get_device() # timestep t # target_t = target_t.cuda(new_device) # pred_t = pred_t.cuda(new_device) target_feats_t = vgg(target_t.cuda()) pred_feats_t = vgg(pred_t.detach() if detach else pred_t) target_feats_t = VGGOutput(**{key: target_feats_t[key] for key in VGGOutput._fields}) pred_feats_t = VGGOutput(**{key: pred_feats_t[key] for key in VGGOutput._fields}) # timestep tk # target_tk = target_tk.cuda(new_device) # pred_tk = pred_tk.cuda(new_device) target_feats_tk = vgg(target_tk) pred_feats_tk = vgg(pred_tk) target_feats_tk = VGGOutput(**{key: target_feats_tk[key] for key in VGGOutput._fields}) pred_feats_tk = VGGOutput(**{key: pred_feats_tk[key] for key in VGGOutput._fields}) names = list(pred_feats_t._asdict().keys()) for i, (tft, pft, tftk, pftk) in enumerate(zip(target_feats_t, pred_feats_t,target_feats_tk, pred_feats_tk)): if diff_pp: loss = get_member(vgg,"loss_weights")[VGGOutput._fields[i]] * torch.mean((torch.abs(tft - tftk) - torch.abs(pft - pftk)) ** 2).unsqueeze(dim=-1) else: loss = get_member(vgg,"loss_weights")[VGGOutput._fields[i]] * (torch.mean(torch.abs(tft - tftk)).unsqueeze(dim=-1) - torch.mean(torch.abs(pft - pftk)).unsqueeze(dim=-1))**2 losses.update({names[i]: loss}) loss_tensor = torch.stack([losses[key] for key in losses], dim=0, ) ptl = loss_tensor.sum() #.cuda(old_device) else: ptl = (vgg_loss_agg(vgg, pred_t.detach(), pred_tk) - vgg_loss_agg(vgg, target_t, target_tk)) ** 2 return ptl def style_loss(vgg,style_target, pred): target_feats = vgg(style_target) pred_feats = vgg(pred) target_feats = StyleLayers(**{key: target_feats[key] for key in StyleLayers._fields}) pred_feats = StyleLayers(**{key: pred_feats[key] for key in StyleLayers._fields}) names = list(pred_feats._asdict().keys()) style_outs = {} # compute gram matrices and take squared frobenius norm for i, (tf,pf) in enumerate(zip(target_feats,pred_feats)): shape = pf.shape pf = pf.reshape(*shape[:2],-1) tf = tf.reshape(*shape[:2],-1) gram_diff = 1. / (shape[1]*shape[2]*shape[3]) * (torch.matmul(pf,pf.permute(0,2,1)) - torch.matmul(tf,tf.permute(0,2,1))) loss = (torch.norm(gram_diff, p="fro",dim=[1,2])**2).mean() style_outs.update({names[i]:loss}) style_outs = torch.stack([style_outs[key] for key in style_outs]) return style_outs.sum() class DynamicsLoss(nn.Module): """ Triplet loss Takes embeddings of an anchor sample, a positive sample and a negative sample """ def __init__(self, config): super().__init__() self.mse = nn.MSELoss() def forward(self, anchor, positive, negative, ): if isinstance(anchor,list) and isinstance(positive,list): losses = [] for a,p in zip(anchor,positive): losses.append(self.mse(a,p)) return torch.stack(losses).mean() else: return self.mse(anchor,positive) def kl_loss_check(latents): """ Estimates a gaussian from the latents and returns the kl_divergence between this gaussian and the standard normal :param latents: :return: """ mu = latents[:,:int(latents.shape[1] / 2)] sigma = latents[:,int(latents.shape[1] / 2):] # reparameterize logstd = nn.Sigmoid()(sigma) return kl_loss(mu,logstd) def kl_loss(mu, logstd): if len(mu.shape) != 2: mu = mu.reshape(mu.shape[0],-1) logstd = logstd.reshape(mu.shape[0],-1) dim = mu.shape[1] std = torch.exp(logstd) kl = torch.sum(-logstd + 0.5 * (std ** 2 + mu ** 2), dim=-1) - (0.5 * dim) return kl.mean()
9,190
33.423221
210
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/metric_fvd.py
import numpy as np import argparse from os import path import torch import ssl from glob import glob from natsort import natsorted ssl._create_default_https_context = ssl._create_unverified_context import cv2 from utils.metrics import compute_fvd from utils.general import get_logger if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument("-s", "--source", type=str, required=True, help="Source directory where the data is stored.") parser.add_argument("--gpu",type=int, required=True, help="The target device.") parser.add_argument("-v","--visualize",default=False,action="store_true") args = parser.parse_args() if not path.isdir(args.source): raise NotADirectoryError(f'The specified, data-holding directory {args.source} is not existing...') file = path.basename(__file__) logger = get_logger(file) logger.info("Read in data...") real_samples_list = natsorted(glob(path.join(args.source, "real_samples_*.npy"))) fake_samples_list = natsorted(glob(path.join(args.source, "fake_samples_*.npy"))) if len(real_samples_list) == 0: fake_samples_list = [path.join(args.source, "fake_samples.npy")] real_samples_list = [path.join(args.source, "real_samples.npy")] for i,(real_samples, fake_samples) in enumerate(zip(real_samples_list,fake_samples_list)): try: length = int(real_samples.split("/")[-1].split(".")[0].split("_")[2]) context = int(real_samples.split("/")[-1].split(".npy")[0].split("_")[-1]) logger.info(f"processing samples of length {length} with {context} context frames.") except: logger.info(f"Processing standard samples") real_samples = np.load(real_samples) fake_samples = np.load(fake_samples) if args.visualize: vis_real = real_samples[0,0] vis_fake = fake_samples[0,0] # visualize writer = cv2.VideoWriter( path.join(args.source, "test_vid_fake.mp4"), cv2.VideoWriter_fourcc(*"MP4V"), 5, (vis_fake.shape[2], vis_fake.shape[1]), ) # writer = vio.FFmpegWriter(savename,inputdict=inputdict,outputdict=outputdict) for frame in vis_fake: frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR) writer.write(frame) writer.release() writer = cv2.VideoWriter( path.join(args.source, "test_vid_real.mp4"), cv2.VideoWriter_fourcc(*"MP4V"), 5, (vis_real.shape[2], vis_real.shape[1]), ) # writer = vio.FFmpegWriter(savename,inputdict=inputdict,outputdict=outputdict) for frame in vis_real: frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR) writer.write(frame) writer.release() real_samples = real_samples[:62] if real_samples.shape[0] > 62 else real_samples fake_samples = fake_samples[:62] if fake_samples.shape[0] > 62 else fake_samples logger.info(f'Number of samples: {len(fake_samples)}') target_device = args.gpu real_samples = list(real_samples) fake_samples = list(fake_samples) real_samples = [torch.from_numpy(r) for r in real_samples] fake_samples = [torch.from_numpy(r) for r in fake_samples] fvd_val = compute_fvd(real_samples,fake_samples,target_device,logger)
3,569
30.59292
107
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/testing.py
import numpy as np import torch from skimage.metrics import structural_similarity as ssim import cv2 import math import imutils import matplotlib.pyplot as plt import wandb from os import path import math def make_flow_grid(src, poke, pred, tgt, n_logged, flow=None): """ :param src: :param poke: :param pred: :param tgt: :param n_logged: :return: """ src = ((src.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] # poke = poke.permute(0, 2, 3, 1).cpu().numpy()[:n_logged] # poke -= poke.min() # poke /= poke.max() # poke = (poke * 255.0).astype(np.uint8) # poke = np.concatenate([poke, np.expand_dims(np.zeros_like(poke).sum(-1), axis=-1)], axis=-1).astype(np.uint8) poke = vis_flow(poke[:n_logged]) poke = np.concatenate(poke,axis=1) # if prediction is image, just take the premuted image if pred.shape[1] == 3: pred = ((pred.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype( np.uint8)[:n_logged] else: # if prediction is flow, you need treat it like that pred = pred.permute(0, 2, 3, 1).cpu().numpy()[:n_logged] pred -= pred.min() pred /= pred.max() pred = (pred * 255.0).astype(np.uint8) pred = np.concatenate([pred, np.expand_dims(np.zeros_like(pred).sum(-1), axis=-1)], axis=-1).astype(np.uint8) tgt = ((tgt.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype( np.uint8)[:n_logged] tgt_gr = [cv2.cvtColor(t,cv2.COLOR_RGB2GRAY) for t in tgt] pred_gr = [cv2.cvtColor(t,cv2.COLOR_RGB2GRAY) for t in pred] ssim_imgs = [ssim(rimg, fimg, multichannel=True, data_range=255, gaussian_weights=True, use_sample_covariance=False, full=True)[1] for rimg, fimg in zip(tgt_gr, pred_gr)] additional = [np.concatenate([cv2.cvtColor((s * 255.).astype(np.uint8),cv2.COLOR_GRAY2RGB) for s in ssim_imgs],axis=1)] if flow is not None: # if provided, use additional flow information (in case of poking, that's the entire flow src --> tgt # add = flow.permute(0, 2, 3, 1).cpu().numpy()[:n_logged] # add -= add.min() # add /= add.max() # add = (add * 255.0).astype(np.uint8) # add = np.concatenate([add, np.expand_dims(np.zeros_like(add).sum(-1), axis=-1)], axis=-1).astype(np.uint8) # additional = additional + [np.concatenate([a for a in add], axis=1)] add = vis_flow(flow[:n_logged]) add = np.concatenate(add,axis=1) additional = additional + [add] # compute ssim_img in grayscale src = np.concatenate([s for s in src], axis=1) #poke = np.concatenate([f for f in poke], axis=1) pred = np.concatenate([p for p in pred], axis=1) tgt = np.concatenate([t for t in tgt], axis=1) grid = np.concatenate([src,poke,pred,tgt,*additional],axis=0) return grid def vis_flow(flow_map, normalize=False): if isinstance(flow_map,torch.Tensor): flow_map = flow_map.cpu().numpy() flows_vis = [] for flow in flow_map: hsv = np.zeros((*flow.shape[1:],3),dtype=np.uint8) hsv[...,1] = 255 mag, ang = cv2.cartToPolar(flow[0], flow[1]) # since 360 is not valid for uint8, 180° corresponds to 360° for opencv hsv representation. Therefore, we're dividing the angle by 2 after conversion to degrees hsv[...,0] = ang * 180 / np.pi / 2 hsv[...,2] = cv2.normalize(mag,None,alpha=0,beta=255, norm_type=cv2.NORM_MINMAX) as_rgb = cv2.cvtColor(hsv,cv2.COLOR_HSV2RGB) if normalize: as_rgb = as_rgb.astype(np.float) - as_rgb.min(axis=(0,1),keepdims=True) as_rgb = (as_rgb / as_rgb.max(axis=(0,1),keepdims=True)*255.).astype(np.uint8) flows_vis.append(as_rgb) return flows_vis def vis_flow_dense(flow_map,**kwargs): if isinstance(flow_map,torch.Tensor): flow_map = flow_map.cpu().numpy() flows_vis = [] for flow in flow_map: h, w = flow.shape[1:] fx, fy = flow[0], flow[1] ang = np.arctan2(fy, fx) + np.pi v = np.sqrt(fx * fx + fy * fy) hsv = np.zeros((h, w, 3), np.uint8) hsv[..., 0] = ang * (180 / np.pi / 2) hsv[..., 1] = 255 hsv[..., 2] = cv2.normalize(v,None,alpha=0,beta=255, norm_type=cv2.NORM_MINMAX) bgr = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR) flows_vis.append(bgr) return flows_vis def make_trf_video(img1,img2,v12,v21,poke,n_logged,logwandb=True,length_divisor=5): """ :param img1: :param img2: :param v12: :param v21: :param poke: :param n_logged: :param lomake_flow_grid()gwandb: :param length_divisor: :return: """ seq_len = v12.shape[1] pokes = vis_flow(poke[:n_logged]) img1 = ((img1.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] img2 = ((img2.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] img1_with_arrow = [] img2_with_arrow = [] eps = 1e-6 for i, (poke_p, img1_i, img2_i) in enumerate(zip(poke[:n_logged], img1, img2)): poke_points = np.nonzero(pokes[i].any(-1) > 0) if poke_points[0].size == 0: img1_with_arrow.append(img1_i) img2_with_arrow.append(img2_i) else: min_y = np.amin(poke_points[0]) max_y = np.amax(poke_points[0]) min_x = np.amin(poke_points[1]) max_x = np.amax(poke_points[1]) # plot mean direction of flow in poke region avg_flow = np.mean(poke_p[:, min_y:max_y, min_x:max_x].cpu().numpy(), axis=(1, 2)) arrow_dir = avg_flow / (np.linalg.norm(avg_flow) + eps) * (poke_p.shape[1] / length_divisor) if not math.isnan(arrow_dir[0]) or not math.isnan(arrow_dir[1]): arrow_start = (int((min_x + max_x) / 2), int((min_y + max_y) / 2)) arrow_end = (arrow_start[0] + int(arrow_dir[0]), arrow_start[1] + int(arrow_dir[1])) img1_with_arrow.append(cv2.UMat.get(cv2.arrowedLine(cv2.UMat(img1_i), arrow_start, arrow_end, (255, 0, 0), max(int(img1_i.shape[0] / 64), 1)))) img2_with_arrow.append(cv2.UMat.get(cv2.arrowedLine(cv2.UMat(img2_i), arrow_start, arrow_end, (255, 0, 0), max(int(img2_i.shape[0] / 64), 1)))) else: img1.append(img1_i) img2.append(img2_i) vid_st1 = np.concatenate(img1_with_arrow, axis=1) vid_st2 = np.concatenate(img2_with_arrow, axis=1) vid_st1 = put_text_to_video_row(np.stack([vid_st1] * seq_len, axis=0), "Image 1", color=(255, 0, 0)) vid_st2 = put_text_to_video_row(np.stack([vid_st2] * seq_len, axis=0), "Image 2", color=(255, 0, 0)) v12 = ((v12.permute(0, 1, 3, 4, 2).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] v12 = np.concatenate(list(v12), axis=2) v12 = put_text_to_video_row(v12, "Vid: FG1-BG2") v21 = ((v21.permute(0, 1, 3, 4, 2).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] v21 = np.concatenate(list(v21), axis=2) v21 = put_text_to_video_row(v21, "Vid: FG2-BG1") full = np.concatenate([vid_st1, vid_st2, v12, v21], axis=1) if logwandb: full = np.moveaxis(full, [0, 1, 2, 3], [0, 2, 3, 1]) return full def draw_arrow(traj): arrow_imgs = [] for c,t in enumerate(traj): active_points = np.nonzero(t.astype(np.uint8).any(0) > 0) img = np.zeros((*t.shape[1:],3),dtype=np.uint8) if active_points[0].size>0: for i in range(active_points[0].shape[0]): y =active_points[0][i] x = active_points[1][i] arrow_dir = t[:,y,x] if not math.isnan(arrow_dir[0]) or not math.isnan(arrow_dir[1]): arrow_start = (x, y) arrow_end = (int(np.clip(x + int(arrow_dir[0]),0,img.shape[0])), int(np.clip(y + int(arrow_dir[1]),0,img.shape[0]))) img = cv2.arrowedLine(img, arrow_start, arrow_end, (255, 0, 0), max(int(traj.shape[1] / 64), 1)) arrow_imgs.append(img) arrow_imgs = np.concatenate(arrow_imgs,axis=1) return arrow_imgs def img_grid_ci(src,traj,pred,tgt,n_logged): src = ((src.permute(0, 2, 3, 1).cpu().numpy()) * 255.).astype(np.uint8)[:n_logged] # if prediction is image, just take the premuted image pred = ((pred.permute(0, 2, 3, 1).cpu().numpy()) * 255).astype( np.uint8)[:n_logged] tgt = ((tgt.permute(0, 2, 3, 1).cpu().numpy()) * 255).astype( np.uint8)[:n_logged] src = np.concatenate([s for s in src], axis=1) # poke = np.concatenate([f for f in poke], axis=1) pred = np.concatenate([p for p in pred], axis=1) tgt = np.concatenate([t for t in tgt], axis=1) arrows = draw_arrow(traj[:n_logged].cpu().numpy()) grid = np.concatenate([src, arrows, pred, tgt], axis=0) return grid def make_video_ci(src,traj,pred,tgt,n_logged,logwandb=True, display_frame_nr=True): seq_len = tgt.shape[1] srcs = np.concatenate([s for s in ((src.permute(0, 2, 3, 1).cpu().numpy()) * 255.).astype(np.uint8)[:n_logged]],axis=1) traj_vis = [] for t in range(traj.shape[1]): arrows = draw_arrow(traj[:n_logged,t].cpu().numpy()) traj_vis.append(arrows) traj_vis = np.stack(traj_vis,axis=0) traj_vis = put_text_to_video_row(traj_vis, "Flow Vectors", display_frame_nr=display_frame_nr) srcs = cv2.UMat.get(cv2.putText(cv2.UMat(srcs), f"Sequence length {seq_len}", (int(srcs.shape[1] // 3), int(srcs.shape[0] / 6)), cv2.FONT_HERSHEY_SIMPLEX, float(srcs.shape[0] / 256), (255, 0, 0), int(srcs.shape[0] / 128))) srcs = np.stack([srcs] * seq_len, axis=0) srcs = put_text_to_video_row(srcs, "Input Image", display_frame_nr=display_frame_nr) pred = ((pred.permute(0, 1, 3, 4, 2).cpu().numpy()) * 255).astype(np.uint8)[:n_logged] pred = np.concatenate(list(pred), axis=2) pred = put_text_to_video_row(pred, "Predicted Video", display_frame_nr=display_frame_nr) tgt = ((tgt.permute(0, 1, 3, 4, 2).cpu().numpy()) * 255).astype(np.uint8)[:n_logged] tgt = np.concatenate(list(tgt), axis=2) tgt = put_text_to_video_row(tgt, "Groundtruth Video", display_frame_nr=display_frame_nr) full = np.concatenate([srcs, pred, tgt, traj_vis], axis=1) if logwandb: full = np.moveaxis(full, [0, 1, 2, 3], [0, 2, 3, 1]) return full def make_video(src,poke,pred,tgt,n_logged,flow=None,length_divisor=5,logwandb=True,flow_weights= None, display_frame_nr=False,invert_poke = False): """ :param src: src image :param poke: poke, also input to the network :param pred: predicted video of the network :param tgt: target video the network was trained to reconstruct :param n_logged: numvber of logged examples :param flow: src flow from which the poke is originating :param length_divisor: divisor for the length of the arrow, that's drawn ti visualize the mean direction of the flow within the poke patch :param logwandb: whether the output video grid is intended to be logged with wandb or not (in this case the grid channels have to be changed) :param flow_weights: Optional weights for the flow which are also displayed if they are not None. :return: """ seq_len = tgt.shape[1] src = ((src.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] pokes = vis_flow(poke[:n_logged]) flows_vis = None if flow is not None: flows = vis_flow(flow[:n_logged]) flows_with_rect = [] for i,(poke_p,flow) in enumerate(zip(pokes,flows)): poke_points = np.nonzero(poke_p.any(-1) > 0) if poke_points[0].size == 0: flows_with_rect.append(np.zeros_like(flow)) else: min_y = np.amin(poke_points[0]) max_y = np.amax(poke_points[0]) min_x = np.amin(poke_points[1]) max_x = np.amax(poke_points[1]) # draw rect flow_with_rect = cv2.rectangle(flow,(min_x,min_y),(max_x,max_y),(255,255,255),max(1,int(flow.shape[0]//64))) # flow_with_rect = cv2.UMat.get(cv2.putText(cv2.UMat(flow_with_rect), f"Flow Complete",(int(flow_with_rect.shape[1] // 3), int(5 * flow_with_rect.shape[0] / 6) ), cv2.FONT_HERSHEY_SIMPLEX, # float(flow_with_rect.shape[0] / 256), (255, 255, 255), int(flow_with_rect.shape[0] / 128))) flows_with_rect.append(flow_with_rect) flow_cat = np.concatenate(flows_with_rect,axis=1) flows_vis= [np.stack([flow_cat]*seq_len,axis=0)] flows_vis[0] = put_text_to_video_row(flows_vis[0], "Flow Complete", color=(255, 255, 255)) if flow_weights is not None: flow_weights = flow_weights.cpu().numpy() heatmaps = [] for i, weight in enumerate(flow_weights): weight_map = ((weight - weight.min()) / weight.max() * 255.).astype(np.uint8) heatmap = cv2.applyColorMap(weight_map, cv2.COLORMAP_HOT) heatmap = cv2.cvtColor(heatmap, cv2.COLOR_RGB2BGR) heatmaps.append(heatmap) heatmaps = np.concatenate(heatmaps, axis=1) heatmaps = np.stack([heatmaps]*seq_len, axis=0) heatmaps = put_text_to_video_row(heatmaps, "Flow Weights", color=(255,255,255)) if flows_vis is None: flows_vis = [heatmaps] else: flows_vis.insert(0,heatmaps) srcs_with_arrow = [] pokes_with_arrow = [] if invert_poke: srcs_with_arrow_inv = [] pokes_with_arrow_inv = [] eps = 1e-6 for i, (poke_p,src_i) in enumerate(zip(poke[:n_logged],src)): poke_points = np.nonzero(pokes[i].any(-1) > 0) if poke_points[0].size==0: pokes_with_arrow.append(np.zeros_like(pokes[i])) srcs_with_arrow.append(src_i) else: min_y = np.amin(poke_points[0]) max_y = np.amax(poke_points[0]) min_x = np.amin(poke_points[1]) max_x = np.amax(poke_points[1]) # plot mean direction of flow in poke region avg_flow = np.mean(poke_p[:, min_y:max_y, min_x:max_x].cpu().numpy(), axis=(1, 2)) arrow_dir = avg_flow / (np.linalg.norm(avg_flow) + eps) * (poke_p.shape[1] / length_divisor) if not math.isnan(arrow_dir[0]) or not math.isnan(arrow_dir[1]): arrow_start = (int((min_x + max_x) / 2), int((min_y + max_y) / 2)) arrow_end = (arrow_start[0] + int(arrow_dir[0]), arrow_start[1] + int(arrow_dir[1])) test = pokes[i] # test = cv2.UMat.get(cv2.putText(cv2.UMat(test), f"Poke", (int(test.shape[1] // 3), int(5 * test .shape[0] / 6)), cv2.FONT_HERSHEY_SIMPLEX, # float(test.shape[0] / 256), (255, 255, 255), int(test.shape[0] / 128))) pokes_with_arrow.append(cv2.arrowedLine(test, arrow_start, arrow_end, (255, 0, 0), max(int(src_i.shape[0] / 64),1))) srcs_with_arrow.append(cv2.UMat.get(cv2.arrowedLine(cv2.UMat(src_i), arrow_start, arrow_end, (255, 0, 0), max(int(src_i.shape[0] / 64),1)))) if invert_poke: arrow_end_inv = (arrow_start[0] - int(arrow_dir[0]), arrow_start[1] - int(arrow_dir[1])) pokes_with_arrow_inv.append(cv2.arrowedLine(test, arrow_start, arrow_end_inv, (0, 255, 0), max(int(src_i.shape[0] / 64), 1))) srcs_with_arrow_inv.append(cv2.UMat.get(cv2.arrowedLine(cv2.UMat(src_i), arrow_start, arrow_end, (0, 255, 0), max(int(src_i.shape[0] / 64), 1)))) else: pokes_with_arrow.append(np.zeros_like(pokes[i])) srcs_with_arrow.append(src_i) poke = np.concatenate(pokes_with_arrow, axis=1) if invert_poke: poke_inv = np.concatenate(pokes_with_arrow_inv, axis=1) poke = put_text_to_video_row(np.stack([*[poke] * int(math.ceil(float(seq_len)/2)),*[poke_inv]*int(seq_len/2)], axis=0),"Pokes",color=(255,255,255)) else: poke = put_text_to_video_row(np.stack([poke] * seq_len, axis=0),"Poke",color=(255,255,255)) if flows_vis is None: flows_vis = [poke] else: flows_vis.append(poke) srcs = np.concatenate(srcs_with_arrow,axis=1) srcs = cv2.UMat.get(cv2.putText(cv2.UMat(srcs), f"Sequence length {seq_len}", (int(srcs.shape[1] // 3), int(srcs.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(srcs.shape[0] / 256), (255, 0, 0), int(srcs.shape[0] / 128))) if invert_poke: srcs_inv = np.concatenate(srcs_with_arrow_inv, axis=1) srcs_inv = cv2.UMat.get(cv2.putText(cv2.UMat(srcs_inv), f"Sequence length {seq_len}", (int(srcs_inv.shape[1] // 3), int(srcs_inv.shape[0] / 6)), cv2.FONT_HERSHEY_SIMPLEX, float(srcs_inv.shape[0] / 256), (255, 0, 0), int(srcs_inv.shape[0] / 128))) srcs = np.stack([*[srcs] * int(math.ceil(float(seq_len)/2)),*[srcs_inv]*int(seq_len/2)],axis=0) else: srcs = np.stack([srcs]*seq_len,axis=0) srcs = put_text_to_video_row(srcs,"Input Image",display_frame_nr=display_frame_nr) pred = ((pred.permute(0, 1, 3, 4, 2).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] pred = np.concatenate(list(pred),axis=2) pred = put_text_to_video_row(pred, "Predicted Video",display_frame_nr=display_frame_nr) tgt = ((tgt.permute(0, 1, 3, 4, 2).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] tgt = np.concatenate(list(tgt),axis=2) tgt = put_text_to_video_row(tgt,"Groundtruth Video",display_frame_nr=display_frame_nr) full = np.concatenate([srcs,pred,tgt,*flows_vis],axis=1) if logwandb: full = np.moveaxis(full,[0,1,2,3],[0,2,3,1]) return full def put_text_to_video_row(video_row,text, color = None,display_frame_nr=False): written = [] for i,frame in enumerate(video_row): current = cv2.UMat.get(cv2.putText(cv2.UMat(frame), text, (int(frame.shape[1] // 3), frame.shape[0] - int(frame.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(frame.shape[0] / 256), (255, 0, 0) if color is None else color, int(frame.shape[0] / 128))) if display_frame_nr: current = cv2.UMat.get(cv2.putText(cv2.UMat(current), str(i+1), (int(frame.shape[1] / 32), frame.shape[0] - int(frame.shape[0] / 6)), cv2.FONT_HERSHEY_SIMPLEX, float(frame.shape[0] / 256), (255, 0, 0) if color is None else color, int(frame.shape[0] / 128))) written.append(current) return np.stack(written) def make_animated_grid(src, poke, pred, tgt, n_logged, flow=None, length_divisor=5,logwandb=True): # visualize flows pokes = vis_flow(poke[:n_logged]) pokes_with_arrow = [] for i,poke_p in enumerate(poke[:n_logged]): poke_points = np.nonzero(pokes[i].any(-1) > 0) min_y = np.amin(poke_points[0]) max_y = np.amax(poke_points[0]) min_x = np.amin(poke_points[1]) max_x = np.amax(poke_points[1]) # plot mean direction of flow in poke region avg_flow = np.mean(poke_p[:,min_y:max_y,min_x:max_x].cpu().numpy(), axis=(1, 2)) arrow_dir = avg_flow / np.linalg.norm(avg_flow) * (poke_p.shape[1] / length_divisor) arrow_start = (int((min_x+max_x)/2),int((min_y+max_y)/2)) arrow_end = (arrow_start[0]+int(arrow_dir[0]),arrow_start[1]+int(arrow_dir[1])) test = pokes[i] pokes_with_arrow.append(cv2.arrowedLine(test,arrow_start,arrow_end,(0,0,255),2)) poke = np.concatenate(pokes_with_arrow, axis=1) flows_vis = [np.stack([poke]*3,axis=0)] if flow is not None: flows = vis_flow(flow[:n_logged]) flows_with_rect = [] for i,(poke_p,flow) in enumerate(zip(pokes,flows)): poke_points = np.nonzero(poke_p.any(-1) > 0) min_y = np.amin(poke_points[0]) max_y = np.amax(poke_points[0]) min_x = np.amin(poke_points[1]) max_x = np.amax(poke_points[1]) # draw rect flow_with_rect = cv2.rectangle(flow,(min_x,min_y),(max_x,max_y),(255,255,255),max(1,int(flow.shape[0]//64))) flows_with_rect.append(flow_with_rect) flow_cat = np.concatenate(flows_with_rect,axis=1) flows_vis.insert(0,np.stack([flow_cat]*3,axis=0)) # visualize images src = ((src.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] pred = ((pred.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] tgt = ((tgt.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] src = np.concatenate(list(src),axis=1) pred = np.concatenate(list(pred), axis=1) tgt = np.concatenate(list(tgt), axis=1) src = cv2.UMat.get(cv2.putText(cv2.UMat(src), "Source", (int(src.shape[0] // 4), 30), cv2.FONT_HERSHEY_SIMPLEX, float(src.shape[0] / 256), (0, 0, 0), int(src.shape[0] / 128))) pred = cv2.UMat.get(cv2.putText(cv2.UMat(pred), "Predicted", (int(pred.shape[0] // 4), 30), cv2.FONT_HERSHEY_SIMPLEX, float(pred.shape[0] / 256), (0, 0, 0), int(pred.shape[0] / 128))) tgt = cv2.UMat.get(cv2.putText(cv2.UMat(tgt), "Target", (int(tgt.shape[0] // 4), 30), cv2.FONT_HERSHEY_SIMPLEX, float(tgt.shape[0] / 256), (0, 0, 0), int(tgt.shape[0] / 128))) animation = np.stack([src,pred,tgt],axis=0) # this generates a video grid which can be used by wandb.Video() full = np.concatenate([animation,*flows_vis],axis=1) # wandb requires video to have shape (time, channels, height, width) if logwandb: full = np.moveaxis(full,[0,1,2,3],[0,2,3,1]) return full def make_generic_grid(data, dtype, n_logged): from utils.visualizer import FlowVisualizer visualizer = FlowVisualizer() final_data = [] assert(len(data)==len(dtype)) for i, batch in enumerate(data): if dtype[i] == "flow": add = batch.permute(0, 2, 3, 1).cpu().numpy()[:n_logged] add -= add.min() add /= add.max() add = (add * 255.0).astype(np.uint8) image = np.concatenate( [add, np.expand_dims(np.zeros_like(add).sum(-1), axis=-1)], axis=-1).astype(np.uint8) elif dtype[i] == "flow_3D": add = batch.permute(0, 2, 3, 1).cpu().numpy()[:n_logged] add -= add.min() add /= add.max() add = (add * 255.0).astype(np.uint8) image = add elif dtype[i] == "img": image = ((batch.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] elif dtype[i] == "diff_flow_amplitude": generated = batch[0][:n_logged].detach().cpu() target = batch[1][:n_logged].detach().cpu() image = visualizer.create_diff_amplitude( visualizer.make_3d_to_2d(generated), visualizer.make_3d_to_2d(target)) image = (image*255).astype(np.uint8)[:, None].repeat(3, axis=1).transpose(0, 2, 3, 1) elif dtype[i] == "diff_flow_direction": generated = batch[0][:n_logged].detach().cpu() target = batch[1][:n_logged].detach().cpu() image = visualizer.create_diff_direction( visualizer.make_3d_to_2d(generated), visualizer.make_3d_to_2d(target)) image = (image * 255).astype(np.uint8)[:, None].repeat(3, axis=1).transpose(0, 2, 3, 1) elif dtype[i] == "diff_flow_clipped": generated = batch[0][:n_logged].permute(0, 2, 3, 1).cpu().numpy() target = batch[1][:n_logged].permute(0, 2, 3, 1).cpu().numpy() image = np.sum(np.abs(generated-target), axis=-1) image = (image[:, :, :, None]).astype(np.uint8) image = np.clip(image, 0, 255) image = np.repeat(image, 3, axis=-1) elif dtype[i] == "diff_scaled": generated = batch[0][:n_logged].permute(0, 2, 3, 1).cpu().numpy() target = batch[1][:n_logged].permute(0, 2, 3, 1).cpu().numpy() image = np.sum(np.abs(generated-target), axis=-1) image /= image.max(axis=0) image = (image[:, :, :, None]*255.0).astype(np.uint8) image = np.repeat(image, 3, axis=-1) image = np.concatenate([s for s in image], axis=1) final_data.append(image) grid = np.concatenate(final_data, axis=0) return grid def make_img_grid(appearance, shape, pred, tgt= None, n_logged=4, target_label="Target Images", label_app = "Appearance Images", label_gen = "Generated Images", label_shape = "Shape Images"): """ :param appearance: :param shape: :param pred: :param tgt: :param n_logged: :return: """ appearance = ((appearance.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] shape = ((shape.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype(np.uint8)[:n_logged] pred = ((pred.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype( np.uint8)[:n_logged] if tgt is not None: tgt = ((tgt.permute(0, 2, 3, 1).cpu().numpy() + 1.) * 127.5).astype( np.uint8)[:n_logged] tgt = np.concatenate([t for t in tgt], axis=1) tgt = cv2.UMat.get(cv2.putText(cv2.UMat(tgt), target_label , (int(tgt.shape[1] // 3), tgt.shape[0] - int(tgt.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(tgt.shape[0] / 256), (255, 0, 0), int(tgt.shape[0] / 128))) appearance = np.concatenate([s for s in appearance], axis=1) appearance = cv2.UMat.get(cv2.putText(cv2.UMat(appearance),label_app, (int(appearance.shape[1] // 3), appearance.shape[0] - int(appearance.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(appearance.shape[0] / 256), (255, 0, 0), int(appearance.shape[0] / 128))) shape = np.concatenate([f for f in shape], axis=1) shape = cv2.UMat.get(cv2.putText(cv2.UMat(shape), label_shape, (int(shape.shape[1] // 3), shape.shape[0] - int(shape.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(shape.shape[0] / 256), (255, 0, 0), int(shape.shape[0] / 128))) pred = np.concatenate([p for p in pred], axis=1) pred = cv2.UMat.get(cv2.putText(cv2.UMat(pred), label_gen, (int(pred.shape[1] // 3), pred.shape[0] - int(pred.shape[0]/6)), cv2.FONT_HERSHEY_SIMPLEX, float(pred.shape[0] / 256), (255, 0, 0), int(pred.shape[0] / 128))) if tgt is None: grid = np.concatenate([appearance, shape, pred], axis=0) else: grid = np.concatenate([appearance, shape, pred, tgt], axis=0) return grid def scale_img(img): """ Rescales an image to the actual pixel domain in [0,255] and converts it to the dtype :param img: :return: """ img = (img + 1.) * 127.5 if isinstance(img, torch.Tensor): img = img.to(torch.uint8) else: # assumed to be numpy array img = img.astype(np.uint8) return img def human_graph_cut_map(img, poke_size): import cv2 import matplotlib.pyplot as plt # make backgound foreground segementation mask = np.zeros(img.shape[:2], np.uint8) rect = (int(img.shape[1] / 5), poke_size, int(3. * img.shape[1] / 5), int(img.shape[0] - 2 * poke_size)) fgm = np.zeros((1, 65), dtype=np.float64) bgm = np.zeros((1, 65), dtype=np.float64) mask2, fgm, bgm = cv2.grabCut(img, mask, rect, fgm, bgm, 5, cv2.GC_INIT_WITH_RECT) mask3 = np.where((mask2 == 2) | (mask2 == 0), 0, 1).astype(np.bool) tuples = np.where(mask3[:, :]) return tuples new_img = np.zeros_like(img) for t in tuples: for i in range(3): new_img[t[0], t[1], i] = 255 # show the output frame plt.imshow(new_img) plt.show() # human_NN_map_weights = "/export/home/jsieber/poking/models/mask-rcnn-coco/frozen_inference_graph.pb" # human_NN_map_classes = "/export/home/jsieber/poking/models/mask-rcnn-coco/object_detection_classes_coco.txt" # human_NN_map_config = "/export/home/jsieber/poking/models/mask-rcnn-coco/mask_rcnn_inception_v2_coco_2018_01_28.pbtxt" # human_NN_map_LABELS = open(human_NN_map_classes).read().strip().split("\n") # # human_NN_map_net = cv2.dnn.readNetFromTensorflow(human_NN_map_weights, human_NN_map_config) # def human_NN_map(frame, conf=0.5, threshold=0.3): # # # # modified from https://www.pyimagesearch.com/2018/11/26/instance-segmentation-with-opencv/ # before_width = frame.shape[1] # frame = imutils.resize(frame, width=600) # (H, W) = frame.shape[:2] # # # construct a blob from the input image and then perform a # # forward pass of the Mask R-CNN, giving us (1) the bounding # # box coordinates of the objects in the image along with (2) # # the pixel-wise segmentation for each specific object # blob = cv2.dnn.blobFromImage(frame, swapRB=True, crop=False) # human_NN_map_net.setInput(blob) # (boxes, masks) = human_NN_map_net.forward(["detection_out_final", # "detection_masks"]) # # # sort the indexes of the bounding boxes in by their corresponding # # prediction probability (in descending order) # idxs = np.argsort(boxes[0, 0, :, 2])[::-1] # # # initialize the mask, ROI, and coordinates of the person for the # # current frame # mask = None # roi = None # coords = None # # # loop over the indexes # for i in idxs: # # extract the class ID of the detection along with the # # confidence (i.e., probability) associated with the # # prediction # classID = int(boxes[0, 0, i, 1]) # confidence = boxes[0, 0, i, 2] # # # if the detection is not the 'person' class, ignore it # if human_NN_map_LABELS[classID] != "person": # continue # # filter out weak predictions by ensuring the detected # # probability is greater than the minimum probability # if confidence > conf: # # scale the bounding box coordinates back relative to the # # size of the image and then compute the width and the # # height of the bounding box # box = boxes[0, 0, i, 3:7] * np.array([W, H, W, H]) # (startX, startY, endX, endY) = box.astype("int") # coords = (startX, startY, endX, endY) # boxW = endX - startX # boxH = endY - startY # # # extract the pixel-wise segmentation for the object, # # resize the mask such that it's the same dimensions of # # the bounding box, and then finally threshold to create # # a *binary* mask # mask = masks[i, classID] # mask = cv2.resize(mask, (boxW, boxH), # interpolation=cv2.INTER_NEAREST) # mask = (mask > threshold) # # # extract the ROI and break from the loop (since we make # # the assumption there is only *one* person in the frame # # who is also the person with the highest prediction # # confidence) # roi = frame[startY:endY, startX:endX][mask] # break # # # initialize our output frame # output = frame.copy() # # # if the mask is not None *and* we are in privacy mode, then we # # know we can apply the mask and ROI to the output image # if mask is not None: # # blur the output frame # output = np.zeros_like(output) # # # add the ROI to the output frame for only the masked region # (startX, startY, endX, endY) = coords # roi = np.ones_like(roi)*255 # output[startY:endY, startX:endX][mask] = roi # output = imutils.resize(output, width=before_width) # # tuples = np.where(output[:, :, 0] > 0) # return tuples # # new_img = np.zeros_like(output) # for t in tuples: # for i in range(3): # new_img[t[0], t[1], i] = 255 # # show the output frame # plt.imshow(new_img) # plt.show() def make_hist(hist, title, ylabel, xlabel="Frame number", bins_edges = None): plt.ioff() if bins_edges is None: bins_edges = np.arange(0, len(hist) + 1).astype(np.float) else: assert len(list(bins_edges)) == len(list(hist)) + 1 centroids = (bins_edges[1:] + bins_edges[:-1]) / 2 hist_, bins_, _ = plt.hist( centroids, bins=len(hist), weights=np.asarray(hist), range=(min(bins_edges), max(bins_edges)), ) plt.xlabel(xlabel) plt.ylabel(ylabel) wandb.log({title: wandb.Image(plt)}) plt.close() def make_plot(x,y,title,ylabel,xlabel="frame idx", savename=None): plt.ioff() fig, ax = plt.subplots() ax.plot(x,y,'rv-') ax.set(xlabel=xlabel, ylabel=ylabel, title = title) ax.grid() if savename is None: wandb.log({title:wandb.Image(plt)}) else: fig.savefig(savename) plt.close() if __name__=="__main__": frame_path = "/export/data/ablattma/Datasets/iPER/processed/001_10_1/frame_277.png" frame = cv2.imread(frame_path) frame = imutils.resize(frame, width=128) import time for i in range(3): start_time = time.time() human_graph_cut_map(frame, 15) print("--- %s seconds ---" % (time.time() - start_time)) for i in range(3): start_time = time.time() human_NN_map(frame) print("--- %s secondss ---" % (time.time() - start_time))
33,806
43.424442
204
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/fvd_models.py
from utils.general import get_logger import os from os import path import argparse if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument("-b", "--base", type=str, default="/export/scratch/ablattma/visual_poking/fixed_length_model/generated", help="Source directory.") parser.add_argument("--gpu", type=int, required=True, help="The target device.") args = parser.parse_args() with open("config/model_names.txt", "r") as f: model_names = f.readlines() model_names = [m for m in model_names if not m.startswith("#")] logger = get_logger("eval-models") base_path = args.base if "DATAPATH" in os.environ: base_path = os.environ['DATAPATH'] + base_path logger.info(f'Base path is "{base_path}"') gpu = args.gpu for n in model_names: n = n.rstrip() logger.info(f"Compute fvd for model {n}") filepath = path.join(base_path, n, "samples_fvd") if not any(map(lambda x: x.endswith("npy"),os.listdir(filepath))): logger.info("no samples were found...skipping") try: test_cmd = f"python -m utils.metric_fvd --source {filepath} --gpu {gpu}" os.system(test_cmd) except Exception as e: logger.error(e) logger.info("next model") continue logger.info("finished")
1,429
25.481481
102
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/flownet_loader.py
import torch from torch.nn import functional as F from PIL import Image from models.flownet2.models import * from torchvision import transforms import matplotlib.pyplot as plt import argparse from utils.general import get_gpu_id_with_lowest_memory class FlownetPipeline: def __init__(self): super(FlownetPipeline, self).__init__() def load_flownet(self, args, device): """ :param args: args from argparser :return: The flownet pytorch model """ # load model savefile save = torch.load( "/export/scratch/compvis/datasets/plants/pretrained_models/FlowNet2_checkpoint.pth.tar") model = FlowNet2(args, batchNorm=False) untrained_statedict = model.state_dict() # load it into proper clean model model.load_state_dict(save["state_dict"]) model.eval() return model.to(device) def preprocess_image(self, img, img2, channelOrder="RGB",spatial_size= None): """ This preprocesses the images for FlowNet input. Preserves the height and width order! :param channelOrder: RGB(A) or BGR :param img: The first image in form of (W x H x RGBA) or (H x W x RGBA) :param img2: The first image in form of (W x H x RGBA) or (H x W x RGBA) :return: The preprocessed input for the prediction (BGR x Img# x W x H) or (BGR x Img# x H x W) """ # ToTensor transforms from (H x W x C) => (C x H x W) # also automatically casts into range [0, 1] if spatial_size is None: img, img2 = transforms.ToTensor()(img)[:3], transforms.ToTensor()(img2)[:3] else: ts = transforms.Compose([transforms.ToPILImage(), transforms.Resize(size=spatial_size,interpolation=Image.BILINEAR),transforms.ToTensor()]) img, img2 = ts(img)[:3],ts(img2)[:3] if channelOrder == "RGB": img, img2 = img[[2, 1, 0]], img2[[2, 1, 0]] # Cast to proper shape (Batch x BGR x #Img x H x W) s = img.shape img, img2 = img[:, :int(s[1] / 64) * 64, :int(s[2] / 64) * 64], \ img2[:, :int(s[1] / 64) * 64,:int(s[2] / 64) * 64] stacked = torch.cat([img[:, None], img2[:, None]], dim=1) return stacked def predict(self, model, stacked, spatial_size=None): """ :param stacked: The two input images. (Batch x BGR x Img# x H x W) :return: The flow result (2 x W x H) """ # predict model.eval() prediction = model(stacked) out_size = float(prediction.shape[-1]) if spatial_size is not None: prediction = F.interpolate( prediction.cpu(), size=(spatial_size,spatial_size), mode="bilinear" ) # rescale to make it fit to new shape (not grave, if this is skipped as flow is normalized anyways later) prediction = prediction / (out_size / spatial_size) flow = prediction[0] return flow def show_results(self, prediction, with_ampl=False): """ prediction (Tensor): The predicted flow (2 x W x H) :return: plots """ zeros = torch.zeros((1, prediction.shape[1], prediction.shape[2])) if with_ampl: ampl = torch.sum(prediction * prediction, dim=0) ampl = ampl.squeeze() else: ampl = torch.cat([prediction, zeros], dim=0) ampl -= ampl.min() ampl /= ampl.max() # show image im = transforms.ToPILImage()(ampl) if with_ampl: plt.imshow(im, cmap='gray') else: plt.imshow(im) if __name__ == "__main__": # parse args parser = argparse.ArgumentParser(description='Process some integers.') # always 1.0, because pytorch toTensor automatically converts into range [0.0, 1.0] parser.add_argument("--rgb_max", type=float, default=1.) parser.add_argument('--fp16', action='store_true', help='Run model in pseudo-fp16 mode (fp16 storage fp32 math).') parser.add_argument('--fp16_scale', type=float, default=1024., help='Loss scaling, positive power of 2 values can improve fp16 convergence.') args = parser.parse_args() # load test images in BGR mode img, img2 = np.asarray(Image.open(f"/export/data/ablattma/Datasets/plants/processed/hoch_misc1/frame_0.png")), \ np.asarray(Image.open(f"/export/data/ablattma/Datasets/plants/processed/hoch_misc1/frame_100.png")) # load Flownet pipeline = FlownetPipeline() flownet_device = get_gpu_id_with_lowest_memory() flownet = pipeline.load_flownet(args, flownet_device) # process to show flow stacked = pipeline.preprocess_image(img, img2).to(flownet_device) prediction = pipeline.predict(flownet, stacked[None]).cpu() pipeline.show_results(prediction) plt.show()
4,882
37.753968
151
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/metrics.py
import torch from torch import nn from torch.nn import functional as F from torchvision.models import inception_v3 import numpy as np from scipy import linalg from skimage.metrics import peak_signal_noise_ratio as compare_psnr from skimage.metrics import structural_similarity as ssim from pytorch_lightning.metrics import functional as PF from tqdm import tqdm class FIDInceptionModel(nn.Module): def __init__(self, normalize_range=True): super().__init__() self.v3 = inception_v3(pretrained=True,aux_logits=False) # self.v3.aux_logits = False self.register_buffer( "mean", torch.tensor([0.485, 0.456, 0.406], dtype=torch.float) .unsqueeze(dim=0) .unsqueeze(dim=-1) .unsqueeze(dim=-1), ) self.register_buffer( "std", torch.tensor([0.229, 0.224, 0.225], dtype=torch.float) .unsqueeze(dim=0) .unsqueeze(dim=-1) .unsqueeze(dim=-1), ) self.resize = nn.Upsample(size=(299,299),mode="bilinear") self.normalize_range = normalize_range def forward(self, x): x = self.resize(x) if self.normalize_range: # normalize in between 0 and 1 x = (x + 1.) / 2. else: x = x.to(torch.float) / 255. # normalize to demanded values x = (x - self.mean) / self.std # this reimpleents the respective layers of the inception model, see model definition for name, submodule in self.v3._modules.items(): x = submodule(x) if name == "Mixed_7c": break elif name == "Conv2d_4a_3x3" or name == "Conv2d_2b_3x3": x = F.avg_pool2d(x, kernel_size=3, stride=2) out = F.adaptive_avg_pool2d(x, (1, 1)) out = torch.flatten(out, 1) return out def metrcis_MAE(tensor1, tensor2, mean=True): value = torch.sum(torch.abs(tensor1 - tensor2)) if mean: value /= tensor1.view((-1)).shape[0] return value.data.cpu().numpy() def metrcis_MSE(tensor1, tensor2, mean=True): diff = tensor1 - tensor2 value = torch.sum(diff * diff) if mean: value /= tensor1.view((-1)).shape[0] return value.data.cpu().numpy() def metrcis_l1(tensor1, tensor2, mean=False): return metrcis_MAE(tensor1, tensor2, mean) def metrcis_l2(tensor1, tensor2, mean=False): diff = tensor1 - tensor2 value = torch.sum(diff * diff) value = torch.sqrt(value) if mean: value /= tensor1.view((-1)).shape[0] return value.data.cpu().numpy() def metric_ssim(real, fake, reduce = True, return_per_frame=False): if real.dim() == 3: real = real[None,None] fake = fake[None,None] elif real.dim() == 4: real = real[None] fake = fake[None] # rescale to valid range real = ((real + 1.) / 2.).permute(0, 1, 3, 4, 2).cpu().numpy() fake = ((fake + 1.) / 2.).permute(0, 1, 3, 4, 2).cpu().numpy() ssim_batch = np.asarray([ssim(rimg, fimg, multichannel=True, data_range=1.0, gaussian_weights=True,use_sample_covariance=False, ) for rimg, fimg in zip(real.reshape(-1,*real.shape[2:]), fake.reshape(-1,*fake.shape[2:]))]) if return_per_frame: ssim_per_frame = {} for i in range(real.shape[1]): real_test = real[:,i] fake_test = fake[:,i] ssim_per_frame[i] = np.asarray([ssim(real_test, fake_test, multichannel=True, data_range=1., gaussian_weights=True, use_sample_covariance=False)]) # ssim_per_frame = {i:np.asarray([ssim(real[:,i], fake[:,i], # multichannel=True, data_range=1., gaussian_weights=True, use_sample_covariance=False)]) for i in range(real.shape[1])} if reduce: if return_per_frame: ssim_pf_reduced = {key: ssim_per_frame[key] for key in ssim_per_frame} return np.mean(ssim_batch), ssim_pf_reduced else: return np.mean(ssim_batch) if return_per_frame: return ssim_batch, ssim_per_frame else: return ssim_batch def ssim_lightning(real, fake, return_per_frame=False, normalize_range=True): if real.dim() == 3: real = real[None, None] fake = fake[None, None] elif real.dim() == 4: real = real[None] fake = fake[None] if normalize_range: real = (real + 1.) /2. fake = (fake + 1.) / 2. ssim_batch = PF.ssim(fake.reshape(-1,*fake.shape[2:]),real.reshape(-1,*real.shape[2:])).cpu().numpy() if return_per_frame: ssim_per_frame = {i: PF.ssim(fake[:,i],real[:,i]).cpu().numpy() for i in range(real.shape[1])} return ssim_batch, ssim_per_frame return ssim_batch def psnr_lightning(real, fake, return_per_frame=False, normalize_range=True): if real.dim() == 3: real = real[None, None] fake = fake[None, None] elif real.dim() == 4: real = real[None] fake = fake[None] if normalize_range: real = (real + 1.) / 2. fake = (fake + 1.) / 2. psnr_batch = PF.psnr(fake.reshape(-1, *fake.shape[2:]), real.reshape(-1, *real.shape[2:])).cpu().numpy() if return_per_frame: psnr_per_frame = {i: PF.psnr(fake[:, i].contiguous(), real[:, i].contiguous()).cpu().numpy() for i in range(real.shape[1])} return psnr_batch, psnr_per_frame return psnr_batch def metric_psnr(im_true, im_test,reduce = True, return_per_frame=False): if im_true.dim() == 3: im_true, im_test = im_true[None,None], im_test[None,None] elif im_true.dim() == 4: im_true, im_test = im_true[None], im_test[None] # make channel last real = ((im_true + 1.) / 2.).permute(0, 1, 3, 4, 2).cpu().numpy() fake = ((im_test + 1.) / 2.).permute(0, 1, 3, 4, 2).cpu().numpy() psnr_batch = np.asarray([compare_psnr(r,f, data_range=1.) for r, f in zip(real.reshape(-1,*real.shape[2:]),fake.reshape(-1,*fake.shape[2:]))]) if return_per_frame: psnr_per_frame = {i: np.asarray([compare_psnr(real[:,i], fake[:,i], data_range=1.)]) for i in range(real.shape[1])} if reduce: if return_per_frame: psnr_pf_reduced = {key: psnr_per_frame[key] for key in psnr_per_frame} return np.mean(psnr_batch), psnr_pf_reduced else: return np.mean(psnr_batch) if return_per_frame: return psnr_batch, psnr_per_frame else: return psnr_batch def metric_lpips(real, fake, lpips_func, reduce=True, return_per_frame=False, normalize=False): if real.dim() == 3: real, fake = real[None,None], fake[None,None] elif real.dim() == 4: real, fake = real[None], fake[None] if normalize: if fake.max() > 1: fake = (fake.to(torch.float) / 127.5) - 1. real = (real.to(torch.float) / 127.5) -1. else: real = (real * 2.) - 1. fake = (fake * 2.) - 1. lpips_batch = lpips_func(real.reshape(-1,*real.shape[2:]),fake.reshape(-1,*fake.shape[2:])).squeeze().cpu().numpy() if return_per_frame: lpips_per_frame = {i: lpips_func(real[:,i],fake[:,i]).squeeze().cpu().numpy() for i in range(real.shape[1])} if reduce: if return_per_frame: lpips_pf_reduced = {key: lpips_per_frame[key].mean() for key in lpips_per_frame} return lpips_batch.mean(), lpips_pf_reduced else: return lpips_batch.mean() if return_per_frame: return lpips_batch, lpips_per_frame else: return lpips_batch def mean_cov(features): mu = np.mean(features, axis=0) cov = np.cov(features, rowvar=False) return mu,cov def metric_fid(real_features, fake_features, eps=1e-6): # Taken and adapted from https://github.com/mseitzer/pytorch-fid/blob/master/fid_score.py if not isinstance(real_features,np.ndarray): real_features = np.concatenate(real_features,axis=0) fake_features = np.concatenate(fake_features,axis=0) mu1, cov1 = mean_cov(real_features) mu2, cov2 = mean_cov(fake_features) mu1 = np.atleast_1d(mu1) mu2 = np.atleast_1d(mu2) sigma1 = np.atleast_2d(cov1) sigma2 = np.atleast_2d(cov2) assert ( mu1.shape == mu2.shape ), "Training and test mean vectors have different lengths" assert ( sigma1.shape == sigma2.shape ), "Training and test covariances have different dimensions" diff = mu1 - mu2 # Product might be almost singular covmean, _ = linalg.sqrtm(sigma1.dot(sigma2), disp=False) if not np.isfinite(covmean).all(): msg = f"fid calculation produces singular product; adding {eps} to diagonal of cov estimates" print(msg) offset = np.eye(sigma1.shape[0]) * eps covmean = linalg.sqrtm((sigma1 + offset).dot(sigma2 + offset)) # Numerical error might give slight imaginary component if np.iscomplexobj(covmean): if not np.allclose(np.diagonal(covmean).imag, 0, atol=1e-3): m = np.max(np.abs(covmean.imag)) raise ValueError(f"Imaginary component {m}") covmean = covmean.real tr_covmean = np.trace(covmean) return diff.dot(diff) + np.trace(sigma1) + np.trace(sigma2) - 2 * tr_covmean def compute_fvd(real_videos,fake_videos, device,logger): import silence_tensorflow.auto import tensorflow.compat.v1 as tf from utils.frechet_video_distance import preprocess,Embedder,calculate_fvd # required for fvd computation # config = tf.ConfigProto() # config.gpu_options.visible_device_list = f"{device}" devs = tf.config.experimental.get_visible_devices("GPU") target_dev = [d for d in devs if d.name.endswith(str(device))][0] tf.config.experimental.set_visible_devices(target_dev, 'GPU') logger.info("Compute fvd score.") #dev = f"/gpu:{device}" logger.info(f"using device {device}") with tf.device("/gpu:0"): with tf.Graph().as_default(): # construct graph sess = tf.Session() input_shape = real_videos[0].shape input_real = tf.placeholder(dtype=tf.uint8, shape=input_shape) input_fake = tf.placeholder(dtype=tf.uint8, shape=input_shape) real_pre = preprocess(input_real, (224, 224)) emb_real = Embedder(real_pre) embed_real = emb_real.create_id3_embedding(real_pre) fake_pre = preprocess(input_fake, (224, 224)) emb_fake = Embedder(fake_pre) embed_fake = emb_fake.create_id3_embedding(fake_pre) sess.run(tf.global_variables_initializer()) sess.run(tf.tables_initializer()) real, fake = [], [] for rv, fv in tqdm(zip(real_videos, fake_videos)): # real_batch = ((rv + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy() # fake_batch = ((fv + 1.) * 127.5).permute(0, 1, 3, 4, 2).cpu().numpy() # real_batch = ((rv + 1.) * 127.5).cpu().numpy() # fake_batch = ((fv + 1.) * 127.5).cpu().numpy() feed_dict = {input_real: rv, input_fake: fv} r, f = sess.run([embed_fake, embed_real], feed_dict) real.append(r) fake.append(f) print('Compute FVD score') real = np.concatenate(real, axis=0) fake = np.concatenate(fake, axis=0) embed_real = tf.placeholder(dtype=tf.float32, shape=(real.shape[0], 400)) embed_fake = tf.placeholder(dtype=tf.float32, shape=(real.shape[0], 400)) result = calculate_fvd(embed_real, embed_fake) feed_dict = {embed_real: real, embed_fake: fake} fvd_val = sess.run(result, feed_dict) sess.close() logger.info(f"Results of fvd computation: fvd={fvd_val}") # for being sure return fvd_val if __name__ == "__main__": z, o = torch.rand((1080, 720, 3)), torch.rand((1080, 720, 3)) o[0, 0, 0], o[1, 0, 0] = 0, 1 z[0, 0, 0], z[1, 0, 0] = 0, 1
12,279
33.985755
153
py
interactive-image2video-synthesis
interactive-image2video-synthesis-main/utils/eval_pretrained.py
import argparse from os import path import yaml import os from experiments import select_experiment def create_dir_structure(model_name, base_dir): subdirs = ["ckpt", "config", "generated", "log"] structure = {subdir: path.join(base_dir,model_name, subdir) for subdir in subdirs} [os.makedirs(structure[s],exist_ok=True) for s in structure] if "DATAPATH" in os.environ: structure = {subdir: os.environ["DATAPATH"] +structure[subdir] for subdir in structure} return structure def load_parameters(model_name, base_dir): dir_structure = create_dir_structure(model_name, base_dir) saved_config = path.join(dir_structure["config"], "config.yaml") if path.isfile(saved_config): with open(saved_config, "r") as f: cdict = yaml.load(f, Loader=yaml.FullLoader) else: raise FileNotFoundError("No saved config file found but model is intended to be restarted. Aborting....") return cdict, dir_structure, if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('-b', '--base_dir', required=True, type=str, help='the base directory, where all logs, configs, checkpoints and evaluation results will be stored.') parser.add_argument("--gpu", type=int, required=True, help="The target device.") parser.add_argument("--mode", default="metrics", type=str, choices=["metrics", "fvd"], help="The mode in which the test-method should be executed.") parser.add_argument("--metrics_on_patches", default=False, action="store_true", help="Whether to run evaluation on patches (if available or not).") parser.add_argument("--best_ckpt", default=False, action="store_true", help="Whether to use the best ckpt as measured by LPIPS (otherwise, latest_ckpt is used)") args = parser.parse_args() with open("config/model_names.txt", "r") as f: model_names = f.readlines() model_names = [m for m in model_names if not m.startswith("#")] gpu = args.gpu for model in model_names: model = model.rstrip() print(f"Evaluate model : {model}") cdict, dirs = load_parameters(model, args.base_dir) cdict["testing"].update({"mode":args.mode}) cdict["general"]["mode"] = "test" cdict["testing"].update({"best_ckpt": args.best_ckpt}) cdict["testing"]["metrics_on_patches"] = args.metrics_on_patches cdict["general"]["restart"] = True experiment = select_experiment(cdict, dirs, args.gpu) try: experiment.test() except FileNotFoundError as e: print(e)
2,683
38.470588
137
py