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
SLIT
SLIT-master/SLIT/Lens.py
import numpy as np import matplotlib.pyplot as plt import pyfits as pf import scipy.signal as scp import warnings warnings.simplefilter("ignore") #Tool box for lensing def SIS(x0,y0,n1,n2,Re): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) count = 0 for i in x: kappa[x[count],y[count]] = Re/(2*np.sqrt((x[count]-x0)**2+(y[count]-y0)**2)) count += 1 if np.isfinite(kappa[x0,y0]) == False: kappa[x0,y0] = 1 return kappa def SIE_xy(x,y,x0,y0,b,beta,q,xc,theta): eps = (1-q**2)/(1+q**2) up = b**(beta-1) pre = up/(2*(1-eps)**((beta-1)/2)) count = 0 theta = theta*np.pi/180. Xr = (x-x0)*np.cos(theta)-(y-y0)*np.sin(theta) Yr = (x-x0)*np.sin(theta)+(y-y0)*np.cos(theta) kappa = pre/((xc**2.)/(1.-eps)+(Xr)**2.+((Yr)**2.)/q**2.)**((beta-1.)/2.) return kappa def SIE(x0,y0,n1,n2,b,beta,q,xc,theta): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) x2d = np.reshape(x, (n1,n2)) y2d = np.reshape(y, (n1,n2)) kappa = SIE_xy(x2d,y2d,x0,y0,b,beta,q,xc,theta) return kappa def alpha_def(kappa, n1,n2,extra=0): #Computes the deflection angle of a single photon at coordinates theta in the source plane and a lens #mass distribution kappa nk1,nk2 = np.shape(kappa) #Coordonnees de la grille de l'espace image [x,y] = np.where(np.zeros([nk1,nk2])==0) x0 = nk1/2 y0 = nk2/2 xc = np.reshape((x)-x0,(nk1,nk2)) yc = np.reshape((y)-y0,(nk1,nk2)) r = (xc**2+yc**2) lx,ly = np.where(r==0) tabx = np.reshape(np.float_(xc)/(r),(nk1,nk2)) taby = np.reshape(np.float_(yc)/(r),(nk1,nk2)) tabx[lx,ly]=0 taby[lx,ly]=0 kappa = kappa.astype(float) tabx = tabx.astype(float) # kappa[rk>(nk1)/2.] = 0 intex = scp.fftconvolve(tabx, (kappa), mode = 'same')/np.pi intey = scp.fftconvolve(taby, (kappa), mode = 'same')/np.pi return intex[x0-(n1)/2:x0+(n1)/2,y0-(n2)/2:y0+(n2)/2], intey[x0-(n1)/2:x0+(n1)/2,y0-(n2)/2:y0+(n2)/2] def beta(kappa,theta): #Computes beta beta = theta-alpha(theta,kappa) return beta def theta(alpha, beta): #Computes beta theta = beta+alpha return beta def F(kappa, nt1,nt2, size, extra=100, x_shear = 0, y_shear = 0, alpha_x_in = [-99], alpha_y_in = [-99]): # Theta positions for each pixel in beta if np.sum(alpha_x_in) != [-99]: alpha_x = alpha_x_in alpha_y = alpha_y_in else: nk1,nk2 = np.shape(kappa) alpha_x,alpha_y = alpha_def(kappa,nt1,nt2,extra = extra) alpha_x = alpha_x+x_shear alpha_y = alpha_y+y_shear na1,na2 = np.shape(alpha_x) xa,ya = np.where(np.zeros((na1,na2)) == 0) nb1=nt1*size nb2=nt2*size xb, yb = np.where(np.zeros((nb1,nb2))==0) #Scaling of the source grid #Scaling of the deflection grid xa = xa*(np.float(nt1)/np.float(na1))#-0.68 ya = ya*(np.float(nt2)/np.float(na2))#-0.68 #Setting images coordinates in 2d xa2d = np.reshape(xa,(na1,na2)) ya2d = np.reshape(ya,(na1,na2)) F2 = [] for i in range(np.size(xb)): #Deflection of photons emitted in xb[i],yb[i] theta_x = (xb[i])*(np.float(nt1)/np.float(nb1))+alpha_x theta_y = (yb[i])*(np.float(nt2)/np.float(nb2))+alpha_y #Matching of arrivals with pixels in image plane xprox = np.int_(np.abs((xa2d-theta_x)*2)) yprox = np.int_(np.abs((ya2d-theta_y)*2)) if np.min(xprox+yprox) <1: loc2 = np.array(np.where((xprox+yprox)==np.min(xprox+yprox)))*np.float(nt1)/np.float(na1)# else: loc2 = [] if (np.size(loc2)==0): F2.append([0]) else: F2.append(np.int_(loc2)) return F2 def source_to_image(Source, nt1,nt2, theta, ones = 1): # Source: Image of the source in the source plane # n1,n2: size in pixels of the postage stamp in image plane # F: the coordinates of the lens mapping F = (theta) nb1,nb2 = np.shape(Source) if ones == 1: onelens = source_to_image(np.ones(Source.shape), nt1,nt2, theta, ones = 0) onelens[np.where(onelens==0)]=1 else: onelens = 1. Image = np.zeros((nt1,nt2)) xb,yb = np.where(np.zeros((nb1,nb2)) == 0) N = np.size(xb) k=0 for pos in F: if np.size(np.shape(pos)) != 1: Image[np.array(pos[0][:]), np.array(pos[1][:])] += Source[xb[k],yb[k]]#fullSource k=k+1 return Image/onelens def image_to_source(Image, size,beta,lensed = 0, square = 0): # Image: postagestamp of the observed image # nsize1,nsize2: size of the postagestamp in source plane # F: lens mapping matrix F = (beta) nt1,nt2 = np.shape(Image) nb1 = nt1*size nb2 = nt2*size Source = np.zeros((nb1,nb2)) xb,yb = np.where(Source == 0) N = np.size(xb) for k in range(N): pos = F[k] if np.size(np.shape(pos)) > 1: if np.sum(lensed) !=0: if square == 0: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])])/np.max([1,np.size(pos[0][:])]) else: Source[xb[k], yb[k]] += np.sum((Image[np.array(pos[0][:]), np.array(pos[1][:])] / np.max([1, np.size(pos[0][:])]))**2) else: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])]) if square == 1: Source = np.sqrt(Source) return Source def image_to_source_bound(Image, size,beta,lensed = 0): # Image: postagestamp of the observed image # nsize1,nsize2: size of the postagestamp in source plane # F: lens mapping matrix F = (beta) nt1,nt2 = np.shape(Image) nb1 = nt1*size nb2 = nt2*size Source = np.zeros((nb1,nb2)) xb,yb = np.where(Source == 0) N = np.size(xb) for k in range(N): pos = F[k] if np.size(np.shape(pos)) > 1: if np.sum(lensed) !=0: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])])/np.max([1,np.size(pos[0][:])]) else: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])]) return Source
6,888
28.440171
123
py
SLIT
SLIT-master/SLIT/__init__.py
from Solve import * import Lens import wave_transform import tools
67
12.6
21
py
SLIT
SLIT-master/Examples/Test_SLIT_forUsers.py
import pyfits as pf import matplotlib.pyplot as plt import numpy as np import matplotlib.cm as cm from scipy import signal as scp import SLIT import time from scipy import signal as scp import warnings warnings.simplefilter("ignore") #Example of a run of the SLIT algorithm on simulated images. #Here the first part of the file shows how simulations are generated. #For users intereseted only in seeing the code run, have a look at the running SLIT section. #The command line that builds the Fkappa operator is also of outmost importance. Image = '''Input 2D image of the lens to invert ''' nt1,nt2 = np.shape(Image) ###############################Mass profile############################### x0,y0 = '''Input the center of mass of the lens with regard to coodinates in Image ''' kappa = '''Input dimensionless pixelated mass density profile here ''' size = '''Input the desired size of the output with regard to Image. Chosing 1 will result in a source with the same number of pixels as in Image. ''' #Mapping between lens and source IMPORTANT Fkappa = SLIT.Lens.F(kappa, nt1,nt2, size,x0,y0) PSF = '''Input the PSF for Image here''' PSFconj = '''Input the conjugate of the PSF here given by np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))), but be carefull that the result is still centered ''' ################################Running SLIT############################ #Parameters kmax = 5 niter =50 #Start clock start = time.clock() #Running SLIT S, FS = SLIT.SLIT(Image, Fkappa, kmax, niter, size, PSF, PSFconj) #Stop clock elapsed = (time.clock()-start) print('execution time:', elapsed, 'seconds') #Reconstruction goodness real_source = newsource source_error = np.sum(np.abs(real_source[np.where(real_source!=0)] -S[np.where(real_source!=0)])**2 /real_source[np.where(real_source!=0)]**2)/(np.size( np.where(real_source!=0))/2.) image_chi2 = np.std(Image-FS)**2/sigma**2 print('Residuals in source space', source_error) print('Residuals in image space',image_chi2) #Display of results for i in [1]: plt.figure(2) # plt.suptitle('FISTA: error per pixel on the source: '+str(source_error)+' image chi2:'+str(image_chi2)) # plt.subplot(2,3,1) plt.title('Source from SLIT') plt.imshow((S), vmin = np.min(real_source), vmax = np.max(real_source),cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,2) plt.figure(3) plt.title('Original source') plt.imshow(real_source, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,3) plt.figure(4) plt.title('Lensed source') plt.imshow(Image, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(41) plt.title('Reconstructed lensed source') plt.imshow(FS, vmin = np.min(Image), vmax = np.max(Image), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,4) plt.figure(5) plt.title('relative difference') diff = (real_source-S)/real_source diff[np.where(real_source==0)] = 0 diff[np.where(diff>1)]= np.log(0.) plt.imshow(np.abs(diff), vmax = 1., vmin = 0., cmap = cm.gist_stern, interpolation = 'nearest' ) plt.axis('off') plt.colorbar() # plt.subplot(2,3,5) plt.figure(6) plt.title('difference with reconstructed lensed image') plt.imshow(Image-FS, vmin = -5*sigma, vmax = 5*sigma, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,6) plt.figure(7) plt.title('difference with true source') plt.imshow((np.abs(real_source-S)), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.show()
3,928
37.145631
150
py
SLIT
SLIT-master/Examples/Test_SLIT_MCA.py
from SLIT import Lens import pyfits as pf import matplotlib.pyplot as plt import numpy as np import matplotlib.cm as cm from scipy import signal as scp import SLIT as slit import time from scipy import signal as scp import warnings warnings.simplefilter("ignore") #Example of a run of the SLIT_MCA algorithm on simulated images. #Here the first part of the file shows how simulations are generated. #For users intereseted only in seeing the code run, have a look at the running SLIT_MCA section. #The command line that builds the Fkappa operator is also of outmost importance. ###############################Simulation############################### def SIE(x0,y0,n1,n2,b,beta,q,xc,theta): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) eps = (1-q**2)/(1+q**2) up = b**(beta-1) pre = up/(2*(1-eps)**((beta-1)/2)) count = 0 theta = theta*np.pi/180. for i in x: Xr = (x[count]-x0)*np.cos(theta)-(y[count]-y0)*np.sin(theta) Yr = (x[count]-x0)*np.sin(theta)+(y[count]-y0)*np.cos(theta) kappa[x[count],y[count]] = pre/((xc**2.)/(1.-eps)+(Xr)**2.+((Yr)**2.)/q**2.)**((beta-1.)/2.) count += 1 return kappa #Source light profile newsource = pf.open('../Files/source.fits')[0].data ##N1,N2 are the numbers of pixels in the image plane. nt1= 100 nt2 = 100 #Size ratio of the source to image number of pixels size = 1 #PSF PSF0 = pf.open('../Files/PSF.fits')[0].data PSF = PSF0[1:,1:] PSFconj = np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))) PSFconj=PSFconj/np.sum(PSFconj) PSF = PSF/np.sum(PSF) ## Lens mass distribution. b = 1.53/0.05 xc = 0.95 q = 0.71 betata = 2.1 thetata = 25.2 kappa = SIE(nt1/2.+50,nt2/2.+50,nt1+100,nt2+100,b,betata,q,xc, thetata) #Mapping between lens and source IMPORTANT Fkappa = slit.Lens.F(kappa, nt1,nt2, size,nt1/2.,nt2/2.) #Lens galaxy light profile gal0 = pf.open('../Files/Galaxy.fits')[0].data #Generation of lensed source I2 = slit.Lens.source_to_image(newsource, nt1 ,nt2 , Fkappa) HI2 = scp.fftconvolve(I2, PSF.astype(float), mode = 'same') #Noise levels SNR = 500 sigma = np.sqrt(np.sum(I2**2)/SNR/(nt1*nt2*size**2)) #Convolution of the observed image simu = scp.fftconvolve(gal0.astype(float)+I2, PSF.astype(float), mode = 'same') #Sotring the convolved lens light profile: gal = scp.fftconvolve(gal0.astype(float), PSF.astype(float), mode = 'same') #Final simulated image Image = simu+np.random.randn(nt1,nt2)*sigma lensed = slit.lens_one(Fkappa, nt1,nt2, size) Test = slit.Lens.image_to_source(I2+gal, size, Fkappa, lensed = lensed) plt.imshow(Image, cmap = cm.gist_stern, interpolation = 'nearest') plt.show() plt.imshow(Test, cmap = cm.gist_stern, interpolation = 'nearest') plt.show() hdus = pf.PrimaryHDU(Image) lists = pf.HDUList([hdus]) lists.writeto('Image.fits', clobber=True) ################################Running SLIT############################ #Parameters kmax = 5 niter =100 riter =100 levels = [0] #Comment the following to have the level estimation routine run (takes more time) levels = pf.open('../Files/Noise_levels_SLIT_MCA.fits')[0].data #Start clock start = time.clock() #Running SLIT_MCA S, FS, G = slit.SLIT_MCA(Image, Fkappa, kmax, niter,riter, size,PSF, PSFconj, levels = levels) #Stop clock elapsed = (time.clock()-start) print('execution time:', elapsed, 'seconds') real_source = newsource source_error = np.sum(np.abs(real_source[np.where(real_source!=0)] -S[np.where(real_source!=0)])**2 /real_source[np.where(real_source!=0)]**2)/(np.size( np.where(real_source!=0))/2.) image_chi2 = np.std(Image-FS-G)**2/sigma**2 print('Residuals in source space', source_error) print('Residuals in image space',image_chi2) for i in [1]: ###Source plt.figure(0) plt.title('Source from SLIT') plt.imshow((S), vmin = np.min(real_source), vmax = np.max(real_source), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(1) plt.title('Original image of the source') plt.imshow(real_source, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(2) plt.title('relative difference') diff = (real_source-S) plt.imshow((np.abs(diff)), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() ####Lensed source plt.figure(3) plt.title('Original lensed galaxy') plt.imshow(HI2, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(4) plt.title('reconstructed lensed source') plt.imshow((FS), vmin = np.min(I2), vmax = np.max(I2), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(5) plt.title('error on the source in image plane') plt.imshow((HI2-FS), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() ###Galaxy plt.figure(6) plt.title('Original galaxy') plt.imshow((gal0), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(12) plt.title('Estimated Galaxy') plt.imshow((G), vmin = np.min(gal0), vmax = np.max(gal0), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(7) plt.title('Error on the galaxy') plt.imshow((gal-G), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() ###Image plt.figure(8) plt.title('Image') plt.imshow(Image, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(9) plt.title('Reconstructed image') plt.imshow(FS+G, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(10) plt.title('difference with reconstructed image') plt.imshow(Image-FS-G,cmap = cm.gist_stern, interpolation = 'nearest', vmin = -5*sigma, vmax = 5*sigma)#slit.fft_convolve(Im,PSF) plt.axis('off') plt.colorbar() plt.show()
6,169
29.544554
133
py
SLIT
SLIT-master/Examples/Test_SLIT.py
import pyfits as pf import matplotlib.pyplot as plt import numpy as np import matplotlib.cm as cm import SLIT import time from scipy import signal as scp import warnings warnings.simplefilter("ignore") #Example of a run of the SLIT algorithm on simulated images. #Here the first part of the file shows how simulations are generated. #For users intereseted only in seeing the code run, have a look at the running SLIT section. #The command line that builds the Fkappa operator is also of outmost importance. ###############################Simulation############################### def SIE(x0,y0,n1,n2,b,beta,q,xc,theta): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) eps = (1-q**2)/(1+q**2) up = b**(beta-1) pre = up/(2*(1-eps)**((beta-1)/2)) count = 0 theta = theta*np.pi/180. for i in x: Xr = (x[count]-x0)*np.cos(theta)-(y[count]-y0)*np.sin(theta) Yr = (x[count]-x0)*np.sin(theta)+(y[count]-y0)*np.cos(theta) kappa[x[count],y[count]] = pre/((xc**2.)/(1.-eps)+(Xr)**2.+((Yr)**2.)/q**2.)**((beta-1.)/2.) count += 1 return kappa #Source light profile newsource = pf.open('../Files/source.fits')[0].data ##N1,N2 are the numbers of pixels in the image plane. nt1= 100 nt2 = 100 #Size ratio of the source to image number of pixels size = 1 #PSF PSF0 = pf.open('../Files/PSF.fits')[0].data PSF = PSF0[1:,1:] PSFconj = np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))) PSFconj=PSFconj/np.sum(PSFconj) PSF = PSF/np.sum(PSF) ## Lens mass distribution. b = 1.53/0.05 xc = 0.95 q = 0.71 betata = 2.1 thetata = 25.2 kappa = SIE(nt1/2.+50,nt2/2.+50,nt1+100,nt2+100,b,betata,q,xc, thetata) #Mapping between lens and source IMPORTANT Fkappa = SLIT.Lens.F(kappa, nt1,nt2, size,nt1/2.,nt2/2.) #Generation of lensed source I2 = SLIT.Lens.source_to_image(newsource, nt1 ,nt2 , Fkappa) #Noise levels SNR = 500 sigma = np.sqrt(np.sum(I2**2)/SNR/(nt1*nt2*size**2)) #Convolution by the PSF and generation of the final image I2 = scp.fftconvolve(I2, PSF, mode = 'same') #Final simulated image Image = I2+np.random.randn(nt1,nt2)*sigma ################################Running SLIT############################ #Parameters kmax = 5 niter =100 levels = [0] #Comment the following to have the level estimation routine run (takes more time) levels = pf.open('../Files/Noise_levels_SLIT.fits')[0].data #Start clock start = time.clock() #Running SLIT sourcesl, Imsl = SLIT.SLIT(Image, Fkappa, kmax, niter, size, PSF, PSFconj, levels = levels, fb =1) #Stop clock elapsed = (time.clock()-start) print('execution time:', elapsed, 'seconds') #Reconstruction goodness real_source = newsource source_error = np.sum(np.abs(real_source[np.where(real_source!=0)] -sourcesl[np.where(real_source!=0)])**2 /real_source[np.where(real_source!=0)]**2)/(np.size( np.where(real_source!=0))/2.) image_chi2 = np.std(Image-Imsl)**2/sigma**2 print('Residuals in source space', source_error) print('Residuals in image space',image_chi2) #Display of results for i in [1]: plt.figure(2) # plt.suptitle('FISTA: error per pixel on the source: '+str(source_error)+' image chi2:'+str(image_chi2)) # plt.subplot(2,3,1) plt.title('Source from SLIT') plt.imshow((sourcesl), vmin = np.min(real_source), vmax = np.max(real_source),cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,2) plt.figure(3) plt.title('Original source') plt.imshow(real_source, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,3) plt.figure(4) plt.title('Lensed source') plt.imshow(Image, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(41) plt.title('Reconstructed lensed source') plt.imshow(Imsl, vmin = np.min(Image), vmax = np.max(Image), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,4) plt.figure(5) plt.title('relative difference') diff = (real_source-sourcesl)/real_source diff[np.where(real_source==0)] = 0 diff[np.where(diff>1)]= np.log(0.) plt.imshow(np.abs(diff), vmax = 1., vmin = 0., cmap = cm.gist_stern, interpolation = 'nearest' ) plt.axis('off') plt.colorbar() # plt.subplot(2,3,5) plt.figure(6) plt.title('difference with reconstructed lensed image') plt.imshow(Image-Imsl, vmin = -5*sigma, vmax = 5*sigma, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,6) plt.figure(7) plt.title('difference with true source') plt.imshow((np.abs(real_source-sourcesl)), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.show()
4,935
31.473684
130
py
SLIT
SLIT-master/Examples/Test_SLIT_HR.py
import pyfits as pf import matplotlib.pyplot as plt import numpy as np import matplotlib.cm as cm from scipy import signal as scp import SLIT import time from scipy import signal as scp import warnings warnings.simplefilter("ignore") #Example of a run of the SLIT algorithm on simulated images. #Here the first part of the file shows how simulations are generated. #For users intereseted only in seeing the code run, have a look at the running SLIT section. #The command line that builds the Fkappa operator is also of outmost importance. ###############################Simulation############################### def SIE(x0,y0,n1,n2,b,beta,q,xc,theta): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) eps = (1-q**2)/(1+q**2) up = b**(beta-1) pre = up/(2*(1-eps)**((beta-1)/2)) count = 0 theta = theta*np.pi/180. for i in x: Xr = (x[count]-x0)*np.cos(theta)-(y[count]-y0)*np.sin(theta) Yr = (x[count]-x0)*np.sin(theta)+(y[count]-y0)*np.cos(theta) kappa[x[count],y[count]] = pre/((xc**2.)/(1.-eps)+(Xr)**2.+((Yr)**2.)/q**2.)**((beta-1.)/2.) count += 1 return kappa #Source light profile newsource = pf.open('../Files/Source_HR.fits')[0].data ##N1,N2 are the numbers of pixels in the image plane. nt1= 100 nt2 = 100 #Size ratio of the source to image number of pixels size = 2 #PSF PSF0 = pf.open('../Files/PSF.fits')[0].data PSF = PSF0[1:,1:] PSFconj = np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))) PSFconj=PSFconj/np.sum(PSFconj) PSF = PSF/np.sum(PSF) ## Lens mass distribution. b = 1.53/0.05 xc = 0.95 q = 0.71 betata = 2.1 thetata = 25.2 kappa = SIE(nt1/2.+50,nt2/2.+50,nt1+100,nt2+100,b,betata,q,xc, thetata) #Mapping between lens and source IMPORTANT Fkappa = SLIT.Lens.F(kappa, nt1,nt2, size,nt1/2.,nt2/2.) #Generation of lensed source I2 = SLIT.Lens.source_to_image(newsource, nt1 ,nt2 , Fkappa) #Noise levels SNR = 500 sigma = np.sqrt(np.sum(I2**2)/SNR/(nt1*nt2*size**2)) #Convolution by the PSF and generation of the final image I2 = scp.fftconvolve(I2, PSF, mode = 'same') #Final simulated image Image = I2+np.random.randn(nt1,nt2)*sigma ################################Running SLIT############################ #Parameters kmax = 5 niter =50 levels = [0] #Comment the following to have the level estimation routine run (takes more time) levels = pf.open('../Files/Noise_levels_SLIT_HR.fits')[0].data #Start clock start = time.clock() #Running SLIT sourcesl, Imsl = SLIT.SLIT(Image, Fkappa, kmax, niter, size, PSF, PSFconj, levels = levels) #Stop clock elapsed = (time.clock()-start) print('execution time:', elapsed, 'seconds') #Reconstruction goodness real_source = newsource source_error = np.sum(np.abs(real_source[np.where(real_source!=0)] -sourcesl[np.where(real_source!=0)])**2 /real_source[np.where(real_source!=0)]**2)/(np.size( np.where(real_source!=0))/2.) image_chi2 = np.std(Image-Imsl)**2/sigma**2 print('Residuals in source space', source_error) print('Residuals in image space',image_chi2) #Display of results for i in [1]: plt.figure(2) # plt.suptitle('FISTA: error per pixel on the source: '+str(source_error)+' image chi2:'+str(image_chi2)) # plt.subplot(2,3,1) plt.title('Source from SLIT') plt.imshow((sourcesl), vmin = np.min(real_source), vmax = np.max(real_source),cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,2) plt.figure(3) plt.title('Original source') plt.imshow(real_source, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,3) plt.figure(4) plt.title('Lensed source') plt.imshow(Image, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.figure(41) plt.title('Reconstructed lensed source') plt.imshow(Imsl, vmin = np.min(Image), vmax = np.max(Image), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,4) plt.figure(5) plt.title('relative difference') diff = (real_source-sourcesl)/real_source diff[np.where(real_source==0)] = 0 diff[np.where(diff>1)]= np.log(0.) plt.imshow(np.abs(diff), vmax = 1., vmin = 0., cmap = cm.gist_stern, interpolation = 'nearest' ) plt.axis('off') plt.colorbar() # plt.subplot(2,3,5) plt.figure(6) plt.title('difference with reconstructed lensed image') plt.imshow(Image-Imsl, vmin = -5*sigma, vmax = 5*sigma, cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() # plt.subplot(2,3,6) plt.figure(7) plt.title('difference with true source') plt.imshow((np.abs(real_source-sourcesl)), cmap = cm.gist_stern, interpolation = 'nearest') plt.axis('off') plt.colorbar() plt.show()
4,965
31.457516
130
py
SLIT
SLIT-master/Examples/Test_sparsity.py
from SLIT import Lens import pyfits as pf import matplotlib.pyplot as plt import numpy as np import matplotlib.cm as cm from scipy import signal as scp from SLIT import wave_transform as mw import time from scipy import signal as scp import SLIT as slit from mpl_toolkits.axes_grid1.inset_locator import zoomed_inset_axes from mpl_toolkits.axes_grid1.inset_locator import mark_inset import warnings warnings.simplefilter("ignore") S = pf.open('../Files/source.fits')[0].data G = pf.open('../Files/Galaxy.fits')[0].data ##Sizes in image and source planes nt1= 100 nt2 = 100 size = 1 #Mass profile of the lens kappa = pf.open('../Files/kappa.fits')[0].data Fkappa = Lens.F(kappa, nt1,nt2, size,nt1/2.,nt2/2.) lensed = slit.lens_one(Fkappa, nt1,nt2, size) #Levels for normalisation lev = slit.level(nt1,nt1) #Starlet transforms of the lens and source in their respective planes wG = mw.wave_transform(G, lvl = 6, newwave =1)/lev wS = mw.wave_transform(S, lvl = 6, newwave =1)/lev #Lensed source FS = Lens.source_to_image(S, nt1, nt2,Fkappa) #Unlensed lens FG = Lens.image_to_source(G, size, Fkappa, lensed=lensed) #Starlet transform of the unlensed lens wFG = mw.wave_transform(FG, 6, newwave =1)/lev #Starlet transform of the lensed wFS = mw.wave_transform(FS, 6, newwave =1)/lev def mk_sort(X): Y = np.sort(np.resize(np.abs(X), X.size)) return Y[::-1] #Function that computes the reconstruction error from the p% highest coefficients def error_rec_from(X, p, wave = 0): Xcopy = np.copy(X) Y = mk_sort(X) ymin = Y[p*Y.size/1000.] Xcopy[np.abs(X)<ymin] = 0 if wave == 1: err = (mw.iuwt(Xcopy)-mw.iuwt(X))**2 else: err = ((Xcopy)-(X))**2 error = np.sum(err) return error #Computation of reconstruction errors for each light profile error_wS = np.zeros(1000) error_S = np.zeros(1000) error_wFS = np.zeros(1000) error_G = np.zeros(1000) error_wG = np.zeros(1000) error_wFG = np.zeros(1000) for i in np.linspace(0,999, 1000): error_wS[i] = error_rec_from(wS, i, wave = 1) error_S[i] = error_rec_from(S, i) error_wFS[i] = error_rec_from(wFS, i, wave = 1) error_G[i] = error_rec_from(G, i) error_wG[i] = error_rec_from(wG, i, wave = 1) error_wFG[i] = error_rec_from(wFG, i, wave = 1) print('NLA on the source at 10%: ',error_wS[100]/np.max(error_wS)) print('NLA on the lens at 10%: ', error_wG[100]/np.max(error_wG)) print('NLA on the lensed source at 10%: ', error_wFS[100]/np.max(error_wFS)) print('NLA on the delensed lens at 10%: ', error_wFG[100]/np.max(error_wFG)) #Display plt.figure(1) plt.plot(np.linspace(0,100, 1000), error_wS/np.max(error_wS), 'r', label = 'Source in starlet space', linewidth = 3) plt.plot(np.linspace(0,100, 1000), error_wFG/np.max(error_wFG), 'c', label = 'Lens in source plane in starlet space', linewidth = 3) plt.xlabel('percentage of coefficients used in reconstruction', fontsize=25) plt.ylabel('Error on reconstruction', fontsize=25) plt.title('Non-linear approximation error in source plane', fontsize=25) plt.legend(fontsize = 25) a = plt.axes([0.4, 0.2, 0.45, 0.4]) plt.semilogy(np.linspace(0,100, 1000), (error_wFG/np.max(error_wFG)), 'c', linewidth = 3) plt.semilogy(np.linspace(0,100, 1000), error_wS/np.max(error_wS), 'r', linewidth = 3) plt.xlim(20,100) plt.figure(2) plt.plot(np.linspace(0,100, 1000), error_wG/np.max(error_wG), 'b', label = 'Galaxy in starlet space', linewidth = 3) plt.plot(np.linspace(0,100, 1000), error_wFS/np.max(error_wFS), 'm', label = 'Lensed source in starlet space', linewidth = 3) plt.xlabel('percentage of coefficients used in reconstruction', fontsize=25) plt.ylabel('Error on reconstruction', fontsize=25) plt.title('Non-linear approximation error in lens plane', fontsize=25) plt.legend(fontsize = 25) a = plt.axes([0.4, 0.2, 0.45, 0.4]) plt.semilogy(np.linspace(0,100, 1000), (error_wFS/np.max(error_wFS)), 'm', linewidth = 3) plt.semilogy(np.linspace(0,100, 1000), error_wG/np.max(error_wG), 'b', linewidth = 3) plt.xlim(20,100) plt.show()
4,021
33.672414
132
py
SLIT
SLIT-master/build/lib/SLIT/Solve.py
#from __future__ import division import wave_transform as mw import numpy as np import matplotlib.pyplot as plt import pyfits as pf import matplotlib.cm as cm from scipy import signal as scp import scipy.ndimage.filters as med import MuSCADeT as wine from numpy import linalg as LA import multiprocess as mtp from pathos.multiprocessing import ProcessingPool as Pool import Lens import warnings import tools warnings.simplefilter("ignore") ##SLIT: Sparse Lens Inversion Technique def SLIT(Y, Fkappa, kmax, niter, size, PSF, PSFconj, S0 = [0], levels = [0], scheme = 'FB', mask = [0], lvl = 0, weightS = 1, noise = 'gaussian', tau = 0): ##DESCRIPTION: ## Function that estimates the source light profile from an image of a lensed source given the mass density profile. ## ##INPUTS: ## -img: a 2-D image of a lensed source given as n1xn2 numpy array. ## -Fkappa: an array giving the mapping between lens and source. This array is calculated from the lens mass density ## using tools from SLIT.Lens ## -kmax: the detection threshold in units of noise levels. We usualy set this value to 5 to get a 5 sigma ## detection threshold. ## -niter: maximal number of iterations of the algorithm. ## -size: resoluution factor between lens and source grids such thathe size of the output source ## will be n1sizexn2size ## -PSF: the point spread function of the observation provided as a 2D array. ## -PSFconj: The conjugate of the PSF. Usually computed via np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))) ## butthe user has to make sure that the conjugate is well centered. ## ##OPTIONS: ## -levels: an array that contains the noise levels at each band of the wavelet decomposition of the source. ## If not provided, the routine will compute the levels and save them in a fits file 'Noise_levels.fits' ## so that they can be used at a later time. This option allows to save time when running the same ## experiment several times. ## -mask: an array of zeros and one with size ns1xns2. The zeros will stand for masked data. ## ##OUTPUTS: ## -S: the source light profile. ## -FS: the lensed version of the estimated source light profile ## ##EXAMPLE: ## S,FS = SLIT(img, Fkappa, 5, 100, 1, PSF, PSFconj) n1,n2 = np.shape(Y) # PSFconj = np.rot90(PSF, 2) #Size of the source ns1,ns2 = n1*size, n2*size #Number of starlet scales in source plane if lvl ==0: lvl = np.int(np.log2(ns2)) else: lvl = np.min([lvl,np.int(np.log2(ns2))]) #Masking if required if np.sum(mask) == 0: mask = np.ones((n1,n2)) img = Y*mask #Noise in image plane sigma0 = MAD(Y) if noise == 'poisson': if tau ==0: print('error: Give exposure time') Y0 =np.copy(Y) sigma = np.copy(sigma0) Y = 2./tau*np.sqrt(tau*np.abs(Y)+tau*3./8.+sigma0)*np.sign(tau*Y+tau*3./8.+sigma0) sigma0 = MAD(Y) #Mapping of an all-at-one image to source plane lensed = lens_one(Fkappa, n1,n2, size) #estimation of the frame of the image in source plane supp = np.zeros((lvl,lensed.shape[0],lensed.shape[1])) supp[:,lensed/lensed ==1] =1 #Useful functions def Finv_apply(I): return Lens.image_to_source(I, size, Fkappa, lensed = lensed) def Lens_op2(I): return Lens.image_to_source(I, size, Fkappa, lensed = lensed, square = 1) def F_apply(Si): return Lens.source_to_image(Si, n1, n2,Fkappa) def PSF_apply(i): return scp.fftconvolve(i,PSF,mode = 'same') def PSFT_apply(ii): return scp.fftconvolve(ii,PSFconj,mode = 'same') def transform(x): return tools.wave_transform(x, lvl, newwave = 1) def inverse(x): return tools.iuwt(x) #Forward operator def F_op(X): return PSF_apply(F_apply(X)) #Inverse operator def I_op(X): return Finv_apply(PSFT_apply(X)) #Regularisation (Backward term) def reg0(X): return tools.Hard(X, transform, inverse,levels, (ks), supp=supp) def reg00(X): return tools.Hard_Threshold(X, transform, inverse,levels, (ks), supp=supp) def reg1(X): return tools.Soft(X, transform, inverse,levels*weightS, kmax, supp=supp) def reg_filter(X): return tools.mr_filter(X,levels,ks,20,transform, inverse, lvl = lvl) #Noise simulations to estimate noise levels in source plane if np.sum(levels)==0: print('Calculating noise levels') #levels = simulate_noise(n1,n2, sigma0, size, I_op, transform, lvl) levels = level_source(n1,n2,sigma0,size,PSFconj, Lens_op2, lensed, lvl) #Saves levels hdus = pf.PrimaryHDU(levels) lists = pf.HDUList([hdus]) lists.writeto('Noise_levels.fits', clobber=True) ##Compute spectral norms op_norm = spectralNorm(ns1,ns2,20,1e-10,F_op,I_op) wave_norm = spectralNorm(ns1,ns2,20,1e-10,transform,inverse) nu = 0.5#op_norm**2/(2*wave_norm**2)-1./(mu) mu = 1./(op_norm+wave_norm) if scheme == 'FB': repeat =1 else: repeat = 2 #Initialisation Res1= [] tau = 0.5 for jr in range(repeat): trans = (transform(I_op(Y))/levels)[:-1,:,:] ks0 = np.max(trans[levels[:-1,:,:]!=0]) ks=np.copy(ks0) steps = (ks0-kmax)/(niter-5) karg = np.log(kmax/ks0)/(niter-5.) i = 0 if np.sum(S0) == 0: S=np.random.randn(ns1,ns2)*np.median(sigma0)*0 else: S = S0 ts = 1 csi = 0 Res1= [] Res2 = [] alpha =transform(S) while i < niter: print(i) if scheme == 'FB': ks = ks0*np.exp(i*karg) ks = np.max([ks, kmax]) S = tools.Forward_Backward(Y, S, F_op, I_op, mu, reg_filter, pos = 1) S[S<0] = 0 FS = F_op(S)*mask else: alpha, csi, ts = tools.FISTA(Y, alpha, F_op, I_op, mu, ts, csi, reg1, transform, inverse, mask = mask) S = inverse(alpha) FS = F_op(S) #Convergence condition Res1.append((np.std(Y-FS)**2)/np.median(sigma0)**2) Res2.append(ks) # ks = ks-steps i = i+1 S[S<0] = 0 # alpha = transform(S) weightS = 1./(1.+np.exp(-10.*(levels*kmax-alpha))) # plt.show() #Final reconstruction of the source plt.plot(Res1, 'b'); plt.show() plt.plot(Res2, 'r'); plt.show() if noise == 'poisson': plt.subplot(211) plt.title('S') plt.imshow(S); plt.colorbar() plt.show() FS = F_op(S)*mask return S, FS #############################SLIT MCA for blended lenses############################ def SLIT_MCA(Y, Fkappa, kmax, niter, riter, size,PSF, PSFconj, lvlg = 0, lvls = 0, levels = [0], mask = [0,0], Ginit=0): ##DESCRIPTION: ## Function that estimates the source and lens light profiles from an image of a ## strong lens system ## ##INPUTS: ## -img: a 2-D image of a lensed source given as n1xn2 numpy array. ## -Fkappa: an array giving the mapping between lens and source. This array is calculated from the lens mass density ## using tools from SLIT.Lens ## -kmax: the detection threshold in units of noise levels. We usualy set this value to 5 to get a 5 sigma ## detection threshold. ## -niter: maximal number of iterations in the main loop over G. ## -riter: maximal number of iterations in the inner loop over S. ## -size: resoluution factor between lens and source grids such thathe size of the output source ## will be n1sizexn2size ## -PSF: the point spread function of the observation provided as a 2D array. ## -PSFconj: The conjugate of the PSF. Usually computed via np.real(np.fft.ifft2(np.conjugate(np.fft.fft2(PSF0[:-1,:-1])))) ## butthe user has to make sure that the conjugate is well centered. ## ##OPTIONS: ## -levels: an array that contains the noise levels at each band of the wavelet decomposition of the source. ## If not provided, the routine will compute the levels and save them in a fits file 'Noise_levels.fits' ## so that they can be used at a later time. This option allows to save time when running the same ## experiment several times. ## -mask: an array of zeros and one with size ns1xns2. The zeros will stand for masked data. ## -Ginit: Educated guedd for the lens galaxy light profile. if set to a 2D numpy array, the array will be used as ## as an initialisation for G. ## ##OUTPUTS: ## -S: the source light profile. ## -G: the convolved lens light profile ## -FS: the lensed version of the estimated source light profile ## ##EXAMPLE: ## S,FS = SLIT(img, Fkappa, 5, 100, 1, PSF, PSFconj) #Shape of the image n1,n2 = np.shape(Y) #Initialisation of the source ns1= n1*size ns2 = n2*size #Number of starlet scales in source and image planes if lvlg ==0: lvlg = np.int(np.log2(n2)) else: lvlg = np.min([lvlg,np.int(np.log2(n2))]) lvls = lvlg if lvls >np.int(np.log2(ns2)): print('Error, too many wavelet levels for the source. Choose a smaller value for lvl') exit #Masking if required if np.sum(mask) == 0: mask = np.ones((n1,n2)) Y = Y*mask #Noise standard deviation in image plane sigma0 = MAD(Y) #Mapping of an all-at-one image lensed = lens_one(Fkappa, n1,n2, size) supp = np.zeros((lvls,lensed.shape[0],lensed.shape[1])) supp[:,lensed/lensed ==1] =1 #Limits of the image plane in source plane bound = mk_bound(Fkappa, n1,n2, size) #Noise levels in image plane in starlet space levelg = level(n1,n2, lvlg)*sigma0 #Useful functions def Finv_apply(I): return Lens.image_to_source(I, size, Fkappa, lensed = lensed) def F_apply(Si): return Lens.source_to_image(Si, n1, n2,Fkappa) def PSF_apply(i): return scp.fftconvolve(i,PSF,mode = 'same') def PSFT_apply(ii): return scp.fftconvolve(ii,PSFconj,mode = 'same') def transform(x): return tools.wave_transform(x, lvlg) def inverse(x): return tools.iuwt(x) #Forward Source operator def FS_op(X): return PSF_apply(F_apply(X)) #Inverse Source operator def IS_op(X): return Finv_apply(PSFT_apply(X)) #Forward Lens operator def FG_op(X): return ((X)) #Inverse Lens operator def IG_op(X): return ((X)) #Regularisation (Backward term) def regG0(X): return tools.Hard_Threshold(X, transform, inverse, levelg*kG) def regS0(X): return tools.Hard_Threshold(X, transform, inverse, levels*kS) def regS1(X): return tools.Soft(X, transform, inverse, levels*kmax*weightS, supp = supp) def regG1(X): return tools.Soft(X, transform, inverse, levelg*(kmax)*weightG, supp = 1) def reg_filter(X): return tools.mr_filter(X, levelg*kG*sigma0, 20, transform, inverse, Soft = 0, pos = 1) #Noise simulations to estimate noise levels in source plane if np.sum(levels)==0: print('Calculating noise levels') levels = simulate_noise(n1,n2, sigma0, size, IS_op, transform, lvls) levels[:,lensed ==0] = np.max(levels*10) #Saves levels hdus = pf.PrimaryHDU(levels) lists = pf.HDUList([hdus]) lists.writeto('Noise_levels_MCA.fits', clobber=True) #Computationt of spectral norms FS_norm = spectralNorm(ns1,ns2,20,1e-10,FS_op,IS_op) Star_norm_im = spectralNorm(n1,n2,20,1e-10,transform,inverse) Star_norm_s = spectralNorm(ns1,ns2,20,1e-10,transform,inverse) muG = 1./(Star_norm_im**2) muS = 1./(Star_norm_s*FS_norm)**2 print(muS, muG) weightS = 1 weightG = 1 #Reweighting loop for it in range(3): #Initialisations FS = 0 G = np.zeros((n1,n2)) S = np.zeros((ns1,ns2)) alphaS = transform(S) csiS = np.copy(alphaS) alphaG = transform(G) csiG = np.copy(alphaG) i = 0 K_s = np.zeros(niter) tg=1 #Beginning of main loop while i < niter: print('main loop: ',i) DS = Y-G ts = 1 for j in range(riter): # S = tools.Forward_Backward(DS, S, FS_op, IS_op, muS, regS0) alphaS, csiS, ts = tools.FISTA(DS, alphaS, FS_op, IS_op, muS, ts, csiS, regS1, transform, inverse, pos = 1) S = inverse(alphaS)#*supp S[S<0] = 0 FS = FS_op(S) DG = Y-FS for j2 in range(riter): alphaG, csiG, tg = tools.FISTA(DG, alphaG, FG_op, IG_op, muG, tg, csiG, regG1, transform, inverse, pos = 1) #Image to approximate by solving the problem in G G = inverse(alphaG) # newres = (np.std(Y-FS-G)**2)/sigma0**2 K_s[i] = newres res = np.copy(newres) plt.figure(0) plt.subplot(221) plt.title('S') plt.imshow(S) plt.subplot(222) plt.title('FS') plt.imshow(FS) plt.subplot(223) plt.title('G') plt.imshow(G) plt.subplot(224) plt.title('Residuals') plt.imshow(Y-FS-G) plt.savefig('Res'+str(i)+'.png') i +=1 #Weighting weightS = 1./(1.+np.exp(-10.*(levels*kmax*sigma0-alphaS))) weightG = 1./(1.+np.exp(-10.*(levelg*kmax*sigma0-alphaG))) S, FS = SLIT(Y-G, Fkappa, kmax, niter, size, PSF, PSFconj, S0 = S, levels = levels, mask = mask, lvl = lvls) #Final reconstructions plt.show() plt.plot(K_s); plt.show() return S, FS,G ################################### TOOLS ################################### def MOM(Y, levels, levelg, kmax, transform, inverse, IS_op, sigma, niter, I = 0): Gw0 = transform((Y))[:-1,:,:] levelg1 = levelg[:-1,:,:] Gw = Gw0[levelg1!=0]/levelg[levelg1!=0]/sigma kG = np.max(Gw) kG0 = kG stepG = (kG-kmax)/(niter-I-5) FS0 = Y Sw0 = transform(IS_op(FS0))[:-1,:,:] levels1 = levels[:-1,:,:] Sw = Sw0[levels1!=0]/levels[levels1!=0]/sigma kS = np.max(Sw) k =np.min([kG,kS]) k = np.max([k,kmax])+(np.abs(kS-kG))/100. step = (k-kmax)/(niter-I-5) return k, step def plot_cube(cube): ##DESCRIPTION: ## Plotting device that displays layers of a cube in different subplot panels. ## ##INPUTS: ## -cube: Cube for which to plot the layers with shape (n,n1,n2) with n, the number of layers and n1xn2, the number of pixels. ## ##OUTPUTS: ## -None n,n1,n2 = np.shape(cube) i = n/2 if i == n/2.+0.5: i+=1 j = 2 for k in range(n): plt.subplot(i,j,k) plt.imshow(cube[k,:,:]); plt.colorbar() return None def level(n1,n2, lvl): ##DESCRIPTION: ## Estimates the noise levels in starlet space in image plane. ## ##INPUTS: ## -n1,n2: shape of the image for which to get noise levels ## ##OUTPUTS: ## -levels: units of noise levels at each scale and location of a starlet transform dirac = np.zeros((n1,n2)) # lvl = np.int(np.log2(n1)) dirac[n1/2,n2/2] = 1 wave_dirac = mw.wave_transform(dirac,lvl, newwave = 0) wave_sum = np.sqrt(np.sum(np.sum(wave_dirac**2,1),1)) levels = np.multiply(np.ones((lvl,n1,n2)).T,wave_sum).T return levels def level_source(n1,n2,sigma,size,PSFT, Lens_op2, lensed, lvl): ns1,ns2 = n1*size, n2*size ones = np.ones((n1,n2)) lensed[lensed == 0] = 1 noise = ones*sigma Hnoise = noise*np.sqrt(np.sum(PSFT**2))#np.sqrt(scp.fftconvolve(noise**2, PSFT**2, mode = 'same'))## FHnoise = Lens_op2(Hnoise) FHnoise[FHnoise==0] = np.mean(FHnoise)*10. dirac = np.zeros((ns1,ns2)) dirac[ns1/2,ns2/2] = 1 wave_dirac = mw.wave_transform(dirac, lvl) levels = np.zeros(wave_dirac.shape) for i in range(lvl): if np.size(noise.shape) > 2: lvlso = (scp.fftconvolve(FHnoise[i, :, :] ** 2, wave_dirac[i, :, :] ** 2, mode='same')) else: lvlso = scp.fftconvolve(FHnoise ** 2, wave_dirac[i,:,:] ** 2, mode='same') levels[i, :, :] = np.sqrt(np.abs(lvlso)) return levels def spectralNorm(nx,ny,Niter,tol,f,finv): ##DESCRIPTION: ## Function that estimates the source light profile from an image of a lensed source given the mass density profile. ## ##INPUTS: ## -nx,ny: shape of the input ## -nz: number of decomposition scales (if the operator tis a multiscale decomposition for instance) ## -Niter: number of iterations ## -tol: tolerance error as a stopping criteria ## -f: operator ## -finv: inverse operator ## ##OUTPUTS: ## -SspNorm: The spectral norm of the operator #### Initilize array with random numbers ### matA = np.random.randn(nx,ny) ### Normalize the input ### spNorm = LA.norm(matA) matA /= spNorm matA = np.array(matA) it = 0 err = abs(tol) while it < Niter and err >= tol: ### Apply operator ### wt = f(matA) ### Apply joint operator ### matA = finv(wt) ### Compute norm ### spNorm_new = LA.norm(matA) matA /= spNorm_new err = abs(spNorm_new - spNorm)/spNorm_new spNorm = spNorm_new it += 1 return spNorm def lens_one(Fkappa, n1,n2,size): ##DESCRIPTION: ## Function that maps an all at one image to source plane. ## ##INPUTS: ## -Fkappa: the mapping between source and image planes ## -n1,n2: the shape of the image. ## -size: the factor that scales the shape of the source relative to the shape of the image ## ##OUTPUTS: ## -lensed: the projection to source plane of an all at aone image. dirac = np.ones((n1,n2)) lensed = Lens.image_to_source(dirac, size,Fkappa,lensed = [0]) return lensed def mk_bound(Fkappa, n1,n2,size): ##DESCRIPTION: ## Function that returns the support of the lens image in source plane. ## ##INPUTS: ## -Fkappa: the mapping between source and image planes ## -n1,n2: the shape of the image. ## -size: the factor that scales the shape of the source relative to the shape of the image ## ##OUTPUTS: ## -lensed: the projection to source plane of an all at aone image. dirac = np.ones((n1,n2)) lensed = Lens.image_to_source_bound(dirac, size,Fkappa,lensed = [0]) bound = lensed/lensed bound[lensed==0]=0 return bound def MAD(x,n=3): ##DESCRIPTION: ## Estimates the noise standard deviation from Median Absolute Deviation ## ##INPUTS: ## -x: a 2D image for which we look for the noise levels. ## ##OPTIONS: ## -n: size of the median filter. Default is 3. ## ##OUTPUTS: ## -S: the source light profile. ## -FS: the lensed version of the estimated source light profile x = mw.wave_transform(x, np.int(np.log2(x.shape[0])))[0,:,:] meda = med.median_filter(x,size = (n,n)) medfil = np.abs(x-meda) sh = np.shape(x) sigma = 1.48*np.median((medfil)) return sigma def MAD_poisson(x,tau,n=3): ##DESCRIPTION: ## Estimates the noise standard deviation from Median Absolute Deviation ## ##INPUTS: ## -x: a 2D image for which we look for the noise levels. ## ##OPTIONS: ## -n: size of the median filter. Default is 3. ## ##OUTPUTS: ## -S: the source light profile. ## -FS: the lensed version of the estimated source light profile n1,n2 = np.shape(x) lvl = np.int(np.log2(x.shape[0]))-1 new_x, y = wine.MCA.mr_filter(x,20,8,MAD(x), lvl = lvl) plt.imshow(new_x); plt.show() sigma = np.sqrt(np.abs(new_x)/tau) return sigma def ST(alpha, k, levels, sigma, hard = 0): ##DESCRIPTION: ## Soft thresholding operator. ## ##INPUTS: ## -alpha: the starlet decomposition to be thresholded. ## -k: the threshold in units of noise levels (usually 5). ## -levels: the noise levels at each scale and location of the starlet decomposition. ## -sigma: the noise standard deviation. ## ##OUTPUTS: ## -alpha: The thresholded coefficients. lvl, n1,n2 = np.shape(alpha) th = np.ones((lvl,n1,n2))*k th[0,:,:] = th[0,:,:]+3 th[-1,:,:] = 0 alpha0 = np.copy(alpha) th = th*levels*sigma if hard == 0: alpha= np.sign(alpha0)*(np.abs(alpha0)-th) alpha[np.where(np.abs(alpha)-th<0)]=0 return alpha def mk_simu(n1,n2,lvl,size, sigma, I_op, transform, n): storage = np.zeros((lvl,n1*size, n2*size, n)) for i in range(n): noise = np.random.randn(n1,n2)*sigma noise_lens = I_op(noise) noise_lens[noise_lens ==0] = 1 storage[:,:,:,i] = transform(noise_lens) return storage def simulate_noise(n1,n2, sigma, size, I_op, transform, lvl, Npar = np.int(mtp.cpu_count()/2)): ##DESCRIPTION: ## Simulates noise levels in source plane from lensing operator and convolution operator. ## ##INPUTS: ## -n1,n2: the shape of the images for which to simulate noise maps. ## -size: scaling factor for the shape of the source. ## -Fkappa: Projection operator between lens and source plane. ## -lensed: mapping of an all at one image to source plane. ## -PSFconj: the conjugate of the PSF ## ##OPTIONS: ## -n: size of the median filter. Default is 3. ## ##OUTPUTS: ## -S: the source light profile. ## -FS: the lensed version of the estimated source light profile n = 500 if Npar>mtp.cpu_count(): Npar = mtp.cpu_count() ns1,ns2 = n1*size, n2*size # lvl = np.int(np.log2(ns1)) w_levels = np.zeros((lvl,ns1,ns2)) p = Pool(Npar) storage = mk_simu(n1,n2,lvl,size, sigma, I_op, transform,n) w_levels = np.std(storage, axis = 3) # w_levels[0,:,:] = w_levels[0,:,:]*6/5 return w_levels
22,735
31.713669
131
py
SLIT
SLIT-master/build/lib/SLIT/wave_transform.py
import numpy as np import scipy.signal as cp import matplotlib.pyplot as plt import scipy.ndimage.filters as sc def symmetrise(img, size): n3, n4 = np.shape(img) n1,n2 = size img[:(n3-n1)/2, :] = np.flipud(img[(n3-n1)/2:(n3-n1),:]) img[:,:(n4-n2)/2] = np.fliplr(img[:,(n4-n2)/2:(n4-n2)]) img[(n3+n1)/2:,:] = np.flipud(img[n1:(n3+n1)/2,:]) img[:,(n4+n2)/2:] = np.fliplr(img[:,n2:(n4+n2)/2]) return img def fft_convolve(X,Y, inv = 0): XF = np.fft.rfft2(X) YF = np.fft.rfft2(Y) # YF0 = np.copy(YF) # YF.imag = 0 # XF.imag = 0 if inv == 1: # plt.imshow(np.real(YF)); plt.colorbar(); plt.show() YF = np.conj(YF) SF = XF*YF S = np.fft.irfft2(SF) n1,n2 = np.shape(S) S = np.roll(S,-n1/2+1,axis = 0) S = np.roll(S,-n2/2+1,axis = 1) return np.real(S) def wave_transform(img, lvl, Filter = 'Bspline', newwave = 1, convol2d = 0): mode = 'nearest' lvl = lvl-1 sh = np.shape(img) if np.size(sh) ==3: mn = np.min(sh) wave = np.zeros([lvl+1,sh[1], sh[1],mn]) for h in np.linspace(0,mn-1, mn): if mn == sh[0]: wave[:,:,:,h] = wave_transform(img[h,:,:],lvl+1, Filter = Filter) else: wave[:,:,:,h] = wave_transform(img[:,:,h],lvl+1, Filter = Filter) return wave n1 = sh[1] n2 = sh[1] if Filter == 'Bspline': h = [1./16, 1./4, 3./8, 1./4, 1./16] else: h = [1./4,1./2,1./4] n = np.size(h) h = np.array(h) if n+2**(lvl-1)*(n-1) >= np.min([n1,n2])/2.: lvl = np.int_(np.log2((n1-1)/(n-1.))+1) c = img ## wavelet set of coefficients. wave = np.zeros([lvl+1,n1,n2]) for i in np.linspace(0,lvl-1,lvl): newh = np.zeros((1,n+(n-1)*(2**i-1))) newh[0,np.int_(np.linspace(0,np.size(newh)-1,len(h)))] = h H = np.dot(newh.T,newh) ######Calculates c(j+1) ###### Line convolution if convol2d == 1: cnew = cp.convolve2d(c, H, mode='same', boundary='symm') else: cnew = sc.convolve1d(c,newh[0,:],axis = 0, mode =mode) ###### Column convolution cnew = sc.convolve1d(cnew,newh[0,:],axis = 1, mode =mode) if newwave ==1: ###### hoh for g; Column convolution if convol2d == 1: hc = cp.convolve2d(cnew, H, mode='same', boundary='symm') else: hc = sc.convolve1d(cnew,newh[0,:],axis = 0, mode = mode) ###### hoh for g; Line convolution hc = sc.convolve1d(hc,newh[0,:],axis = 1, mode = mode) ###### wj+1 = cj-hcj+1 wave[i,:,:] = c-hc else: ###### wj+1 = cj-cj+1 wave[i,:,:] = c-cnew c = cnew wave[i+1,:,:] = c return wave def iuwt(wave, convol2d =0): mode = 'nearest' lvl,n1,n2 = np.shape(wave) h = np.array([1./16, 1./4, 3./8, 1./4, 1./16]) n = np.size(h) cJ = np.copy(wave[lvl-1,:,:]) for i in np.linspace(1,lvl-1,lvl-1): newh = np.zeros((1,n+(n-1)*(2**(lvl-1-i)-1))) newh[0,np.int_(np.linspace(0,np.size(newh)-1,len(h)))] = h H = np.dot(newh.T,newh) ###### Line convolution if convol2d == 1: cnew = cp.convolve2d(cJ, H, mode='same', boundary='symm') else: cnew = sc.convolve1d(cJ,newh[0,:],axis = 0, mode = mode) ###### Column convolution cnew = sc.convolve1d(cnew,newh[0,:],axis = 1, mode = mode) cJ = cnew+wave[lvl-1-i,:,:] return np.reshape(cJ,(n1,n2))
3,715
25.169014
81
py
SLIT
SLIT-master/build/lib/SLIT/tools.py
import numpy as np import matplotlib.pyplot as plt import pyfits as pf import scipy.ndimage.filters as sc import scipy.ndimage.filters as med import scipy.signal as cp def MAD(x,n=3): ##DESCRIPTION: ## Estimates the noise standard deviation from Median Absolute Deviation ## ##INPUTS: ## -x: a 2D image for which we look for the noise levels. ## ##OPTIONS: ## -n: size of the median filter. Default is 3. ## ##OUTPUTS: ## -S: the source light profile. ## -FS: the lensed version of the estimated source light profile x = wave_transform(x, np.int(np.log2(x.shape[0])))[0,:,:] meda = med.median_filter(x,size = (n,n)) medfil = np.abs(x-meda) sh = np.shape(x) sigma = 1.48*np.median((medfil)) return sigma def Forward_Backward(Y, X, F_op, I_op, mu, reg, pos = 1): R = mu*I_op(Y-F_op(X)) Xnew = np.copy(X+R) Xnew, M = reg(Xnew) if pos == 1: Xnew[Xnew<0] = 0 return Xnew, M def Primal_dual(Y, X, U, mu, nu, tau, F_op, I_op, transform, inverse, reg): p = X+mu*I_op(Y-F_op(X))-mu*inverse(U) # plot_cube(U+nu*transform(2*p-X)); plt.show() q = reg(U+nu*transform(2*p-X)) X =tau*p+(1-tau)*X U =tau*q+(1-tau)*U # plot_cube(q); plt.show() return X,U def FISTA(Y, alphaX, F_op, I_op, mu, ts, csi, reg, transform, inverse, pos = 1, mask = 1): S = inverse(alphaX) #S[S>size*np.max(Y)] = np.max(Y) # S[S<0] = 0 R = mu*I_op(Y-F_op(S)*mask) alpha = transform(R)+csi alpha, M = reg(alpha) tsnew = (1.+np.sqrt(1.+4.*ts**2))/2. csi = alpha+((ts-1)/tsnew)*(alpha-alphaX) return alpha, csi, tsnew def Soft(X, transform, inverse, level, k, supp =1): Xnew = np.sign(X)*(np.abs(X)-level*k) Xnew[np.where((np.abs(X)-level*k)<0)] = 0 Xnew[-1,:,:] = X[-1,:,:] #print(Xnew.shape, supp.shape) Xnew = Xnew*supp return Xnew def Soft_Threshold(X, transform, inverse, level, k, supp =1): X = transform(X) alpha = wave_transform(X,Xw.shape[0],newwave = 0) M = np.zeros(alpha.shape) M[np.abs(alpha)-level*k>0] = 1 M[0,:,:] = 0 M[0,np.abs(alpha[0,:,:]) - level[0,:,:] * (k+1) > 0] = 1 Xnew = np.sign(X)*(np.abs(X)-level*k) Xnew = Xnew*M Xnew[-1,:,:] = X[-1,:,:] Xnew = Xnew*supp return inverse(Xnew), M def Hard(X, transform, inverse, level, k, supp=1): Xnew = np.copy(X) Xnew[np.where((np.abs(X)-level*k)<0)] = 0 Xnew[-1,:,:] = X[-1,:,:] Xnew = Xnew*supp ## plt.figure(0) ## plot_cube(X) ## plt.figure(1) ## plot_cube(Xnew) ## plt.show() return Xnew, M def Hard_Threshold(X, transform, inverse, level, k, supp=1): Xw = transform(X) alpha = wave_transform(X,Xw.shape[0],newwave = 0) M = np.zeros(alpha.shape) M[np.abs(alpha)-level*k>0] = 1 M[0,:,:] = 0 M[0,np.abs(alpha[0,:,:]) - level[0,:,:] * (k+1) > 0] = 1 Xnew=M*Xw Xnew[-1,:,:] = Xw[-1,:,:] Xnew = Xnew*supp return inverse(Xnew), M def mr_filter(Y, level, k, niter, transform, inverse, lvl = 6, Soft = 0, pos = 1): Xnew = 0 alpha = wave_transform(Y, lvl, newwave=0) M = np.zeros(alpha.shape) M[np.abs(alpha)-level*k>0] = 1 M[0,:,:] = 0 M[0,np.abs(alpha[0,:,:]) - level[0,:,:] * (k+1) > 0] = 1 M[-1,:,:] =1 i=0 while i < niter: R = Y-Xnew if Soft == True : Rnew= Soft_threshold(R, transform, inverse, level,k) else: Rnew = Hard_Threshold(R, transform, inverse, level,k) # Rnew = inverse(transform(R)*M) Xnew = Xnew+Rnew if pos == True: Xnew[Xnew < 0] = 0 i = i+1 return (Xnew), M def wave_transform(img, lvl, Filter = 'Bspline', newwave = 1, convol2d = 0): mode = 'nearest' lvl = lvl-1 sh = np.shape(img) if np.size(sh) ==3: mn = np.min(sh) wave = np.zeros([lvl+1,sh[1], sh[1],mn]) for h in np.linspace(0,mn-1, mn): if mn == sh[0]: wave[:,:,:,h] = wave_transform(img[h,:,:],lvl+1, Filter = Filter) else: wave[:,:,:,h] = wave_transform(img[:,:,h],lvl+1, Filter = Filter) return wave n1 = sh[1] n2 = sh[1] if Filter == 'Bspline': h = [1./16, 1./4, 3./8, 1./4, 1./16] else: h = [1./4,1./2,1./4] n = np.size(h) h = np.array(h) if n+2**(lvl-1)*(n-1) >= np.min([n1,n2])/2.: lvl = np.int_(np.log2((n1-1)/(n-1.))+1) c = img ## wavelet set of coefficients. wave = np.zeros([lvl+1,n1,n2]) for i in np.linspace(0,lvl-1,lvl): newh = np.zeros((1,n+(n-1)*(2**i-1))) newh[0,np.int_(np.linspace(0,np.size(newh)-1,len(h)))] = h H = np.dot(newh.T,newh) ######Calculates c(j+1) ###### Line convolution if convol2d == 1: cnew = cp.convolve2d(c, H, mode='same', boundary='symm') else: cnew = sc.convolve1d(c,newh[0,:],axis = 0, mode =mode) ###### Column convolution cnew = sc.convolve1d(cnew,newh[0,:],axis = 1, mode =mode) if newwave ==1: ###### hoh for g; Column convolution if convol2d == 1: hc = cp.convolve2d(cnew, H, mode='same', boundary='symm') else: hc = sc.convolve1d(cnew,newh[0,:],axis = 0, mode = mode) ###### hoh for g; Line convolution hc = sc.convolve1d(hc,newh[0,:],axis = 1, mode = mode) ###### wj+1 = cj-hcj+1 wave[i,:,:] = c-hc else: ###### wj+1 = cj-cj+1 wave[i,:,:] = c-cnew c = cnew wave[i+1,:,:] = c return wave def iuwt(wave, convol2d =0): mode = 'nearest' lvl,n1,n2 = np.shape(wave) h = np.array([1./16, 1./4, 3./8, 1./4, 1./16]) n = np.size(h) cJ = np.copy(wave[lvl-1,:,:]) for i in np.linspace(1,lvl-1,lvl-1): newh = np.zeros((1,n+(n-1)*(2**(lvl-1-i)-1))) newh[0,np.int_(np.linspace(0,np.size(newh)-1,len(h)))] = h H = np.dot(newh.T,newh) ###### Line convolution if convol2d == 1: cnew = cp.convolve2d(cJ, H, mode='same', boundary='symm') else: cnew = sc.convolve1d(cJ,newh[0,:],axis = 0, mode = mode) ###### Column convolution cnew = sc.convolve1d(cnew,newh[0,:],axis = 1, mode = mode) cJ = cnew+wave[lvl-1-i,:,:] return np.reshape(cJ,(n1,n2)) def plot_cube(cube): ##DESCRIPTION: ## Plotting device that displays layers of a cube in different subplot panels. ## ##INPUTS: ## -cube: Cube for which to plot the layers with shape (n,n1,n2) with n, the number of layers and n1xn2, the number of pixels. ## ##OUTPUTS: ## -None n,n1,n2 = np.shape(cube) i = n/2 if i == n/2.+0.5: i+=1 j = 2 for k in range(n): plt.subplot(i,j,k) plt.imshow(cube[k,:,:]); plt.colorbar() return None
7,049
26.539063
131
py
SLIT
SLIT-master/build/lib/SLIT/Lens.py
import numpy as np import matplotlib.pyplot as plt import pyfits as pf import scipy.signal as scp import warnings warnings.simplefilter("ignore") #Tool box for lensing def SIS(x0,y0,n1,n2,Re): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) count = 0 for i in x: kappa[x[count],y[count]] = Re/(2*np.sqrt((x[count]-x0)**2+(y[count]-y0)**2)) count += 1 if np.isfinite(kappa[x0,y0]) == False: kappa[x0,y0] = 1 return kappa def SIE_xy(x,y,x0,y0,b,beta,q,xc,theta): eps = (1-q**2)/(1+q**2) up = b**(beta-1) pre = up/(2*(1-eps)**((beta-1)/2)) count = 0 theta = theta*np.pi/180. Xr = (x-x0)*np.cos(theta)-(y-y0)*np.sin(theta) Yr = (x-x0)*np.sin(theta)+(y-y0)*np.cos(theta) kappa = pre/((xc**2.)/(1.-eps)+(Xr)**2.+((Yr)**2.)/q**2.)**((beta-1.)/2.) return kappa def SIE(x0,y0,n1,n2,b,beta,q,xc,theta): kappa = np.zeros((n1,n2)) x,y = np.where(kappa == 0) x2d = np.reshape(x, (n1,n2)) y2d = np.reshape(y, (n1,n2)) kappa = SIE_xy(x2d,y2d,x0,y0,b,beta,q,xc,theta) return kappa def alpha_def(kappa, n1,n2,x0,y0,extra): #Computes the deflection angle of a single photon at coordinates theta in the source plane and a lens #mass distribution kappa nk1,nk2 = np.shape(kappa) x0+=extra/2. y0+=extra/2. #Coordonnees de la grille de l'espace image [x,y] = np.where(np.zeros([nk1,nk2])==0) [xk,yk] = np.where(np.zeros([nk1,nk2])==0) xc = np.reshape((x)-x0,(nk1,nk2)) yc = np.reshape((y)-y0,(nk1,nk2)) xkc = np.reshape((xk)-(nk1/2.),(nk1,nk2)) ykc = np.reshape((yk)-(nk2/2.),(nk1,nk2)) r = (xc**2+yc**2) rk = np.sqrt(xkc**2+ykc**2) lx,ly = np.where(r==0) tabx = np.reshape((xc)/r,(nk1,nk2)) taby = np.reshape((yc)/r,(nk1,nk2)) tabx[lx,ly]=0 taby[lx,ly]=0 l = 0 kappa = kappa.astype(float) tabx = tabx.astype(float) # kappa[rk>(nk1)/2.] = 0 intex = scp.fftconvolve(tabx, (kappa), mode = 'same')/np.pi intey = scp.fftconvolve(taby, (kappa), mode = 'same')/np.pi return intex[x0-(n1)/2.:x0+(n1)/2,y0-(n2)/2.:y0+(n2)/2.],intey[x0-(n1)/2.:x0+(n1)/2,y0-(n2)/2.:y0+(n2)/2.] def beta(kappa,theta): #Computes beta beta = theta-alpha(theta,kappa) return beta def theta(alpha, beta): #Computes beta theta = beta+alpha return beta def F(kappa, nt1,nt2, size, x0, y0, extra=100, x_shear = 0, y_shear = 0, alpha_x_in = [-99], alpha_y_in = [-99]): # Theta positions for each pixel in beta if np.sum(alpha_x_in) != [-99]: alpha_x = alpha_x_in alpha_y = alpha_y_in else: nk1,nk2 = np.shape(kappa) alpha_x,alpha_y = alpha_def(kappa,nt1,nt2,x0,y0,extra) alpha_x = alpha_x+x_shear alpha_y = alpha_y+y_shear na1,na2 = np.shape(alpha_x) xa,ya = np.where(np.zeros((na1,na2)) == 0) # xa = xa-(x0+extra/2) #ya = ya-(y0+extra/2) nb1=nt1*size nb2=nt2*size xb, yb = np.where(np.zeros((nb1,nb2))==0) #Scaling of the source grid #Scaling of the deflection grid xa = xa*(np.float(nt1)/np.float(na1))#-0.68 ya = ya*(np.float(nt2)/np.float(na2))#-0.68 #Setting images coordinates in 2d xa2d = np.reshape(xa,(na1,na2)) ya2d = np.reshape(ya,(na1,na2)) F2 = [] for i in range(np.size(xb)): #Deflection of photons emitted in xb[i],yb[i] theta_x = (xb[i])*(np.float(nt1)/np.float(nb1))+alpha_x theta_y = (yb[i])*(np.float(nt2)/np.float(nb2))+alpha_y #Matching of arrivals with pixels in image plane xprox = np.int_(np.abs((xa2d-theta_x)*2)) yprox = np.int_(np.abs((ya2d-theta_y)*2)) if np.min(xprox+yprox) <1.5: loc2 = np.where((xprox+yprox)==np.min(xprox+yprox))# else: loc2 = [] if (np.size(loc2)==0): F2.append([0]) else: F2.append((np.array(loc2))-1) return F2 def source_to_image(Source, nt1,nt2, theta, ones = 1): # Source: Image of the source in the source plane # n1,n2: size in pixels of the postage stamp in image plane # F: the coordinates of the lens mapping F = (theta) nb1,nb2 = np.shape(Source) if ones == 1: onelens = source_to_image(np.ones(Source.shape), nt1,nt2, theta, ones = 0) onelens[np.where(onelens==0)]=1 else: onelens = 1. Image = np.zeros((nt1,nt2)) xb,yb = np.where(np.zeros((nb1,nb2)) == 0) N = np.size(xb) k=0 for pos in F: if np.size(np.shape(pos)) != 1: Image[np.array(pos[0][np.where((pos[0][:]<nt1))]), np.array(pos[1][np.where(pos[1][:]<nt2)])] += Source[xb[k],yb[k]]#fullSource k=k+1 return Image/onelens def image_to_source(Image, size,beta,lensed = 0, square = 0): # Image: postagestamp of the observed image # nsize1,nsize2: size of the postagestamp in source plane # F: lens mapping matrix F = (beta) nt1,nt2 = np.shape(Image) nb1 = nt1*size nb2 = nt2*size Source = np.zeros((nb1,nb2)) xb,yb = np.where(Source == 0) N = np.size(xb) for k in range(N): pos = F[k] if np.size(np.shape(pos)) > 1: if np.sum(lensed) !=0: if square == 0: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])])/np.max([1,np.size(pos[0][:])]) else: Source[xb[k], yb[k]] += np.sum((Image[np.array(pos[0][:]), np.array(pos[1][:])] / np.max([1, np.size(pos[0][:])]))**2) else: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])]) if square == 1: Source = np.sqrt(Source) return Source def image_to_source_bound(Image, size,beta,lensed = 0): # Image: postagestamp of the observed image # nsize1,nsize2: size of the postagestamp in source plane # F: lens mapping matrix F = (beta) nt1,nt2 = np.shape(Image) nb1 = nt1*size nb2 = nt2*size Source = np.zeros((nb1,nb2)) xb,yb = np.where(Source == 0) N = np.size(xb) for k in range(N): pos = F[k] if np.size(np.shape(pos)) > 1: if np.sum(lensed) !=0: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])])/np.max([1,np.size(pos[0][:])]) else: Source[xb[k],yb[k]] += np.sum(Image[np.array(pos[0][:]), np.array(pos[1][:])]) return Source
7,136
28.987395
123
py
SLIT
SLIT-master/build/lib/SLIT/__init__.py
from Solve import * import Lens import wave_transform import tools
67
12.6
21
py
eEVM
eEVM-main/3rdparty/intx/test/fuzzer/decode.py
#!/usr/bin/env python3 import os import sys ops_filter = () ops = ('/', '*', '<<', '>>', '+', '-', 's/') def err(*args, **kwargs): print(*args, file=sys.stderr, **kwargs) def decode_file(file): with open(file, 'rb') as f: print("Decoding {}".format(file)) decode_data(f.read()) def decode_data(data): arg_size = (len(data) - 1) // 2 if arg_size not in (16, 32, 64): err("Incorrect argument size: {}".format(arg_size)) return op_index = int(data[0]) if op_index >= len(ops): return op = ops[op_index] if ops_filter and op not in ops_filter: return x = int.from_bytes(data[1:1 + arg_size], byteorder='big') y = int.from_bytes(data[1 + arg_size:], byteorder='big') print("argument size: {}".format(arg_size)) print(x, op, y) print(hex(x), op, hex(y)) if op in ('/', 's/'): print("Test:") print("{") print(" {}_u512,".format(hex(x))) print(" {}_u512,".format(hex(y))) print(" {}_u512,".format(hex(x // y))) print(" {}_u512,".format(hex(x % y))) print("},") if op == 's/': ax = (-x) % 2**512 ay = (-y) % 2**512 print("Test:") print("{") print(" {}_u512,".format(hex(ax))) print(" {}_u512,".format(hex(ay))) print(" {}_u512,".format(hex(ax // ay))) print(" {}_u512,".format(hex(ax % ay))) print("},") assert len(sys.argv) > 1 path = sys.argv[1] if (os.path.isfile(path)): decode_file(path) else: for root, _, files in os.walk(path): for file in files: decode_file(os.path.join(root, file))
1,690
22.486111
61
py
frodo
frodo-main/setup.py
from setuptools import setup, find_packages install_requires=[ 'flask', 'rdflib', 'nltk', 'shortuuid', 'fredclient @ git+https://github.com/anuzzolese/fredclient' ] setup(name='frodo', version='1.0.0', packages=find_packages(), install_requires=install_requires)
284
20.923077
64
py
frodo
frodo-main/demo/frodo_webapp.py
from flask import Flask, Response, render_template, request from frodo import Frodo import shortuuid import webapp_conf myapp = Flask(__name__) @myapp.route("/") def index(): text = request.args.get("text") ontology_id = request.args.get("ontology-id") if text: print(text) #namespace = ''.join([webapp_conf.NS, shortuuid.uuid(text)[:8], '/']) namespace = webapp_conf.NS frodo = Frodo( namespace=namespace, fred_uri=webapp_conf.FRED_ENDPOINT, fred_key=webapp_conf.FRED_KEY ) ontology = frodo.generate(text, ontology_id) return Response(ontology.serialize(format='text/turtle'), mimetype='text/turtle', headers={"Content-disposition": "attachment; filename=ontology.ttl"}) else: return render_template("index.html", basepath=webapp_conf.BASEPATH) if __name__ == '__main__': myapp.run(port=webapp_conf.PORT)
972
28.484848
77
py
frodo
frodo-main/demo/webapp_conf.py
import shortuuid FRED_ENDPOINT = 'http://wit.istc.cnr.it/stlab-tools/fred' FRED_KEY = '' NS = 'https://w3id.org/stlab/ontology/' PORT = 5555 BASEPATH = './' shortuuid.set_alphabet("0123456789abcdefghijkmnopqrstuvwxyz")
220
23.555556
61
py
frodo
frodo-main/demo/demo.py
from frodo import Frodo import shortuuid import webapp_conf #shortuuid.set_alphabet("0123456789abcdefghijkmnopqrstuvwxyz") sentence = 'What cars cost more than $50,000?' namespace = ''.join([webapp_conf.NS, shortuuid.uuid(sentence)[:8], '/']) frodo = Frodo( namespace=namespace, fred_uri=webapp_conf.FRED_ENDPOINT, fred_key=webapp_conf.FRED_KEY ) #sentence = 'What are the contaminated sites in the area of Pavia recorded in 2020?' #sentence = 'Who well commissioned a cultural property at a certain time?' ontology = frodo.generate(sentence) print(ontology.serialize(format='text/turtle'))
609
26.727273
84
py
frodo
frodo-main/frodo/taxonomic_queries.py
update = { 'alias_instance': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?alias_instance a fred:Alias;\n fred:aliasOf ?instance.\n\n ?s ?p ?o. # remove all statements about the fred:Alias class\n}\nINSERT {\n ?alias_instance owl:sameAs ?instance.\n}\nWHERE {\n OPTIONAL {\n ?alias_instance a fred:Alias;\n fred:aliasOf ?instance.\n\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?alias_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Alias || ?p = fred:aliasOf || ?o = fred:Alias).\n}\n', 'declaration': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX boxing: <http://www.ontologydesignpatterns.org/ont/boxer/boxing.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance ?p ?o.\n}\nINSERT {\n [ a ?sub_class].\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n # OPTIONAL {\n ?instance a fred:Topic;\n boxing:declaration/rdf:type ?super_class.\n\n ?instance ?p ?o.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n BIND(?instance AS ?sub_class).\n # }\n}\n', 'genre_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Genre;\n fred:genreOf ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?instance a ?super_class.\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Genre.\n\n OPTIONAL {\n ?instance a ?sub_class.\n FILTER(?sub_class != fred:Genre).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n }\n\n ?instance fred:genreOf ?super_class_instance.\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n OPTIONAL {\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n }\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Genre || ?p = fred:genreOf || ?o = fred:Genre).\n}\n', 'kind_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Kind;\n fred:kindOf ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?instance a ?super_class.\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Kind.\n\n OPTIONAL {\n ?instance a ?sub_class.\n FILTER(?sub_class != fred:Kind).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n }\n\n ?instance fred:kindOf ?super_class_instance.\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n OPTIONAL {\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n }\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Kind || ?p = fred:kindOf || ?o = fred:Kind).\n}\n', 'name_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Name;\n fred:nameOf ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?instance a ?super_class.\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Name.\n\n OPTIONAL {\n ?instance a ?sub_class.\n FILTER(?sub_class != fred:Name).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n }\n\n ?instance fred:nameOf ?super_class_instance.\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n OPTIONAL {\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n }\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Name || ?p = fred:nameOf || ?o = fred:Name).\n}\n', 'quantity_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Quantity;\n ?typeOf_property ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Quantity, ?sub_class.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n\n ?instance ?typeOf_property ?super_class_instance.\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?typeOf_property), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(?typeOf_property = IRI(CONCAT("http://www.ontologydesignpatterns.org/ont/fred/", LCASE(REPLACE(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/", "")), "Of"))).\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Quantity || ?o = fred:Quantity).\n}\n', 'series_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Series;\n ?typeOf_property ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Series, ?sub_class.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n\n ?instance ?typeOf_property ?super_class_instance.\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?typeOf_property), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(?typeOf_property = IRI(CONCAT("http://www.ontologydesignpatterns.org/ont/fred/", LCASE(REPLACE(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/", "")), "Of"))).\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Series || ?o = fred:Series).\n}\n', 'species_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Species;\n ?typeOf_property ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Species, ?sub_class.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n\n ?instance ?typeOf_property ?super_class_instance.\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?typeOf_property), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(?typeOf_property = IRI(CONCAT("http://www.ontologydesignpatterns.org/ont/fred/", LCASE(REPLACE(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/", "")), "Of"))).\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Species || ?o = fred:Species).\n}\n', 'type_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance ?typeOf_property ?super_class_instance.\n}\nINSERT {\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n # OPTIONAL {\n ?instance a ?sub_class.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n\n ?instance ?typeOf_property ?super_class_instance.\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?typeOf_property), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(?typeOf_property = IRI(CONCAT("http://www.ontologydesignpatterns.org/ont/fred/", LCASE(REPLACE(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/", "")), "Of"))).\n # }\n\n}\n', 'variety_of': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\nPREFIX fred: <http://www.ontologydesignpatterns.org/ont/fred/>\n\nDELETE {\n ?instance a fred:Variety;\n fred:varietyOf ?super_class_instance.\n\n ?s ?p ?o.\n}\nINSERT {\n ?instance a ?super_class.\n ?sub_class rdfs:subClassOf ?super_class.\n}\nWHERE {\n OPTIONAL {\n ?instance a fred:Variety.\n\n OPTIONAL {\n ?instance a ?sub_class.\n FILTER(?sub_class != fred:Variety).\n FILTER(STRSTARTS(STR(?sub_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n MINUS {\n ?sub_class rdfs:subClassOf+ dul:Event.\n }\n }\n\n ?instance fred:varietyOf ?super_class_instance.\n FILTER(STRSTARTS(STR(?super_class_instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n OPTIONAL {\n ?super_class_instance a ?super_class.\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n }\n }\n\n # Clean up the knowledge graph\n ?s ?p ?o.\n FILTER(?s = fred:Variety || ?p = fred:varietyOf || ?o = fred:Variety).\n}\n', } construct = { 'fred_taxonomy': 'PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#>\n\nCONSTRUCT {\n ?domain_class_FrODO a owl:Class;\n rdfs:label ?domain_class_label;\n rdfs:subClassOf ?super_class_FrODO;\n owl:equivalentClass ?equiv_class_FrODO.\n\n ?super_class_FrODO a owl:Class;\n rdfs:label ?super_class_label;\n rdfs:subClassOf ?ancestor_class_FrODO.\n\n ?equiv_class_FrODO a owl:Class;\n rdfs:label ?equiv_class_label.\n\n ?ancestor_class_FrODO a owl:Class;\n rdfs:label ?ancestor_class_label.\n}\nWHERE {\n ?instance a ?domain_class.\n FILTER(!STRENDS(STR(?domain_class), "Disjunct")). # remove this when the disjunction pattern is implemented\n FILTER(STRSTARTS(STR(?domain_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?domain_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?domain_class_label).\n BIND(IRI(REPLACE(STR(?domain_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?domain_class_FrODO).\n\n MINUS {\n ?domain_class rdfs:subClassOf+ dul:Event.\n }\n\n OPTIONAL {\n ?domain_class rdfs:subClassOf+ ?super_class.\n FILTER(STRSTARTS(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?super_class_label).\n BIND(IRI(REPLACE(STR(?super_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?super_class_FrODO).\n }\n\n OPTIONAL {\n ?domain_class owl:equivalentClass+ ?equiv_class.\n FILTER(STRSTARTS(STR(?equiv_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?equiv_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?equiv_class_label).\n BIND(IRI(REPLACE(STR(?equiv_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?equiv_class_FrODO).\n }\n\n OPTIONAL {\n ?super_class rdfs:subClassOf+ ?ancestor_class.\n FILTER(STRSTARTS(STR(?ancestor_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?ancestor_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?ancestor_class_label).\n BIND(IRI(REPLACE(STR(?ancestor_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?ancestor_class_FrODO).\n }\n}\n', 'sameas_instance': 'CONSTRUCT {\n ?instance_FrODO a owl:NamedIndividual, ?instance_class_FrODO;\n rdfs:label ?instance_label;\n owl:sameAs ?alias_FrODO.\n\n ?alias_FrODO a owl:NamedIndividual, ?alias_class_FrODO;\n rdfs:label ?alias_label;\n owl:sameAs ?instance_FrODO.\n\n ?instance_class_FrODO a owl:Class;\n rdfs:label ?instance_class_label.\n\n ?alias_class_FrODO a owl:Class;\n rdfs:label ?alias_class_label.\n}\nWHERE {\n ?alias owl:sameAs ?instance.\n FILTER(STRSTARTS(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/")).\n FILTER(STRSTARTS(STR(?alias), "http://www.ontologydesignpatterns.org/ont/fred/")).\n\n BIND(STRLANG(REPLACE(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?instance_label).\n BIND(IRI(REPLACE(STR(?instance), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?instance_FrODO).\n\n BIND(STRLANG(REPLACE(STR(?alias), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?alias_label).\n BIND(IRI(REPLACE(STR(?alias), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?alias_FrODO).\n\n OPTIONAL {\n ?instance a ?instance_class.\n FILTER(STRSTARTS(STR(?instance_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?instance_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?instance_class_label).\n BIND(IRI(REPLACE(STR(?instance_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?instance_class_FrODO).\n }\n\n OPTIONAL {\n ?alias a ?alias_class.\n FILTER(STRSTARTS(STR(?alias_class), "http://www.ontologydesignpatterns.org/ont/fred/")).\n BIND(STRLANG(REPLACE(STR(?alias_class), "http://www.ontologydesignpatterns.org/ont/fred/", ""), "en") AS ?alias_class_label).\n BIND(IRI(REPLACE(STR(?alias_class), "http://www.ontologydesignpatterns.org/ont/fred/", "https://w3id.org/stlab/ontology/")) AS ?alias_class_FrODO).\n }\n}\n', }
16,514
916.5
2,617
py
frodo
frodo-main/frodo/frodo.py
import os import re from fredclient import FREDClient, FREDParameters, FREDDefaults import nltk from nltk.stem import WordNetLemmatizer from rdflib import RDFS, RDF, OWL, XSD, URIRef, Literal, BNode, Graph, Namespace from rdflib.paths import evalPath, OneOrMore from abc import ABC, abstractmethod from typing import List, Dict, NoReturn, Tuple from rdflib.term import URIRef import frodo.taxonomic_queries as taxonomic_queries nltk.download('wordnet') class MorphUtils: LEMMATIZER: WordNetLemmatizer = WordNetLemmatizer() @staticmethod def get_namespace(uriref: URIRef) -> str: uriref_str = str(uriref) last_hash = uriref_str.rfind('#') last_slash = uriref_str.rfind('/') return uriref_str[:last_hash+1] if last_hash > last_slash else uriref_str[:last_slash+1] @staticmethod def get_id(uriref: URIRef) -> str: uriref_str = str(uriref) last_hash = uriref_str.rfind('#') last_slash = uriref_str.rfind('/') return uriref_str[last_hash+1:] if last_hash > last_slash else uriref_str[last_slash+1:] @staticmethod def labelize_uriref(uriref: URIRef, lang: str = None, datatype: URIRef = None) -> Literal: ns = MorphUtils.get_namespace(uriref) uriref_str = str(uriref) term_id = uriref_str.replace(ns, '') label = term_id[0:1] + re.sub('([A-Z]{1})', r' \1', term_id[1:]).lower() return Literal(label, lang) if lang else Literal(label, datatype) if datatype else Literal(label) @staticmethod def migrate_taxonomy(g: Graph, source_cls: URIRef, target_cls: URIRef, gerundify: bool = False, predicate: URIRef = RDFS.subClassOf) -> NoReturn: ontology = Graph() ns = MorphUtils.get_namespace(target_cls) for subclass in g.objects(source_cls, predicate): if str(subclass).startswith(FREDDefaults.DEFAULT_FRED_NAMESPACE): if gerundify: sc = URIRef(ns + MorphUtils.gerundify(subclass)) else: sc = URIRef(ns + MorphUtils.get_id(subclass)) ontology.add((target_cls, RDFS.subClassOf, sc)) ontology.add((sc, RDF.type, OWL.Class)) label = MorphUtils.labelize_uriref(sc, 'en') ontology.add((sc, RDFS.label, label)) ontology += MorphUtils.migrate_taxonomy(g, subclass, sc, gerundify) return ontology @staticmethod def inverse(predicate: URIRef) -> URIRef: ns = MorphUtils.get_namespace(predicate) predicate_id = predicate.replace(ns, '') add_of = False if predicate_id.startswith('involves'): predicate_id = predicate_id.replace('involves', '') + 'InvolvedIn' else: add_of = True inverse_predicate_id = 'is' + predicate_id.capitalize() if add_of: inverse_predicate_id += 'Of' return URIRef(ns + inverse_predicate_id) @staticmethod def gerundify(term: URIRef) -> Graph: class_label = MorphUtils.get_id(term) class_label_terms = class_label[0:1] + re.sub('([A-Z]{1})', r' \1', class_label[1:]) class_label_terms = class_label_terms.split() lemma = MorphUtils.LEMMATIZER.lemmatize(class_label_terms[-1], 'v') isVowel = lambda c: c.lower() in 'aeiou' isConsonant = lambda c: c.lower() not in 'aeiou' if len(lemma) >= 3 and \ isVowel(lemma[-3]) and isVowel(lemma[-2]) and isConsonant(lemma[-1]): # e.g. 'speak' -> 'speaking' gerundive = f'{lemma}ing' elif len(lemma) >= 3 and \ isConsonant(lemma[-3]) and isVowel(lemma[-2]) and isConsonant(lemma[-1]): # TODO: Verbs of this kind whose last syllable is not stressed are not # subject to a doubling of the final consonant! # (e.g. 'render' -> 'rendering') # e.g. 'run' -> 'running' gerundive = f'{lemma}{lemma[-1]}ing' elif lemma.endswith('ic'): # e.g. panic -> panicking gerundive = f'{lemma}king' elif lemma.endswith('ee'): # e.g. 'see' -> 'seeing' gerundive = f'{lemma}ing' elif lemma.endswith('ie'): # e.g. 'die' -> 'dying' gerundive = f'{lemma[:-2]}ying' elif lemma.endswith('e'): # e.g. 'make' -> 'making' gerundive = f'{lemma[:-1]}ing' else: # e.g. 'study' -> 'studying' gerundive = f'{lemma}ing' gerundive = gerundive.capitalize() return "".join(class_label_terms[:-1]) + gerundive class MorphismI(ABC): def __init__(self, ns: Namespace): self._ns = ns @abstractmethod def morph(self, g: Graph) -> Graph: pass class TaxonomyMorphism(MorphismI): def morph(self, g: Graph) -> Graph: ontology = Graph() ontology.bind('owl', OWL) ontology.bind('rdf', RDF) ontology.bind('rdf', RDFS) # Apply UPDATE queries to convert graph patterns to taxonomy axioms for update_query in taxonomic_queries.update: g = self._update_query(g, update_query) # Apply CONSTRUCT queries to generate the ontology draft for construct_query in taxonomic_queries.construct: ontology += self._construct_query(g, construct_query) return g, ontology def _update_query(self, g: Graph, query_name: str): sparql = taxonomic_queries.update[query_name] sparql = sparql.replace('$FRED_NS', FREDDefaults.DEFAULT_FRED_NAMESPACE) g.update(sparql) return g def _construct_query(self, g: Graph, query_name: str): sparql = taxonomic_queries.construct[query_name] sparql = sparql.replace('$FRED_NS', FREDDefaults.DEFAULT_FRED_NAMESPACE) sparql = sparql.replace('$FrODO_NS', self._ns) # ontology namespace sparql = sparql.replace('$LANG', 'en') # language for labels result = g.query(sparql) return result.graph class BinaryRelationMorphism(MorphismI): def morph(self, g: Graph) -> Graph: ontology = Graph() sparql = f''' PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#> PREFIX owl: <{OWL._NS}> SELECT ?subj ?subjtype ?subjsameastype ?rel ?obj ?objtype ?objsameastype WHERE{{ ?subj ?rel ?obj OPTIONAL{{ ?subj a ?subjtype FILTER(REGEX(STR(?subjtype), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}')) }} OPTIONAL{{ ?obj a ?objtype FILTER(REGEX(STR(?objtype), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}')) }} OPTIONAL{{ ?subj owl:sameAs/rdf:type ?subjsameastype }} OPTIONAL{{ ?obj owl:sameAs/rdf:type ?objsameastype }} OPTIONAL{{ ?subjtype rdfs:subClassOf ?subtype . ?subtype rdfs:subClassOf* dul:Event}} FILTER(!BOUND(?subtype)) FILTER(REGEX(STR(?rel), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}')) }} ''' resultset = g.query(sparql) for row in resultset: subjtype = row.subjtype subjsameastype = row.subjsameastype binary_predicate = row.rel objtype = row.objtype objsameastype = row.objsameastype if subjtype: subj = subjtype elif subjsameastype: subj = subjsameastype else: subj = None if objtype: obj = objtype elif objsameastype: obj = objsameastype else: obj = None if subj and obj: subject_id = MorphUtils.get_id(subj) predicate_id = MorphUtils.get_id(binary_predicate) object_id = MorphUtils.get_id(obj) subject_class = URIRef(self._ns + subject_id) ontology.add((subject_class, RDF.type, OWL.Class)) ontology.add((subject_class, RDFS.label, MorphUtils.labelize_uriref(subject_class, 'en'))) ontology += MorphUtils.migrate_taxonomy(g, subj, subject_class) object_class = URIRef(self._ns + object_id) ontology.add((object_class, RDF.type, OWL.Class)) ontology.add((object_class, RDFS.label, MorphUtils.labelize_uriref(object_class, 'en'))) ontology += MorphUtils.migrate_taxonomy(g, obj, object_class) object_property = URIRef("".join([self._ns, predicate_id, object_id])) ontology.add((object_property, RDF.type, OWL.ObjectProperty)) ontology.add((object_property, RDFS.domain, OWL.Thing)) ontology.add((object_property, RDFS.range, object_class)) ontology.add((object_property, RDFS.label, MorphUtils.labelize_uriref(object_property, 'en'))) inverse_object_property = MorphUtils.inverse(object_property) ontology.add((inverse_object_property, RDF.type, OWL.ObjectProperty)) ontology.add((inverse_object_property, RDFS.domain, object_class)) ontology.add((inverse_object_property, RDFS.range, OWL.Thing)) ontology.add((inverse_object_property, RDFS.label, MorphUtils.labelize_uriref(inverse_object_property, 'en'))) restriction = BNode() ontology.add((restriction, RDF.type, OWL.Restriction)) ontology.add((restriction, OWL.onProperty, object_property)) ontology.add((restriction, OWL.someValuesFrom, object_class)) ontology.add((subject_class, RDFS.subClassOf, restriction)) return g, ontology class RoleType: PASSIVE = 'PASSIVE' AGENTIVE = 'AGENTIVE' CONDITIONAL_AGENTIVE = 'CONDITIONAL_AGENTIVE' OBLIQUE = 'OBLIQUE' FRED_ROLE = 'FRED_ROLE' @staticmethod def get_role_type(role: URIRef): role = str(role) role_type = None if role in FREDDefaults.ROLES['passive']: role_type = RoleType.PASSIVE elif role in FREDDefaults.ROLES['agentive']: role_type = RoleType.AGENTIVE elif role in FREDDefaults.ROLES['conditional_agentive']: role_type = RoleType.CONDITIONAL_AGENTIVE elif role in FREDDefaults.ROLES['oblique']: role_type = RoleType.OBLIQUE elif role.startswith(FREDDefaults.DEFAULT_FRED_NAMESPACE): role_type = RoleType.FRED_ROLE return role_type class RoleMap: def __init__(self, role: URIRef, ontology_class: URIRef, fred_class: URIRef, role_type: str): self.__role = role self.__ontology_class = ontology_class self.__fred_class = fred_class self.__role_type = role_type def get_role(self) -> URIRef: return self.__role def get_ontology_class(self) -> URIRef: return self.__ontology_class def get_fred_class(self) -> URIRef: return self.__fred_class def get_role_type(self) -> str: return self.__role_type class SituationDigest: def __init__(self, source_graph: Graph, situation: URIRef, situation_type: URIRef): self.__source_graph = source_graph self.__situation = situation self.__situation_type = situation_type self.__passive_roles: List[RoleMap] = [] self.__agentive_roles: List[RoleMap] = [] self.__conditional_agentive_roles: List[RoleMap] = [] self.__oblique_roles: List[RoleMap] = [] self.__fred_roles: List[RoleMap] = [] self.__class_label: str = None def get_source_graph(self) -> Graph: return self.__source_graph def get_situation(self) -> URIRef: return self.__situation def get_situation_type(self) -> URIRef: return self.__situation_type def add_role_map(self, participant: URIRef, role_type: str): if role_type == RoleType.AGENTIVE: self.__agentive_roles.append(participant) elif role_type == RoleType.PASSIVE: self.__passive_roles.append(participant) elif role_type == RoleType.CONDITIONAL_AGENTIVE: self.__conditional_agentive_roles.append(participant) elif role_type == RoleType.OBLIQUE: self.__oblique_roles.append(participant) elif role_type == RoleType.FRED_ROLE: self.__fred_roles.append(participant) def get_participants_of_role_type(self, role_type: str) -> str: participants = None if role_type == RoleType.AGENTIVE: participants = self.__agentive_roles elif role_type == RoleType.PASSIVE: participants = self.__passive_roles elif role_type == RoleType.CONDITIONAL_AGENTIVE: participants = self.__conditional_agentive_roles elif role_type == RoleType.OBLIQUE: participants = self.__oblique_roles elif role_type == RoleType.FRED_ROLE: participants = self.__fred_roles return participants def get_class_label(self) -> str: return self.__class_label def set_class_label(self, class_label: str): self.__class_label = class_label def update_class_label(self, class_label: str): if not self.__class_label: self.__class_label = '' self.__class_label = class_label + self.__class_label def formalise(self, namespace: str) -> Graph: ontology = Graph() class_iri = URIRef(namespace+self.__class_label) ontology.add((class_iri, RDF.type, OWL.Class)) ontology.add((class_iri, RDFS.label, MorphUtils.labelize_uriref(class_iri, 'en'))) ontology += MorphUtils.migrate_taxonomy(self.__source_graph, self.__situation, class_iri, True, RDF.type) ''' fred_situation_type = self.__situation_type gerund = MorphUtils.gerundify(fred_situation_type) gerundive_res = URIRef("".join([namespace, gerund])) ontology.add((gerundive_res, RDF.type, OWL.Class)) ontology.add((gerundive_res, RDFS.label, MorphUtils.labelize_uriref(gerundive_res))) ontology.add((class_iri, RDFS.subClassOf, gerundive_res)) ''' role_maps: List[RoleMap] = [*self.__agentive_roles, *self.__passive_roles, *self.__conditional_agentive_roles, *self.__oblique_roles, *self.__fred_roles] for role_map in role_maps: role_predicate = role_map.get_role() role_actor = role_map.get_ontology_class() fred_class = role_map.get_fred_class() role_type = role_map.get_role_type() ''' if fred_class == OWL.Thing: role_actor = URIRef(''.join([namespace, MorphUtils.get_id(role_predicate)])) ''' role_actor_iri = str(role_actor) if role_type == RoleType.FRED_ROLE: role_id = str(role_actor_iri).replace(namespace, '').replace(FREDDefaults.DEFAULT_FRED_NAMESPACE, '') predicate_id = str(role_predicate).replace(namespace, '').replace(FREDDefaults.DEFAULT_FRED_NAMESPACE, '') predicate = URIRef(''.join([namespace, predicate_id, role_id])) else: predicate_id = str(role_actor_iri).replace(namespace, '').replace(FREDDefaults.DEFAULT_FRED_NAMESPACE, '') predicate = URIRef(''.join([namespace, 'involves', predicate_id])) ontology.add((predicate, RDF.type, OWL.ObjectProperty)) ontology.add((predicate, RDFS.label, MorphUtils.labelize_uriref(predicate))) restriction = BNode() ontology.add((restriction, RDF.type, OWL.Restriction)) ontology.add((restriction, OWL.onProperty, predicate)) ontology.add((restriction, OWL.someValuesFrom, role_actor)) ontology.add((class_iri, RDFS.subClassOf, restriction)) ontology.add((predicate, RDFS.domain, OWL.Thing)) ontology.add((predicate, RDFS.range, role_actor)) # Inverse predicate inverse_predicate = MorphUtils.inverse(predicate) ontology.add((inverse_predicate, RDF.type, OWL.ObjectProperty)) ontology.add((inverse_predicate, RDFS.label, MorphUtils.labelize_uriref(inverse_predicate))) ontology.add((inverse_predicate, RDFS.domain, role_actor)) ontology.add((inverse_predicate, RDFS.range, OWL.Thing)) ontology.add((inverse_predicate, OWL.inverseOf, predicate)) ontology.add((predicate, OWL.inverseOf, inverse_predicate)) ontology.add((role_actor, RDF.type, OWL.Class)) ontology.add((role_actor, RDFS.label, MorphUtils.labelize_uriref(role_actor))) ontology += MorphUtils.migrate_taxonomy(self.__source_graph, fred_class, role_actor) return ontology class NAryRelationMorphism(MorphismI): def __init__(self, ns): super().__init__(ns) self.__lemmatizer: WordNetLemmatizer = WordNetLemmatizer() def morph(self, g: Graph) -> Graph: ontology = Graph() ''' We first detect frame occurrences for the base case. Frame occurrences are detected by querying the graph with the following property path: ?x rdf:type/rdfs:subClassOf* dul:Event ''' sparql = f''' PREFIX dul: <http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#> PREFIX owl: <{OWL._NS}> PREFIX xsd: <{XSD._NS}> SELECT ?situation ?situationtype ?role ?participant ?participanttype ?participantsameastype WHERE{{ ?situation a ?situationtype ; ?role ?participant . OPTIONAL{{ ?participant a ?participanttype FILTER(REGEX(STR(?participanttype), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}') || ?participanttype = owl:Thing) }} OPTIONAL{{ {{?participant owl:sameAs/a ?participantsameastype}} UNION {{ ?situation ?role2 ?participant2 FILTER(datatype(?participant2) = xsd:date || datatype(?participant2) = xsd:dateTime) BIND(dul:Time AS ?participantsameastype) }} }} ?situationtype rdfs:subClassOf+ dul:Event FILTER(REGEX(STR(?situation), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}')) FILTER(REGEX(STR(?situationtype), '^{FREDDefaults.DEFAULT_FRED_NAMESPACE}')) FILTER(REGEX(STR(?role), '^http://www.ontologydesignpatterns.org/ont/') ) }} ''' resultset = g.query(sparql) #paths = evalPath(g, (None, RDF.type/(RDFS.subClassOf*OneOrMore), URIRef('http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#Event'))) #for path in paths: situations_registry: Dict[URIRef, SituationDigest] = dict() for row in resultset: situation = row.situation situation_type = row.situationtype role = row.role role_type = RoleType.get_role_type(role) if row.participanttype == OWL.Thing: if row.participantsameastype: participant_type = row.participantsameastype else: participant_type = role else: if row.participanttype: participant_type = row.participanttype elif row.participantsameastype: participant_type = row.participantsameastype else: participant_type = None if participant_type and role_type: if situation in situations_registry: situation_digest = situations_registry[situation] else: situation_digest = SituationDigest(g, situation, situation_type) situations_registry.update({situation: situation_digest}) situation_digest.set_class_label(MorphUtils.gerundify(situation_type)) local_class_id = MorphUtils.get_id(participant_type) ontology_class = URIRef(self._ns + local_class_id) role_map = RoleMap(role, ontology_class, participant_type, role_type) situation_digest.add_role_map(role_map, role_type) if role_type == RoleType.PASSIVE: situation_digest.update_class_label(local_class_id) for situation, digest in situations_registry.items(): ontology += digest.formalise(self._ns) return g, ontology class Frodo: def __init__(self, namespace: str, fred_uri: str, fred_key: str, morphisms: Tuple[MorphismI] = None): self.__g: Graph = None self.__ns: Namespace = Namespace(namespace) self.__fred_uri = fred_uri self.__fred_key = fred_key defaultMorphisms = ( TaxonomyMorphism(self.__ns), BinaryRelationMorphism(self.__ns), NAryRelationMorphism(self.__ns) ) self.__morphisms: Tuple[MorphismI] = morphisms if morphisms else defaultMorphisms def get_namespace(self) -> str: return self.__ns def set_namespace(self, namespace: str): self.__ns = namespace def generate(self, cq: str, ontology_id: str = None) -> Graph: if not ontology_id: ontology_id = self.__ns fredclient = FREDClient(self.__fred_uri, key=self.__fred_key) self.__g = fredclient.execute_request(cq, FREDParameters(semantic_subgraph=True)) ontology = Graph() ontology.add((URIRef(ontology_id), RDF.type, OWL.Ontology)) ontology.bind("owl", Namespace('http://www.w3.org/2002/07/owl#')) ontology.bind("rdf", Namespace('http://www.w3.org/1999/02/22-rdf-syntax-ns#')) ontology.bind("rdfs", Namespace('http://www.w3.org/2000/01/rdf-schema#')) ontology.bind("", Namespace(self.__ns)) for morphism in self.__morphisms: self.__g, new_axioms = morphism.morph(self.__g) ontology += new_axioms triples_to_del = [] for s, _, o in ontology.triples((None, RDFS.subClassOf, None)): if s == o: triples_to_del.append((s, RDFS.subClassOf, o)) for triple in triples_to_del: ontology.remove(triple) return ontology
22,735
37.601019
161
py
frodo
frodo-main/frodo/__init__.py
from frodo.frodo import *
26
12.5
25
py
minkasi
minkasi-master/setup.py
from setuptools import Extension, setup import ctypes import subprocess import os try: mylib=ctypes.cdll.LoadLibrary("libminkasi.so") except OSError: os.environ["prefix"] = "minkasi" subprocess.check_call(["make", "-e", "libminkasi"]) try: mylib=ctypes.cdll.LoadLibrary("libmkfftw.so") except OSError: os.environ["prefix"] = "minkasi" subprocess.check_call(["make", "-e", "libmkfftw"]) setup( name='minkasi', version='1.0.0', install_requires=[ 'requests', 'importlib-metadata', 'numpy', 'astropy', 'scipy' ], packages=['minkasi'], package_data={"minkasi": ["*.so"]}, )
642
20.433333
55
py
minkasi
minkasi-master/examples/minkasi_mpi_example.py
import numpy from matplotlib import pyplot as plt import minkasi,pyfftw import time import glob reload(minkasi) plt.ion() #find tod files we want to map dir='../data/m0717_raw/' tod_names=glob.glob(dir+'*.fits') #if running MPI, you would want to split up files between processes #one easy way is to say to this: tod_names=tod_names[minkasi.myrank::minkasi.nproc] #NB - minkasi checks to see if MPI is around, if not #it sets rank to 0 an nproc to 1, so this would still #run in a non-MPI environment todvec=minkasi.TodVec() #loop over each file, and read it. for fname in tod_names: t1=time.time() dat=minkasi.read_tod_from_fits(fname) t2=time.time() minkasi.truncate_tod(dat) #truncate_tod chops samples from the end to make #the length happy for ffts minkasi.downsample_tod(dat) #sometimes we have faster sampled data than we need. #this fixes that. You don't need to, though. minkasi.truncate_tod(dat) #since our length changed, make sure we have a happy length #figure out a guess at common mode #and (assumed) linear detector drifts/offset #drifts/offsets are removed, which is important for mode finding. CM is *not* removed. dd=minkasi.fit_cm_plus_poly(dat['dat_calib']) dat['dat_calib']=dd t3=time.time() tod=minkasi.Tod(dat) todvec.add_tod(tod) print('took ',t2-t1,' ',t3-t2,' seconds to read and downsample file ',fname) #make a template map with desired pixel size an limits that cover the data #todvec.lims() is MPI-aware and will return global limits, not just #the ones from private TODs lims=todvec.lims() pixsize=3.0/3600*numpy.pi/180 map=minkasi.SkyMap(lims,pixsize) #once we have a map, we can figure out the pixellization of the data. Save that #so we don't have to recompute. Also calculate a noise model. The one here #(and currently the only supported one) is to rotate the data into SVD space, then #smooth the power spectrum of each mode. Other models would not be hard #to implement. The smoothing could do with a bit of tuning as well. for tod in todvec.tods: ipix=map.get_pix(tod) tod.info['ipix']=ipix tod.set_noise_smoothed_svd() #get the hit count map. We use this as a preconditioner #which helps small-scale convergence quite a bit. hits=minkasi.make_hits(todvec,map) #setup the mapset. In general this can have many things #in addition to map(s) of the sky, but for now we'll just #use a single skymap. mapset=minkasi.Mapset() mapset.add_map(map) #make A^T N^1 d. TODs need to understand what to do with maps #but maps don't necessarily need to understand what to do with TODs, #hence putting make_rhs in the vector of TODs. #Again, make_rhs is MPI-aware, so this should do the right thing #if you run with many processes. rhs=mapset.copy() todvec.make_rhs(rhs) #this is our starting guess. Default to starting at 0, #but you could start with a better guess if you have one. x0=rhs.copy() x0.clear() #preconditioner is 1/ hit count map. helps a lot for #convergence. precon=mapset.copy() tmp=hits.map.copy() ii=tmp>0 tmp[ii]=1.0/tmp[ii] #precon.maps[0].map[:]=numpy.sqrt(tmp) precon.maps[0].map[:]=tmp[:] #run PCG! mapset_out=minkasi.run_pcg(rhs,x0,todvec,precon,maxiter=50) if minkasi.myrank==0: mapset_out.maps[0].write('first_map_precon_mpi.fits') #and write out the map as a FITS file else: print('not writing map on process ',minkasi.myrank) #if you wanted to run another round of PCG starting from the previous solution, #you could, replacing x0 with mapset_out. #mapset_out2=minkasi.run_pcg(rhs,mapset_out,todvec,mapset,maxiter=50) #mapset_out2.maps[0].write('second_map.fits')
3,708
34.32381
95
py
minkasi
minkasi-master/examples/tsBowl_fitter.py
import minkasi import numpy as np from matplotlib import pyplot as plt import time fname = '/scratch/r/rbond/jorlo/MS0735/TS_EaCMS0f0_51_5_Oct_2021/Signal_TOD-AGBT19A_092_08-s26.fits' dat = minkasi.read_tod_from_fits(fname) minkasi.truncate_tod(dat) # figure out a guess at common mode and (assumed) linear detector drifts/offset # drifts/offsets are removed, which is important for mode finding. CM is *not* removed. dd, pred2, cm = minkasi.fit_cm_plus_poly(dat["dat_calib"], cm_ord=3, full_out=True) dat["dat_calib"] = dd dat["pred2"] = pred2 dat["cm"] = cm flatten = True if flatten: dat['dat_calib'] -= pred2 tod = minkasi.Tod(dat) tod.set_apix() todvec=minkasi.TodVec() todvec.add_tod(tod) for tod in todvec.tods: tod.set_noise(minkasi.NoiseSmoothedSVD) tsBowl = minkasi.tsBowl(tod) mapset=minkasi.Mapset() mapset.add_map(tsBowl) rhs=mapset.copy() todvec.make_rhs(rhs) x0=rhs.copy() x0.clear() t1=time.time() mapset_out=minkasi.run_pcg(rhs,x0,todvec,maxiter=50) t2 = time.time() print('Took {} sec to fit 1 tod'.format(t2-t1)) print(mapset_out.maps[0].params[0]) plt.plot(tod.info['apix'][0], tod.info['dat_calib'][0]) plt.plot(tod.info['apix'][0], np.dot(mapset_out.maps[0].params[0], mapset_out.maps[0].vecs[0].T)) plt.ylim(-0.05, 0.05) plt.xlabel('apix') if flatten: plt.ylabel('dat_calib - pred2') else: plt.ylabel('dat_calib') plt.show()
1,384
20.984127
100
py
minkasi
minkasi-master/examples/zw3146_tsBowl.py
#This is a template script to show how to fit multi-component models #directly to timestreams. The initial part (where the TODs, noise model etc.) #are set up is the same as general mapping scripts, although we don't #need to bother with setting a map/pixellization in general (although if #your timestream model used a map you might). This script works under #MPI as well. import numpy as np from matplotlib import pyplot as plt import minkasi import time import glob from importlib import reload reload(minkasi) plt.ion() #find tod files we want to map outroot='maps/zw3146/zw3146' dir='../data/Zw3146/' tod_names=glob.glob(dir+'Sig*.fits') tod_names.sort() #make sure everyone agrees on the order of the file names tod_names=tod_names[minkasi.myrank::minkasi.nproc] todvec=minkasi.TodVec() #loop over each file, and read it. for fname in tod_names: t1=time.time() dat=minkasi.read_tod_from_fits(fname) t2=time.time() minkasi.truncate_tod(dat) #truncate_tod chops samples from the end to make dd=minkasi.fit_cm_plus_poly(dat['dat_calib']) dat['dat_calib']=dd t3=time.time() tod=minkasi.Tod(dat) todvec.add_tod(tod) print('took ',t2-t1,' ',t3-t2,' seconds to read and downsample file ',fname) minkasi.barrier() lims=todvec.lims() pixsize=2.0/3600*numpy.pi/180 map=minkasi.SkyMap(lims,pixsize) for tod in todvec.tods: ipix=map.get_pix(tod) #don't need pixellization for a timestream fit... tod.info['ipix']=ipix tod.set_noise_smoothed_svd() tod.set_noise(minkasi.NoiseSmoothedSVD) tsVec = minkasi.tsModel(todvec = todvec, modelclass = minkasi.tsBowl) #We add two things for pcg to do simulatenously here: make the maps from the tods #and fit the polynomials to tsVec mapset.add_map(map) mapset.add_map(tsVec) hits=minkasi.make_hits(todvec,map) rhs=mapset.copy() todvec.make_rhs(rhs) x0=rhs.copy() x0.clear() #preconditioner is 1/ hit count map. helps a lot for #convergence. precon=mapset.copy() tmp=hits.map.copy() ii=tmp>0 tmp[ii]=1.0/tmp[ii] precon.maps[0].map[:]=tmp[:] #tsBowl precon is 1/todlength if len(mapset.maps) > 1: for key in precon.maps[1].data.keys(): temp = np.ones(precon.maps[1].data[key].params.shape) temp /= precon.maps[1].data[key].vecs.shape[1] precon.maps[1].data[key].params = temp mapset_out=minkasi.run_pcg(rhs,x0,todvec,precon,maxiter=50) if minkasi.myrank==0: if len(mapset.maps)==1: mapset_out.maps[0].write('/scratch/r/rbond/jorlo/noBowl_map_precon_mpi_py3.fits') #and write out the map as a FITS file else: mapset_out.maps[0].write('/scratch/r/rbond/jorlo/tsBowl_map_precon_mpi_py3.fits') else: print('not writing map on process ',minkasi.myrank) if minkasi.nproc>1: minkasi.MPI.Finalize() d2r=np.pi/180 sig=9/2.35/3600*d2r theta_0=40/3600*d2r beta_pars=np.asarray([155.91355*d2r,4.1877*d2r,theta_0,0.7,-8.2e-4]) src1_pars=np.asarray([155.9374*d2r,4.1775*d2r,3.1e-5,9.15e-4]) src2_pars=np.asarray([155.90447*d2r,4.1516*d2r,2.6e-5,5.1e-4]) pars=np.hstack([beta_pars,src1_pars,src2_pars]) #we need to combine parameters into a single vector npar=np.hstack([len(beta_pars),len(src1_pars),len(src2_pars)]) #and the fitter needs to know how many per function #this array of functions needs to return the model timestreams and derivatives w.r.t. parameters #of the timestreams. funs=[minkasi.derivs_from_isobeta_c,minkasi.derivs_from_gauss_c,minkasi.derivs_from_gauss_c] #we can keep some parameters fixed at their input values if so desired. to_fit=np.ones(len(pars),dtype='bool') to_fit[3]=False #Let's keep beta pegged to 0.7 t1=time.time() pars_fit,chisq,curve,errs=minkasi.fit_timestreams_with_derivs_manyfun(funs,pars,npar,todvec,to_fit) t2=time.time() if minkasi.myrank==0: print('No Sub: ') print('took ',t2-t1,' seconds to fit timestreams') for i in range(len(pars_fit)): print('parameter ',i,' is ',pars_fit[i],' with error ',errs[i]) minkasi.comm.barrier() for tod in todvec.tods(): todname = tod.info['fname'] tod.info['dat_calib'] -= np.dot(mapset_out.maps[1].data[fname].params, mapset_out[1].data[fname].vecs.T)
4,147
30.907692
127
py
minkasi
minkasi-master/examples/fit_zw3146_multi_component.py
#This is a template script to show how to fit multi-component models #directly to timestreams. The initial part (where the TODs, noise model etc.) #are set up is the same as general mapping scripts, although we don't #need to bother with setting a map/pixellization in general (although if #your timestream model used a map you might). This script works under #MPI as well. import numpy import numpy as np from matplotlib import pyplot as plt import minkasi import time import glob from importlib import reload reload(minkasi) plt.ion() #find tod files we want to map outroot='maps/zw3146/zw3146' dir='../data/Zw3146/' #dir='../data/moo1110/' #dir='../data/moo1046/' tod_names=glob.glob(dir+'Sig*.fits') tod_names.sort() #make sure everyone agrees on the order of the file names #tod_names=tod_names[:20] #you can cut the number of TODs here for testing purposes #if running MPI, you would want to split up files between processes #one easy way is to say to this: tod_names=tod_names[minkasi.myrank::minkasi.nproc] #NB - minkasi checks to see if MPI is around, if not #it sets rank to 0 an nproc to 1, so this would still #run in a non-MPI environment todvec=minkasi.TodVec() #loop over each file, and read it. for fname in tod_names: t1=time.time() dat=minkasi.read_tod_from_fits(fname) t2=time.time() minkasi.truncate_tod(dat) #truncate_tod chops samples from the end to make #the length happy for ffts #minkasi.downsample_tod(dat) #sometimes we have faster sampled data than we need. # #this fixes that. You don't need to, though. #minkasi.truncate_tod(dat) #since our length changed, make sure we have a happy length# # #figure out a guess at common mode #and (assumed) linear detector drifts/offset #drifts/offsets are removed, which is important for mode finding. CM is *not* removed. dd=minkasi.fit_cm_plus_poly(dat['dat_calib']) dat['dat_calib']=dd t3=time.time() tod=minkasi.Tod(dat) todvec.add_tod(tod) print('took ',t2-t1,' ',t3-t2,' seconds to read and downsample file ',fname) #make a template map with desired pixel size an limits that cover the data #todvec.lims() is MPI-aware and will return global limits, not just #the ones from private TODs lims=todvec.lims() pixsize=2.0/3600*numpy.pi/180 #map=minkasi.SkyMap(lims,pixsize) #we don't need a map when fitting timestreams #once we have a map, we can figure out the pixellization of the data. Save that #so we don't have to recompute. Also calculate a noise model. The one here #is to rotate the data into SVD space, then smooth the power spectrum of each mode. # The smoothing could do with a bit of tuning.. for tod in todvec.tods: #ipix=map.get_pix(tod) #don't need pixellization for a timestream fit... #tod.info['ipix']=ipix #tod.set_noise_smoothed_svd() tod.set_noise(minkasi.NoiseSmoothedSVD) #we need an initial guess since this fitting routine is #for nonlinear models. This guess came from looking #at a map/some initial fits. The better the guess, the #faster the convergence. d2r=np.pi/180 sig=9/2.35/3600*d2r theta_0=40/3600*d2r beta_pars=np.asarray([155.91355*d2r,4.1877*d2r,theta_0,0.7,-8.2e-4]) src1_pars=np.asarray([155.9374*d2r,4.1775*d2r,3.1e-5,9.15e-4]) src2_pars=np.asarray([155.90447*d2r,4.1516*d2r,2.6e-5,5.1e-4]) pars=np.hstack([beta_pars,src1_pars,src2_pars]) #we need to combine parameters into a single vector npar=np.hstack([len(beta_pars),len(src1_pars),len(src2_pars)]) #and the fitter needs to know how many per function #this array of functions needs to return the model timestreams and derivatives w.r.t. parameters #of the timestreams. funs=[minkasi.derivs_from_isobeta_c,minkasi.derivs_from_gauss_c,minkasi.derivs_from_gauss_c] #we can keep some parameters fixed at their input values if so desired. to_fit=np.ones(len(pars),dtype='bool') to_fit[3]=False #Let's keep beta pegged to 0.7 t1=time.time() pars_fit,chisq,curve,errs=minkasi.fit_timestreams_with_derivs_manyfun(funs,pars,npar,todvec,to_fit) t2=time.time() if minkasi.myrank==0: print('took ',t2-t1,' seconds to fit timestreams') for i in range(len(pars_fit)): print('parameter ',i,' is ',pars_fit[i],' with error ',errs[i]) minkasi.comm.barrier()
4,311
38.2
114
py
minkasi
minkasi-master/examples/minkasi_map_moo_python3.py
import numpy import numpy as np from matplotlib import pyplot as plt import minkasi import time import glob from importlib import reload reload(minkasi) plt.ion() def smooth_map(map,npix=3): n=map.shape[0] m=map.shape[1] v1=np.fft.fftfreq(n)*n v2=np.fft.fftfreq(m)*m rmat=np.outer(v1,np.ones(m))**2+np.outer(np.ones(n),v2)**2 kernel=np.exp(-0.5*rmat/npix**2) mapft=np.fft.rfft2(map) kft=np.fft.rfft2(kernel) return np.fft.irfft2(mapft*kft/kft[0,0]) #assert(1==0) #find tod files we want to map #dir='../data/moo1110/' #dir='/Users/sievers/mustang/data/moo1110/' #dir='../data/moo1046/' dir = '/scratch/s/sievers/skh/tods/MOO1142/TS_EaCMS0f0_51_16_Feb_2022/' tod_names=glob.glob(dir+'Sig*.fits') #if running MPI, you would want to split up files between processes #one easy way is to say to this: tod_names=tod_names[minkasi.myrank::minkasi.nproc] #NB - minkasi checks to see if MPI is around, if not #it sets rank to 0 an nproc to 1, so this would still #run in a non-MPI environment todvec=minkasi.TodVec() #loop over each file, and read it. for fname in tod_names: t1=time.time() dat=minkasi.read_tod_from_fits(fname) t2=time.time() minkasi.truncate_tod(dat) #truncate_tod chops samples from the end to make #the length happy for ffts minkasi.downsample_tod(dat) #sometimes we have faster sampled data than we need. #this fixes that. You don't need to, though. minkasi.truncate_tod(dat) #since our length changed, make sure we have a happy length #figure out a guess at common mode #and (assumed) linear detector drifts/offset #drifts/offsets are removed, which is important for mode finding. CM is *not* removed. dd=minkasi.fit_cm_plus_poly(dat['dat_calib']) dat['dat_calib']=dd t3=time.time() tod=minkasi.Tod(dat) todvec.add_tod(tod) print('took ',t2-t1,' ',t3-t2,' seconds to read and downsample file ',fname) #make a template map with desired pixel size an limits that cover the data #todvec.lims() is MPI-aware and will return global limits, not just #the ones from private TODs lims=todvec.lims() pixsize=2.0/3600*numpy.pi/180 map=minkasi.SkyMap(lims,pixsize) #once we have a map, we can figure out the pixellization of the data. Save that #so we don't have to recompute. Also calculate a noise model. The one here #(and currently the only supported one) is to rotate the data into SVD space, then #smooth the power spectrum of each mode. Other models would not be hard #to implement. The smoothing could do with a bit of tuning as well. for tod in todvec.tods: ipix=map.get_pix(tod) tod.info['ipix']=ipix #tod.set_noise_smoothed_svd() tod.set_noise(minkasi.NoiseSmoothedSVD) ''' inds=[0,1,-1] for ind in inds: tod=todvec.tods[ind] plt.clf(); dd=tod.info['dat_calib'].copy() tvec=np.arange(dd.shape[1])*tod.info['dt'] plt.plot(tvec,dd.T) plt.xlabel("Time (s)") plt.ylabel("Detector Temperature (K)") tt=tod.info['fname'] mystr=tod.info['fname'] i1=mystr.find('-');i2=mystr.find('.',3);tag=mystr[i1+1:i2] plt.savefig('all_detectors_' + tag +'.png') plt.clf(); dat_filt=tod.apply_noise(tod.info['dat_calib']) plt.clf(); plt.plot(tvec,dat_filt[::10,:].T/np.sqrt(tod.info['dt'])/1000) plt.savefig('noise_filtered_dets_'+tag+'.png') #assert(1==0) ''' #get the hit count map. We use this as a preconditioner #which helps small-scale convergence quite a bit. hits=minkasi.make_hits(todvec,map) #setup the mapset. In general this can have many things #in addition to map(s) of the sky, but for now we'll just #use a single skymap. mapset=minkasi.Mapset() mapset.add_map(map) #make A^T N^1 d. TODs need to understand what to do with maps #but maps don't necessarily need to understand what to do with TODs, #hence putting make_rhs in the vector of TODs. #Again, make_rhs is MPI-aware, so this should do the right thing #if you run with many processes. rhs=mapset.copy() todvec.make_rhs(rhs) #this is our starting guess. Default to starting at 0, #but you could start with a better guess if you have one. x0=rhs.copy() x0.clear() #preconditioner is 1/ hit count map. helps a lot for #convergence. precon=mapset.copy() tmp=hits.map.copy() ii=tmp>0 tmp[ii]=1.0/tmp[ii] #precon.maps[0].map[:]=numpy.sqrt(tmp) precon.maps[0].map[:]=tmp[:] #run PCG! plot_info={} plot_info['vmin']=-6e-4 plot_info['vmax']=6e-4 plot_iters=[1,2,3,5,10,15,20,25,30,35,40,45,49] mapset_out=minkasi.run_pcg(rhs,x0,todvec,precon,maxiter=50)#,plot_iters=plot_iters,plot_info=plot_info) if minkasi.myrank==0: print(type(mapset_out.maps[0])) mapset_out.maps[0].write('/scratch/r/rbond/jorlo/first_map_precon_mpi_py3.fits') #and write out the map as a FITS file else: print('not writing map on process ',minkasi.myrank) if minkasi.nproc>1: minkasi.MPI.Finalize() #if you wanted to run another round of PCG starting from the previous solution, #you could, replacing x0 with mapset_out. #mapset_out2=minkasi.run_pcg(rhs,mapset_out,todvec,mapset,maxiter=50) #mapset_out2.maps[0].write('second_map.fits')
5,186
30.628049
122
py
minkasi
minkasi-master/examples/tsBowl_map_maker.py
import minkasi import numpy as np from matplotlib import pyplot as plt import glob import time import minkasi_jax.presets_by_source as pbs import os from astropy.coordinates import Angle from astropy import units as u dir = '/scratch/r/rbond/jorlo/MS0735//TS_EaCMS0f0_51_5_Oct_2021/' tod_names=glob.glob(dir+'Sig*.fits') bad_tod, addtag = pbs.get_bad_tods("MS0735", ndo=False, odo=False) #bad_tod.append('Signal_TOD-AGBT21A_123_03-s20.fits') tod_names = minkasi.cut_blacklist(tod_names, bad_tod) print("nproc: ", minkasi.nproc) tod_names = tod_names[minkasi.myrank::minkasi.nproc] todvec=minkasi.TodVec() flatten = True #loop over each file, and read it. for i, fname in enumerate(tod_names): if fname == '/scratch/r/rbond/jorlo/MS0735//TS_EaCMS0f0_51_5_Oct_2021/Signal_TOD-AGBT21A_123_03-s20.fits': continue t1=time.time() dat=minkasi.read_tod_from_fits(fname) t2=time.time() minkasi.truncate_tod(dat) #truncate_tod chops samples from the end to make #the length happy for ffts minkasi.downsample_tod(dat) #sometimes we have faster sampled data than we need. #this fixes that. You don't need to, though. minkasi.truncate_tod(dat) #since our length changed, make sure we have a happy length #figure out a guess at common mode #and (assumed) linear detector drifts/offset #drifts/offsets are removed, which is important for mode finding. CM is *not* removed. dd, pred2, cm = minkasi.fit_cm_plus_poly(dat["dat_calib"], cm_ord=3, full_out=True) dat['dat_calib']=dd if flatten: dat['dat_calib'] -= pred2 t3=time.time() tod=minkasi.Tod(dat) todvec.add_tod(tod) print('took ',t2-t1,' ',t3-t2,' seconds to read and downsample file ',fname) minkasi.barrier() lims=todvec.lims() pixsize=2.0/3600*np.pi/180 map=minkasi.SkyMap(lims,pixsize) mapset = minkasi.Mapset() for i, tod in enumerate(todvec.tods): #print(i) #print(tod.info['fname']) ipix=map.get_pix(tod) tod.info['ipix']=ipix try: tod.set_noise(minkasi.NoiseSmoothedSVD) except: print(i, tod.info['fname']) tsVec = minkasi.tsModel(todvec = todvec, modelclass = minkasi.tsBowl) #We add two things for pcg to do simulatenously here: make the maps from the tods #and fit the polynomials to tsVec mapset.add_map(map) mapset.add_map(tsVec) hits=minkasi.make_hits(todvec,map) rhs=mapset.copy() todvec.make_rhs(rhs) x0=rhs.copy() x0.clear() #preconditioner is 1/ hit count map. helps a lot for #convergence. precon=mapset.copy() tmp=hits.map.copy() ii=tmp>0 tmp[ii]=1.0/tmp[ii] precon.maps[0].map[:]=tmp[:] #tsBowl precon is 1/todlength if len(mapset.maps) > 1: for key in precon.maps[1].data.keys(): temp = np.ones(precon.maps[1].data[key].params.shape) temp /= precon.maps[1].data[key].vecs.shape[1] precon.maps[1].data[key].params = temp todvec_copy = minkasi.TodVec() for tod in todvec.tods: todvec_copy.add_tod(tod.copy()) #run PCG! plot_info={} plot_info['vmin']=-6e-4 plot_info['vmax']=6e-4 plot_iters=[1,2,3,5,10,15,20,25,30,35,40,45,49] mapset_out=minkasi.run_pcg(rhs,x0,todvec,precon,maxiter=200) if minkasi.myrank==0: if len(mapset.maps)==1: mapset_out.maps[0].write('/scratch/r/rbond/jorlo/{}_noBowl_map_precon_mpi_py3.fits'.format(flatten)) #and write out the map as a FITS file else: mapset_out.maps[0].write('/scratch/r/rbond/jorlo/{}_tsBowl_map_precon_mpi_py3.fits'.format(flatten)) else: print('not writing map on process ',minkasi.myrank) #if you wanted to run another round of PCG starting from the previous solution, #you could, replacing x0 with mapset_out. #mapset_out2=minkasi.run_pcg(rhs,mapset_out,todvec,mapset,maxiter=50) #mapset_out2.maps[0].write('second_map.fits') d2r=np.pi/180 sig=9/2.35/3600*d2r theta0 = np.deg2rad(97) x0 = Angle('07 41 44.5 hours').to(u.radian).value y0 = Angle('74:14:38.7 degrees').to(u.radian).value beta_pars=np.asarray([x0,y0,theta0,0.98,-8.2e-1]) x0_src = Angle('07 41 44.5 hours').to(u.radian).value y0_src = Angle('74:14:38.7 degrees').to(u.radian).value src1_pars=np.asarray([x0_src, y0_src,1.37e-5,1.7e-4]) #src2_pars=np.asarray([155.90447*d2r,4.1516*d2r,2.6e-5,5.1e-4]) pars=np.hstack([beta_pars,src1_pars]) #we need to combine parameters into a single vector npar=np.hstack([len(beta_pars),len(src1_pars)]) #and the fitter needs to know how many per function #this array of functions needs to return the model timestreams and derivatives w.r.t. parameters #of the timestreams. funs=[minkasi.derivs_from_isobeta_c,minkasi.derivs_from_gauss_c] #we can keep some parameters fixed at their input values if so desired. to_fit=np.ones(len(pars),dtype='bool') to_fit[[0,1,2,5,6]]=False #Let's keep beta pegged to 0.7 ''' t1=time.time() pars_fit,chisq,curve,errs=minkasi.fit_timestreams_with_derivs_manyfun(funs,pars,npar,todvec_copy,to_fit) t2=time.time() if minkasi.myrank==0: print('No Sub: ') print('took ',t2-t1,' seconds to fit timestreams') for i in range(len(pars_fit)): print('parameter ',i,' is ',pars_fit[i],' with error ',errs[i]) ''' minkasi.comm.barrier() if minkasi.myrank==0: for tod in todvec.tods: todname = tod.info['fname'] temp = minkasi.map2todbowl(mapset_out.maps[1].data[todname].vecs, mapset_out.maps[1].data[todname].params) tod.info['dat_calib'] -= temp ''' t1=time.time() pars_fit,chisq,curve,errs=minkasi.fit_timestreams_with_derivs_manyfun(funs,pars,npar,todvec,to_fit) t2=time.time() if minkasi.myrank==0: print('Sub: ') print('took ',t2-t1,' seconds to fit timestreams') for i in range(len(pars_fit)): print('parameter ',i,' is ',pars_fit[i],' with error ',errs[i]) ''' lims=todvec.lims() pixsize=2.0/3600*np.pi/180 map=minkasi.SkyMap(lims,pixsize) mapset = minkasi.Mapset() mapset.add_map(map) hits=minkasi.make_hits(todvec,map) rhs=mapset.copy() todvec.make_rhs(rhs) x0=rhs.copy() x0.clear() #preconditioner is 1/ hit count map. helps a lot for #convergence. precon=mapset.copy() tmp=hits.map.copy() ii=tmp>0 tmp[ii]=1.0/tmp[ii] precon.maps[0].map[:]=tmp[:] mapset_out=minkasi.run_pcg(rhs,x0,todvec,precon,maxiter=50) if minkasi.myrank==0: mapset_out.maps[0].write('/scratch/r/rbond/jorlo/{}_sub_tsBowl_map_precon_mpi_py3.fits'.format(flatten)) #and write out the map as a FITS file else: print('not writing map on process ',minkasi.myrank)
6,463
28.788018
146
py
minkasi
minkasi-master/minkasi/code_scrapyard.py
def __run_pcg_old(b,x0,tods,mapset,precon): Ax=mapset.dot(x0) r=b-Ax z=precon*r p=z.copy() k=0 zr=r.dot(z) x=x0.copy() for iter in range(25): print(iter,zr) Ap=mapset.dot(p) pAp=p.dot(Ap) alpha=zr/pAp x_new=x+p*alpha r_new=r-Ap*alpha z_new=precon*r_new zr_new=r_new.dot(z_new) beta=zr_new/zr p_new=z_new+p*beta p=p_new z=z_new r=r_new zr=zr_new x=x_new return x def run_pcg_wprior_old(b,x0,tods,prior,precon=None,maxiter=25): t1=time.time() Ax=tods.dot(x0) #prior.apply_prior(Ax,x0) flub=prior.apply_prior(x0.maps[0].map) print('means of flub and Ax are ',np.mean(np.abs(Ax.maps[0].map)),np.mean(np.abs(flub))) Ax.maps[0].map=Ax.maps[0].map+prior.apply_prior(x0.maps[0].map) try: r=b.copy() r.axpy(Ax,-1) except: r=b-Ax if not(precon is None): z=precon*r else: z=r.copy() p=z.copy() k=0.0 zr=r.dot(z) x=x0.copy() t2=time.time() for iter in range(maxiter): if myrank==0: if iter>0: print(iter,zr,alpha,t2-t1,t3-t2,t3-t1) else: print(iter,zr,t2-t1) t1=time.time() Ap=tods.dot(p) fwee=prior.apply_prior(p.maps[0].map) #print 'means are ',np.mean(np.abs(Ap.maps[0].map)),np.mean(np.abs(fwee)) Ap.maps[0].map=Ap.maps[0].map+fwee #prior.apply_prior(Ap,p) t2=time.time() pAp=p.dot(Ap) alpha=zr/pAp try: x_new=x.copy() x_new.axpy(p,alpha) except: x_new=x+p*alpha try: r_new=r.copy() r_new.axpy(Ap,-alpha) except: r_new=r-Ap*alpha if not(precon is None): z_new=precon*r_new else: z_new=r_new.copy() zr_new=r_new.dot(z_new) beta=zr_new/zr try: p_new=z_new.copy() p_new.axpy(p,beta) except: p_new=z_new+p*beta p=p_new z=z_new r=r_new zr=zr_new x=x_new t3=time.time() return x class tsStripes_old(tsGeneric): def __init__(self,tod,seg_len=100,do_slope=True): dims=tod.get_data_dims() nseg=dims[1]//seg_len if nseg*seg_len<dims[1]: nseg=nseg+1 #this says to connect segments with straight lines as #opposed to simple horizontal offsets if do_slope: nseg=nseg+1 self.nseg=nseg self.seg_len=seg_len self.params=np.zeros([dims[0],self.nseg]) self.do_slope=do_slope def tod2map(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: dat=tod.get_data() tmp=np.zeros(self.params.shape) if self.do_slope: vec=np.arange(self.seg_len)/self.seg_len vec2=1-vec vv=np.vstack([vec2,vec]) nseg=dat.shape[1]//self.seg_len imax=nseg if nseg*self.seg_len<dat.shape[1]: have_extra=True nextra=dat.shape[1]-imax*self.seg_len else: have_extra=False nseg=nseg-1 nextra=0 for i in range(imax): #tmp[:,i:i+2]=tmp[:,i:i+2]+dat[:,i*self.seg_len:(i+1)*self.seg_len]@(vv.T) tmp[:,i:i+2]=tmp[:,i:i+2]+np.dot(dat[:,i*self.seg_len:(i+1)*self.seg_len],(vv.T)) if have_extra: vec=np.arange(nextra)/nextra vec2=1-vec vv=np.vstack([vec2,vec]) #tmp[:,-2:]=tmp[:,-2:]+dat[:,self.seg_len*nseg:]@(vv.T) tmp[:,-2:]=tmp[:,-2:]+np.dot(dat[:,self.seg_len*nseg:],(vv.T)) else: nseg=dat.shape[1]//self.seg_len if nseg*self.seg_len<dat.shape[1]: nseg=nseg+1 vec=np.zeros(nseg*self.seg_len) ndet=dat.shape[0] ndat=dat.shape[1] for i in range(ndet): vec[:ndat]=dat[i,:] vv=np.reshape(vec,[nseg,self.seg_len]) tmp[i,:]=np.sum(vv,axis=1) if do_add: self.params[:]=self.params[:]+tmp else: self.params[:]=tmp def map2tod(self,tod,dat=None,do_add=True,do_omp=False): tmp=tod.get_empty() ndet=tmp.shape[0] ndat=tmp.shape[1] if self.do_slope: vec=np.arange(self.seg_len)/self.seg_len vec2=1-vec vv=np.vstack([vec2,vec]) nseg=tmp.shape[1]//self.seg_len imax=nseg if imax*self.seg_len<tmp.shape[1]: have_extra=True nextra=tmp.shape[1]-imax*self.seg_len else: have_extra=False nseg=nseg-1 #imax=imax+1 for i in range(imax): #tmp[:,i*self.seg_len:(i+1)*self.seg_len]=self.params[:,i:i+2]@vv tmp[:,i*self.seg_len:(i+1)*self.seg_len]=np.dot(self.params[:,i:i+2],vv) if have_extra: vec=np.arange(nextra)/nextra vec2=1-vec vv=np.vstack([vec2,vec]) #tmp[:,self.seg_len*nseg:]=self.params[:,-2:]@vv tmp[:,self.seg_len*nseg:]=np.dot(self.params[:,-2:],vv) else: ndet=tmp.shape[0] ndat=tmp.shape[1] for i in range(ndet): pars=self.params[i,:] mymod=np.repeat(pars,self.seg_len) tmp[i,:]=mymod[:ndat] if dat is None: dat=tmp return dat else: if do_add: dat[:]=dat[:]+tmp else: dat[:]=tmp class __tsAirmass_old: def __init__(self,tod=None,order=5): if tod is None: self.sz=np.asarray([0,0],dtype='int') self.params=np.zeros(1) self.fname='' self.order=0 self.airmass=None else: #self.sz=tod.info['dat_calib'].shape self.sz=tod.get_data_dims() self.fname=tod.info['fname'] self.order=order self.params=np.zeros(order+1) self.airmass=scaled_airmass_from_el(tod.info['elev']) def copy(self,copyMat=False): cp=tsAirmass() cp.sz=self.sz cp.params=self.params.copy() cp.fname=self.fname cp.order=self.order if copyMat: cp.airmass=self.airmass.copy() else: cp.airmass=self.airmass #since this shouldn't change, use a pointer to not blow up RAM return cp def clear(self): self.params[:]=0.0 def dot(self,ts): return np.sum(self.params*ts.params) def axpy(self,ts,a): self.params=self.params+a*ts.params def _get_current_legmat(self): x=np.linspace(-1,1,self.sz[1]) m1=np.polynomial.legendre.legvander(x,self.order) return m1 def _get_current_model(self): x=np.linspace(-1,1,self.sz[1]) m1=self._get_current_legmat() poly=np.dot(m1,self.params) mat=np.repeat([poly],self.sz[0],axis=0) mat=mat*self.airmass return mat def tod2map(self,tod,dat,do_add=True,do_omp=False): poly=self._get_current_legmat() vec=np.sum(dat*self.airmass,axis=0) atd=np.dot(vec,poly) if do_add: self.params[:]=self.params[:]+atd else: self.params[:]=atd def map2tod(self,tod,dat,do_add=True,do_omp=False): mat=self._get_current_model() if do_add: dat[:]=dat[:]+mat else: dat[:]=mat def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params return tt def write(self,fname=None): pass def _fit_timestreams_with_derivs_old(func,pars,tods,to_fit=None,to_scale=None,tol=1e-2,maxiter=10,scale_facs=None): '''Fit a model to timestreams. func should return the model and the derivatives evaluated at the parameter values in pars. to_fit says which parameters to float. 0 to fix, 1 to float, and anything larger than 1 is expected to vary together (e.g. shifting a TOD pointing mode you could put in a 2 for all RA offsets and a 3 for all dec offsets.). to_scale will normalize by the input value, so one can do things like keep relative fluxes locked together.''' if not(to_fit is None): #print 'working on creating rotmat' to_fit=np.asarray(to_fit,dtype='int64') inds=np.unique(to_fit) nfloat=np.sum(to_fit==1) ncovary=np.sum(inds>1) nfit=nfloat+ncovary rotmat=np.zeros([len(pars),nfit]) solo_inds=np.where(to_fit==1)[0] icur=0 for ind in solo_inds: rotmat[ind,icur]=1.0 icur=icur+1 if ncovary>0: group_inds=inds[inds>1] for ind in group_inds: ii=np.where(to_fit==ind)[0] rotmat[ii,icur]=1.0 icur=icur+1 iter=0 converged=False pp=pars.copy() while (converged==False) and (iter<maxiter): curve=0.0 grad=0.0 chisq=0.0 for tod in tods.tods: #sz=tod.info['dat_calib'].shape sz=tod.get_data_dims() derivs,pred=func(pp,tod) if not (to_fit is None): derivs=np.dot(rotmat.transpose(),derivs) derivs_filt=0*derivs tmp=np.zeros(sz) npp=derivs.shape[0] nn=derivs.shape[1] #delt=tod.info['dat_calib']-pred delt=tod.get_data()-pred delt_filt=tod.apply_noise(delt) for i in range(npp): tmp[:,:]=np.reshape(derivs[i,:],sz) tmp_filt=tod.apply_noise(tmp) derivs_filt[i,:]=np.reshape(tmp_filt,nn) delt=np.reshape(delt,nn) delt_filt=np.reshape(delt_filt,nn) grad1=np.dot(derivs,delt_filt) grad2=np.dot(derivs_filt,delt) grad=grad+0.5*(grad1+grad2) curve=curve+np.dot(derivs,derivs_filt.transpose()) chisq=chisq+np.dot(delt,delt_filt) if iter==0: chi_ref=chisq curve=0.5*(curve+curve.transpose()) curve=curve+2.0*np.diag(np.diag(curve)) #double the diagonal for testing purposes curve_inv=np.linalg.inv(curve) errs=np.sqrt(np.diag(curve_inv)) shifts=np.dot(curve_inv,grad) #print errs,shifts conv_fac=np.max(np.abs(shifts/errs)) if conv_fac<tol: print('We have converged.') converged=True if not (to_fit is None): shifts=np.dot(rotmat,shifts) if not(scale_facs is None): if iter<len(scale_facs): print('rescaling shift by ',scale_facs[iter]) shifts=shifts*scale_facs[iter] to_print=np.asarray([3600*180.0/np.pi,3600*180.0/np.pi,3600*180.0/np.pi,1.0,1.0,3600*180.0/np.pi,3600*180.0/np.pi,3600*180.0/np.pi*np.sqrt(8*np.log(2)),1.0])*(pp-pars) print('iter ',iter,' max shift is ',conv_fac,' with chisq improvement ',chi_ref-chisq,to_print) #converged,pp,shifts pp=pp+shifts iter=iter+1 return pp,chisq #class Cuts: # def __init__(self,tod): # self.tag=tod.info['tag'] # self.ndet=tod.info['dat_calib'].shape[0] # self.cuts=[None]*self.ndet # #class CutsVec: # def __init__(self,todvec): # self.ntod=todvec.ntod # self.cuts=[None]*self.ntod # for tod in todvec.tods: # self.cuts[tod.info['tag']]=Cuts(tod) #this class is pointless, as you can get the same functionality with the tsModel class, which will be #consistent with other timestream model classes. #class CutsVecs: # def __init__(self,todvec,do_add=True): # #if class(todvec)==CutsVecs: #for use in copy # if isinstance(todvec,CutsVecs): # self.cuts=[None]*todvec.ntod # self.ntod=todvec.ntod # for i in range(todvec.ntod): # self.cuts[i]=todvec.cuts[i].copy() # return # #if class(todvec)!=TodVec: # if not(isinstance(todvec,TodVec)): # print('error in CutsVecs init, must pass in a todvec class.') # return None # self.cuts=[None]*todvec.ntod # self.ntod=todvec.ntod # for i in range(todvec.ntod): # tod=todvec.tods[i] # if tod.info['tag']!=i: # print('warning, tag mismatch in CutsVecs.__init__') # print('continuing, but you should be careful...') # if 'bad_samples' in tod.info: # self.cuts[i]=Cuts(tod,do_add) # elif 'mask' in tod.info: # self.cuts[i]=CutsCompact(tod) # self.cuts[i].cuts_from_array(tod.info['mask']) # self.cuts[i].get_imap() # def copy(self): # return CutsVecs(self) # def clear(self): # for cuts in self.cuts: # cuts.clear() # def axpy(self,cutsvec,a): # assert(self.ntod==cutsvec.ntod) # for i in range(ntod): # self.cuts[i].axpy(cutsvec.cuts[i],a) # def map2tod(self,todvec): # assert(self.ntod==todvec.ntod) # for i in range(self.ntod): # self.cuts[i].map2tod(todvec.tods[i]) # def tod2map(self,todvec,dat): # assert(self.ntod==todvec.ntod) # assert(self.ntod==dat.ntod) # for i in range(self.ntod): # self.cuts[i].tod2map(todvec.tods[i],dat.tods[i]) # def dot(self,cutsvec): # tot=0.0 # assert(self.ntod==cutsvec.ntod) # for i in range(self.ntod): # tot+=self.cuts[i].dot(cutsvec.cuts[i]) # return tot
14,023
31.843091
175
py
minkasi
minkasi-master/minkasi/minkasi_nb.py
import numpy as np import numba as nb @nb.njit(parallel=True) def map2tod_destriped(mat,pars,lims,do_add=True): ndet=mat.shape[0] nseg=len(lims)-1 for seg in nb.prange(nseg): for det in range(ndet): if do_add: for i in range(lims[seg],lims[seg+1]): mat[det,i]=mat[det,i]+pars[det,seg] else: for i in range(lims[seg],lims[seg+1]): mat[det,i]=pars[det,seg] @nb.njit(parallel=True) def tod2map_destriped(mat,pars,lims,do_add=True): ndet=mat.shape[0] nseg=len(lims)-1 for seg in nb.prange(nseg): for det in range(ndet): if do_add==False: pars[det,seg]=0 for i in range(lims[seg],lims[seg+1]): pars[det,seg]=pars[det,seg]+mat[det,i] @nb.njit(parallel=True) def __map2tod_binned_det_loop(pars,inds,mat,ndet,n): for det in nb.prange(ndet): for i in range(n): mat[det][i]=mat[det][i]+pars[det][inds[i]] #pars[det][inds[i]]=pars[det][inds[i]]+mat[det][i] def map2tod_binned_det(mat,pars,vec,lims,nbin,do_add=True): n=mat.shape[1] #print('range is ',pars.min(),pars.max()) #inds=np.empty(n,dtype='int64') fac=nbin/(lims[1]-lims[0]) inds=np.asarray((vec-lims[0])*fac,dtype='int64') #print('ind range is ',inds.min(),inds.max()) #for i in nb.prange(n): # inds[i]=(vec[i]-lims[0])*fac ndet=mat.shape[0] if do_add==False: mat[:]=0 __map2tod_binned_det_loop(pars,inds,mat,ndet,n) #for det in nb.prange(ndet): # for i in np.arange(n): # pars[det][inds[i]]=pars[det][inds[i]]+mat[det][i] @nb.njit(parallel=True) def __tod2map_binned_det_loop(pars,inds,mat,ndet,n): for det in nb.prange(ndet): for i in range(n): pars[det][inds[i]]=pars[det][inds[i]]+mat[det][i] def tod2map_binned_det(mat,pars,vec,lims,nbin,do_add=True): #print('dims are ',mat.shape,pars.shape,vec.shape) #print('lims are ',lims,nbin,vec.min(),vec.max()) n=mat.shape[1] fac=nbin/(lims[1]-lims[0]) #inds=np.empty(n,dtype='int64') #for i in nb.prange(n): # inds[i]=(vec[i]-lims[0])*fac inds=np.asarray((vec-lims[0])*fac,dtype='int64') #print('max is ',inds.max()) ndet=mat.shape[0] if do_add==False: mat[:]=0 __tod2map_binned_det_loop(pars,inds,mat,ndet,n) #for det in nb.prange(ndet): # for i in np.arange(n): # pars[det][inds[i]]=pars[det][inds[i]]+mat[det][i] return 0 #@nb.njit(parallel=True) #def map2tod_binned_det(mat,pars,vec,lims,nbin,do_add=True): # n=mat.shape[1] # inds=np.empty(n,dtype='int') # fac=nbin/(lims[1]-lims[0]) # for i in nb.prange(n): # inds[i]=(vec[i]-lims[0])*fac # ndet=mat.shape[0] # if do_add==False: # mat[:]=0 # for det in np.arange(ndet): # for i in nb.prange(n): # mat[det][i]=mat[det][i]+pars[det][inds[i]] @nb.njit(parallel=True) def fill_elliptical_isobeta(params,dx,dy,pred): ndet=dx.shape[0] n=dx.shape[1] x0=params[0] y0=params[1] theta1=params[2] theta2=params[3] theta1_inv=1/theta1 theta2_inv=1/theta2 theta1_inv_sqr=theta1_inv**2 theta2_inv_sqr=theta2_inv**2 psi=params[4] beta=params[5] amp=params[6] cosdec=np.cos(y0) cospsi=np.cos(psi) sinpsi=np.sin(psi) mypow=0.5-1.5*beta for det in nb.prange(ndet): for j in np.arange(n): delx=(dx[det,j]-x0)*cosdec dely=dy[det,j]-y0 xx=delx*cospsi+dely*sinpsi yy=dely*cospsi-delx*sinpsi rr=1+theta1_inv_sqr*xx*xx+theta2_inv_sqr*yy*yy pred[det,j]=amp*(rr**mypow) @nb.njit(parallel=True) def fill_elliptical_isobeta_derivs(params,dx,dy,pred,derivs): """Fill model/derivatives for an isothermal beta model. Parameters should be [ra,dec,theta axis 1,theta axis 2,angle,beta,amplitude. Beta should be positive (i.e. 0.7, not -0.7).""" ndet=dx.shape[0] n=dx.shape[1] x0=params[0] y0=params[1] theta1=params[2] theta2=params[3] theta1_inv=1/theta1 theta2_inv=1/theta2 theta1_inv_sqr=theta1_inv**2 theta2_inv_sqr=theta2_inv**2 psi=params[4] beta=params[5] amp=params[6] cosdec=np.cos(y0) sindec=np.sin(y0)/np.cos(y0) #cosdec=np.cos(dy[0,0]) #cosdec=1.0 cospsi=np.cos(psi) cc=cospsi**2 sinpsi=np.sin(psi) ss=sinpsi**2 cs=cospsi*sinpsi mypow=0.5-1.5*beta for det in nb.prange(ndet): for j in np.arange(n): delx=(dx[det,j]-x0)*cosdec dely=dy[det,j]-y0 xx=delx*cospsi+dely*sinpsi yy=dely*cospsi-delx*sinpsi xfac=theta1_inv_sqr*xx*xx yfac=theta2_inv_sqr*yy*yy #rr=1+theta1_inv_sqr*xx*xx+theta2_inv_sqr*yy*yy rr=1+xfac+yfac rrpow=rr**mypow pred[det,j]=amp*rrpow dfdrr=rrpow/rr*mypow drdx=-2*delx*(cc*theta1_inv_sqr+ss*theta2_inv_sqr)-2*dely*(theta1_inv_sqr-theta2_inv_sqr)*cs #drdy=-2*dely*(cc*theta2_inv_sqr+ss*theta1_inv_sqr)-2*delx*(theta1_inv_sqr-theta2_inv_sqr)*cs drdy=-(2*xx*theta1_inv_sqr*(cospsi*sindec*delx+sinpsi)+2*yy*theta2_inv_sqr*(-sinpsi*sindec*delx+cospsi)) drdtheta=2*(theta1_inv_sqr-theta2_inv_sqr)*(cs*(dely**2-delx**2)+delx*dely*(cc-ss)) #drdtheta=-2*delx**2*cs*(theta_1_inv_sqr-theta_2_inv_sqr)+2*dely*delx*(theta_1_inv_sqr-theta_2_inv_sqr)*(cc-ss)+2*dely**2*cs*( derivs[0,det,j]=dfdrr*drdx*cosdec derivs[1,det,j]=dfdrr*drdy derivs[2,det,j]=dfdrr*xfac*(-2*theta1_inv) derivs[3,det,j]=dfdrr*yfac*(-2*theta2_inv) derivs[4,det,j]=dfdrr*drdtheta derivs[5,det,j]=-1.5*np.log(rr)*amp*rrpow derivs[6,det,j]=rrpow @nb.njit(parallel=True) def fill_elliptical_gauss_derivs(params,dx,dy,pred,derivs): """Fill model/derivatives for an elliptical gaussian model. Parameters should be [ra,dec,sigma axis 1,sigmaaxis 2,angle,amplitude.""" ndet=dx.shape[0] n=dx.shape[1] x0=params[0] y0=params[1] theta1=params[2] theta2=params[3] theta1_inv=1/theta1 theta2_inv=1/theta2 theta1_inv_sqr=theta1_inv**2 theta2_inv_sqr=theta2_inv**2 psi=params[4] amp=params[5] cosdec=np.cos(y0) sindec=np.sin(y0)/np.cos(y0) cospsi=np.cos(psi) cc=cospsi**2 sinpsi=np.sin(psi) ss=sinpsi**2 cs=cospsi*sinpsi for det in nb.prange(ndet): for j in np.arange(n): delx=(dx[det,j]-x0)*cosdec dely=dy[det,j]-y0 xx=delx*cospsi+dely*sinpsi yy=dely*cospsi-delx*sinpsi xfac=theta1_inv_sqr*xx*xx yfac=theta2_inv_sqr*yy*yy #rr=1+theta1_inv_sqr*xx*xx+theta2_inv_sqr*yy*yy rr=xfac+yfac rrpow=np.exp(-0.5*rr) pred[det,j]=amp*rrpow dfdrr=-0.5*rrpow drdx=-2*delx*(cc*theta1_inv_sqr+ss*theta2_inv_sqr)-2*dely*(theta1_inv_sqr-theta2_inv_sqr)*cs #drdy=-2*dely*(cc*theta2_inv_sqr+ss*theta1_inv_sqr)-2*delx*(theta1_inv_sqr-theta2_inv_sqr)*cs drdy=-(2*xx*theta1_inv_sqr*(cospsi*sindec*delx+sinpsi)+2*yy*theta2_inv_sqr*(-sinpsi*sindec*delx+cospsi)) drdtheta=2*(theta1_inv_sqr-theta2_inv_sqr)*(cs*(dely**2-delx**2)+delx*dely*(cc-ss)) #drdtheta=-2*delx**2*cs*(theta_1_inv_sqr-theta_2_inv_sqr)+2*dely*delx*(theta_1_inv_sqr-theta_2_inv_sqr)*(cc-ss)+2*dely**2*cs*( derivs[0,det,j]=dfdrr*drdx*cosdec #derivs[1,det,j]=dfdrr*(drdy-2*sindec*delx**2*theta1_inv_sqr) derivs[1,det,j]=dfdrr*drdy derivs[2,det,j]=dfdrr*xfac*(-2*theta1_inv) derivs[3,det,j]=dfdrr*yfac*(-2*theta2_inv) derivs[4,det,j]=dfdrr*drdtheta derivs[5,det,j]=rrpow @nb.njit(parallel=True) def radec2pix_car(ra,dec,ipix,lims,pixsize,cosdec,ny): ra=np.ravel(ra) dec=np.ravel(dec) ipix=np.ravel(ipix) n=len(ipix) for i in nb.prange(n): xpix=int((ra[i]-lims[0])*cosdec/pixsize+0.5) ypix=int((dec[i]-lims[2])/pixsize+0.5) ipix[i]=xpix*ny+ypix @nb.njit(parallel=True) def axpy_in_place(y,x,a=1.0): #add b into a n=x.shape[0] m=x.shape[1] assert(n==y.shape[0]) assert(m==y.shape[1]) #Numba has a bug, as of at least 0.53.1 (an 0.52.0) where #both parts of a conditional can get executed, so don't #try to be fancy. Lower-down code can be used in the future. for i in nb.prange(n): for j in np.arange(m): y[i,j]=y[i,j]+x[i,j]*a #isone=(a==1.0) #if isone: # for i in nb.prange(n): # for j in np.arange(m): # y[i,j]=y[i,j]+x[i,j] #else: # for i in nb.prange(n): # for j in np.arange(m): # y[i,j]=y[i,j]+x[i,j]*a @nb.njit(parallel=True) def scale_matrix_by_vector(mat,vec,axis=1): n=mat.shape[0] m=mat.shape[1] if axis==1: assert(len(vec)==n) for i in nb.prange(n): for j in np.arange(m): mat[i,j]=mat[i,j]*vec[i] elif axis==0: assert(len(vec)==m) for i in nb.prange(n): for j in np.arange(m): mat[i,j]=mat[i,j]*vec[j] else: print('unsupported number of dimensions in scale_matrix_by_vector')
9,537
31.114478
138
py
minkasi
minkasi-master/minkasi/minkasi.py
import os import numpy as np import ctypes import time from . import mkfftw #import pyfits from astropy.io import fits as pyfits import astropy from astropy import wcs from astropy.io import fits from astropy.cosmology import WMAP9 as cosmo #choose your cosmology here import scipy import copy import sys from numba import jit try: import healpy have_healpy=True except: have_healpy=False try: import numba as nb from . import minkasi_nb have_numba=True except: have_numba=False try: import qpoint as qp have_qp=True except: have_qp=False print('importing mpi4py') try: import mpi4py.rc mpi4py.rc.threads = False from mpi4py import MPI print('mpi4py imported') comm=MPI.COMM_WORLD myrank = comm.Get_rank() nproc=comm.Get_size() print('nproc:, ', nproc) if nproc>1: have_mpi=True else: have_mpi=False except: have_mpi=False myrank=0 nproc=1 #try: # import numba as nb # have_numba=True #else: # have_numba=False try: mylib=ctypes.cdll.LoadLibrary("libminkasi.so") except OSError: mylib=ctypes.cdll.LoadLibrary(os.path.join(os.path.dirname(os.path.abspath(__file__)), "libminkasi.so")) tod2map_simple_c=mylib.tod2map_simple tod2map_simple_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p] tod2map_atomic_c=mylib.tod2map_atomic tod2map_atomic_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p] #tod2map_everyone_c=mylib.tod2map_everyone #tod2map_everyone_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] tod2map_omp_c=mylib.tod2map_omp tod2map_omp_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] tod2map_cached_c=mylib.tod2map_cached tod2map_cached_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] map2tod_simple_c=mylib.map2tod_simple map2tod_simple_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] map2tod_omp_c=mylib.map2tod_omp map2tod_omp_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int, ctypes.c_int, ctypes.c_void_p,ctypes.c_int] map2tod_iqu_omp_c=mylib.map2tod_iqu_omp map2tod_iqu_omp_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] map2tod_qu_omp_c=mylib.map2tod_qu_omp map2tod_qu_omp_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p,ctypes.c_int] tod2map_iqu_simple_c=mylib.tod2map_iqu_simple tod2map_iqu_simple_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p] tod2map_qu_simple_c=mylib.tod2map_qu_simple tod2map_qu_simple_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p] tod2map_iqu_precon_simple_c=mylib.tod2map_iqu_precon_simple tod2map_iqu_precon_simple_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p] tod2map_qu_precon_simple_c=mylib.tod2map_qu_precon_simple tod2map_qu_precon_simple_c.argtypes=[ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int,ctypes.c_int,ctypes.c_void_p] scan_map_c=mylib.scan_map scan_map_c.argtypes=[ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int] tod2cuts_c=mylib.tod2cuts tod2cuts_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int] cuts2tod_c=mylib.cuts2tod cuts2tod_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int] set_nthread_c=mylib.set_nthread set_nthread_c.argtypes=[ctypes.c_int] get_nthread_c=mylib.get_nthread get_nthread_c.argtypes=[ctypes.c_void_p] fill_isobeta_c=mylib.fill_isobeta fill_isobeta_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int] fill_isobeta_derivs_c=mylib.fill_isobeta_derivs fill_isobeta_derivs_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int] fill_gauss_derivs_c=mylib.fill_gauss_derivs fill_gauss_derivs_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int] fill_gauss_src_c=mylib.fill_gauss_src fill_gauss_src_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int] outer_c=mylib.outer_block outer_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int] def y2rj(freq=90): """conversion to multiply a y map by to get a Rayleigh-Jeans normalized map note that it doesn't have the T_cmb at the end, so the value for low frequencies is -2.""" kb=1.38064852e-16 h=6.62607004e-27 T=2.725 x=freq*1e9*h/kb/T ex=np.exp(x) f=x**2*ex/(ex-1)**2*( x*(ex+1)/(ex-1)-4) return f def planck_g(freq=90): """conversion between T_CMB and T_RJ as a function of frequency.""" kb=1.38064852e-16 h=6.62607004e-27 T=2.725 x=freq*1e9*h/kb/T ex=np.exp(x) return x**2*ex/( (ex-1)**2) def report_mpi(): if have_mpi: print('myrank is ',myrank,' out of ',nproc) else: print('mpi not found') def barrier(): if have_mpi: comm.barrier() else: pass def invsafe(mat,thresh=1e-14): u,s,v=np.linalg.svd(mat,0) ii=np.abs(s)<thresh*s.max() #print ii s_inv=1/s s_inv[ii]=0 tmp=np.dot(np.diag(s_inv),u.transpose()) return np.dot(v.transpose(),tmp) def tod2map_simple(map,dat,ipix): ndet=dat.shape[0] ndata=dat.shape[1] if not(ipix.dtype=='int32'): print("Warning - ipix is not int32 in tod2map_simple. this is likely to produce garbage results.") tod2map_simple_c(map.ctypes.data,dat.ctypes.data,ndet,ndata,ipix.ctypes.data) def tod2map_everyone(map,dat,ipix,edges): assert(len(edges)==get_nthread()+1) tod2map_everyone_c(map.ctypes.data,dat.ctypes.data,dat.shape[0],dat.shape[1],ipix.ctypes.data,map.size,edges.ctypes.data,len(edges)) def tod2map_omp(map,dat,ipix,atomic=False): ndet=dat.shape[0] ndata=dat.shape[1] if not(ipix.dtype=='int32'): print("Warning - ipix is not int32 in tod2map_omp. this is likely to produce garbage results.") if atomic: tod2map_atomic_c(map.ctypes.data,dat.ctypes.data,ndet,ndata,ipix.ctypes.data,map.size) else: tod2map_omp_c(map.ctypes.data,dat.ctypes.data,ndet,ndata,ipix.ctypes.data,map.size) def tod2map_cached(map,dat,ipix): ndet=dat.shape[0] ndata=dat.shape[1] if not(ipix.dtype=='int32'): print("Warning - ipix is not int32 in tod2map_cached. this is likely to produce garbage results.") tod2map_cached_c(map.ctypes.data,dat.ctypes.data,ndet,ndata,ipix.ctypes.data,map.shape[1]) def tod2polmap(map,dat,poltag,twogamma,ipix): ndet=dat.shape[0] ndata=dat.shape[1] fun=None if poltag=='QU': fun=tod2map_qu_simple_c if poltag=='IQU': fun=tod2map_iqu_simple_c if poltag=='QU_PRECON': fun=tod2map_qu_precon_simple_c if poltag=='IQU_PRECON': fun=tod2map_iqu_precon_simple_c if fun is None: print('unrecognized poltag ' + repr(poltag) + ' in tod2polmap.') #print('calling ' + repr(fun)) fun(map.ctypes.data,dat.ctypes.data,twogamma.ctypes.data,ndet,ndata,ipix.ctypes.data) def map2tod(dat,map,ipix,do_add=False,do_omp=True): ndet=dat.shape[0] ndata=dat.shape[1] if do_omp: map2tod_omp_c(dat.ctypes.data, map.ctypes.data, ndet, ndata, ipix.ctypes.data, do_add) else: map2tod_simple_c(dat.ctypes.data,map.ctypes.data,ndet,ndata,ipix.ctypes.data,do_add) def polmap2tod(dat,map,poltag,twogamma,ipix,do_add=False,do_omp=True): ndet=dat.shape[0] ndata=dat.shape[1] fun=None if poltag=='QU': fun=map2tod_qu_omp_c if poltag=='IQU': fun=map2tod_iqu_omp_c if poltag=='QU_PRECON': fun=map2tod_qu_precon_omp_c if poltag=='IQU_PRECON': fun=map2tod_iqu_precon_omp_c if fun is None: print('unknown poltag ' + repr(poltag) + ' in polmap2tod.') return #print('calling ' + repr(fun)) fun(dat.ctypes.data,map.ctypes.data,twogamma.ctypes.data,ndet,ndata,ipix.ctypes.data,do_add) @jit(nopython=True) def map2todbowl(vecs, params): """ Converts parameters to tods for the tsBowl class. Parameters ---------- vecs: np.array(order, ndata, ndet) pseudo-Vandermonde matrix params: np.array(order, ndet) corresponding weights for pseudo-Vandermonde matrix """ #Return tod should have shape ndet x ndata to_return = np.zeros((vecs.shape[0], vecs.shape[-2])) for i in range(vecs.shape[0]): to_return[i] = np.dot(vecs[i,...], params[i,...]) return to_return @jit(nopython=True) def tod2mapbowl(vecs, mat): """ transpose of map2tod for bowling Parameters ---------- vecs: np.array(ndet, ndata, order) pseudo-Vandermonde matrix mat: np.array(ndet, ndata) tod data """ #Return tod should have shape ndet x ndata to_return = np.zeros((vecs.shape[0], vecs.shape[-1])) for i in range(vecs.shape[0]): to_return[i] = np.dot(vecs[i,...].T, mat[i,...]) return to_return def read_fits_map(fname,hdu=0,do_trans=True): f=fits.open(fname) raw=f[hdu].data tmp=raw.copy() f.close() if do_trans: tmp=(tmp.T).copy() return tmp def write_fits_map_wheader(map,fname,header,do_trans=True): if do_trans: map=(map.T).copy() hdu=fits.PrimaryHDU(map,header=header) try: hdu.writeto(fname,overwrite=True) except: hdu.writeto(fname,clobber=True) def get_ft_vec(n): x=np.arange(n) x[x>n/2]=x[x>n/2]-n return x def set_nthread(nthread): set_nthread_c(nthread) def get_nthread(): nthread=np.zeros([1,1],dtype='int32') get_nthread_c(nthread.ctypes.data) return nthread[0,0] def segs_from_vec(vec,pad=True): """ segs_from_vec(vec,pad=True) return the starting/stopping points of regions marked False in vec. For use in e.g. generating cuts from a vector/array. If pad is False, assume vector is already True-padded""" #insert input vector into a True-padded vector do make reasoning about starting/stopping points #of False regions easier. if pad: vv=np.ones(len(vec)+2,dtype='bool') vv[1:-1]=vec else: if vec.dtype=='bool': vv=vec else: vv=np.ones(len(vec),dtype='bool') vv[:]=vec if vv.min()==True: nseg=0 istart=[] istop=[] else: inds=np.where(np.diff(vv))[0] assert(len(inds)%2==0) nseg=len(inds)//2 istart=[] istop=[] for i in range(nseg): istart.append(inds[2*i]) istop.append(inds[2*i+1]) return nseg,istart,istop def cut_blacklist(tod_names,blacklist): mydict={} for nm in tod_names: tt=nm.split('/')[-1] mydict[tt]=nm ncut=0 for nm in blacklist: tt=nm.split('/')[-1] #if mydict.has_key(tt): if tt in mydict: ncut=ncut+1 del(mydict[tt]) if ncut>0: print('deleted ',ncut,' bad files.') mynames=mydict.values() mynames.sort() return mynames else: return tod_names def find_spikes(dat,inner=1,outer=10,rad=0.25,thresh=8,pad=2): #find spikes in a block of timestreams n=dat.shape[1]; ndet=dat.shape[0] x=np.arange(n); filt1=np.exp(-0.5*x**2/inner**2) filt1=filt1+np.exp(-0.5*(x-n)**2/inner**2); filt1=filt1/filt1.sum() filt2=np.exp(-0.5*x**2/outer**2) filt2=filt2+np.exp(-0.5*(x-n)**2/outer**2); filt2=filt2/filt2.sum() filt=filt1-filt2 #make a filter that is the difference of two Gaussians, one narrow, one wide filtft=np.fft.rfft(filt) datft=np.fft.rfft(dat,axis=1) datfilt=np.fft.irfft(filtft*datft,axis=1,n=n) jumps=[None]*ndet mystd=np.median(np.abs(datfilt),axis=1) for i in range(ndet): while np.max(np.abs(datfilt[i,:]))>thresh*mystd[i]: ind=np.argmax(np.abs(datfilt[i,:])) if jumps[i] is None: jumps[i]=[ind] else: jumps[i].append(ind) datfilt[i,ind]=0 return jumps,datfilt return mystd def make_rings_wSlope(edges,cent,vals,map,pixsize=2.0,fwhm=10.0,amps=None,aa=1.0,bb=1.0,rot=0.0): xvec=np.arange(map.nx) yvec=np.arange(map.ny) xvec[map.nx//2:]=xvec[map.nx//2:]-map.nx yvec[map.ny//2:]=yvec[map.ny//2:]-map.ny xmat=np.repeat([xvec],map.ny,axis=0).transpose() ymat=np.repeat([yvec],map.nx,axis=0) rmat=np.sqrt(xmat**2+ymat**2)*pixsize if isinstance(fwhm,int)|isinstance(fwhm,float): sig=fwhm/np.sqrt(8*np.log(2.)) src_map=np.exp(-0.5*rmat**2./sig**2) src_map=src_map/src_map.sum() else: sig=fwhm[0]/np.sqrt(8*np.log(2)) src_map=np.exp(-0.5*rmat**2/sig**2)*amps[0] for i in range(1,len(fwhm)): sig=fwhm[i]/np.sqrt(8*np.log(2)) src_map=src_map+np.exp(-0.5*rmat**2/sig**2)*amps[i] src_map=src_map/src_map.sum() beam_area=pixsize**2/src_map.max() beam_area=beam_area/3600**2/(360**2/np.pi) print('beam_area is ',beam_area*1e9,' nsr') nring=len(edges)-1 rings=np.zeros([nring,map.nx,map.ny]) mypix=map.wcs.wcs_world2pix(cent[0],cent[1],1) print('mypix is ',mypix) xvec=np.arange(map.nx) yvec=np.arange(map.ny) xmat=np.repeat([xvec],map.ny,axis=0).transpose() ymat=np.repeat([yvec],map.nx,axis=0) srcft=np.fft.fft2(src_map) xtr = (xmat-mypix[0])*np.cos(rot) + (ymat-mypix[1])*np.sin(rot) # Rotate and translate x coords ytr = (ymat-mypix[1])*np.cos(rot) - (xmat-mypix[0])*np.sin(rot) # Rotate and translate y coords rmat = np.sqrt( (xtr/aa)**2 + (ytr/bb)**2 ) * pixsize # Elliptically scale x,y myvals = vals[:nring]*1.0 # Get just the values that correspond to rings myvals -= np.max(myvals) # Set it such that the maximum value approaches 0 pk2pk = np.max(myvals) - np.min(myvals) myvals -= pk2pk/50.0 # Let's assume we're down about a factor of 50 at the outskirts. for i in range(nring): #rings[i,(rmat>=edges[i])&(rmat<edges[i+1]=1.0 if i == nring-1: slope=0.0 else: slope = (myvals[i]-myvals[i+1])/(edges[i+1]-edges[i]) # expect positve slope; want negative one. rgtinedge = (rmat>=edges[i]) rfromin = (rmat-edges[i]) initline = rfromin[rgtinedge]*slope if vals[i] != 0: rings[i,rgtinedge] = (myvals[i] - initline)/myvals[i] # Should be normalized to 1 now. else: rings[i,rgtinedge] = 1.0 rgtoutedge = (rmat>=edges[i+1]) rings[i,rgtoutedge]=0.0 myannul = [ c1 and not(c2) for c1,c2 in zip(rgtinedge.ravel(),rgtoutedge.ravel())] rannul = rmat.ravel()[myannul] rmin = (rmat == np.min(rannul)) rmout = (rmat == np.max(rannul)) rings[i,:,:]=np.real(np.fft.ifft2(np.fft.fft2(rings[i,:,:])*srcft)) return rings def make_rings(edges,cent,map,pixsize=2.0,fwhm=10.0,amps=None,iswcs=True): xvec=np.arange(map.nx) yvec=np.arange(map.ny) ix=int(map.nx/2) iy=int(map.ny/2) xvec[ix:]=xvec[ix:]-map.nx yvec[iy:]=yvec[iy:]-map.ny #xvec[map.nx/2:]=xvec[map.nx/2:]-map.nx #yvec[map.ny/2:]=yvec[map.ny/2:]-map.ny xmat=np.repeat([xvec],map.ny,axis=0).transpose() ymat=np.repeat([yvec],map.nx,axis=0) rmat=np.sqrt(xmat**2+ymat**2)*pixsize if isinstance(fwhm,int)|isinstance(fwhm,float): sig=fwhm/np.sqrt(8*np.log(2)) src_map=np.exp(-0.5*rmat**2/sig**2) src_map=src_map/src_map.sum() else: sig=fwhm[0]/np.sqrt(8*np.log(2)) src_map=np.exp(-0.5*rmat**2/sig**2)*amps[0] for i in range(1,len(fwhm)): sig=fwhm[i]/np.sqrt(8*np.log(2)) src_map=src_map+np.exp(-0.5*rmat**2/sig**2)*amps[i] src_map=src_map/src_map.sum() beam_area=pixsize**2/src_map.max() beam_area=beam_area/3600**2/(360**2/np.pi) print('beam_area is ',beam_area*1e9,' nsr') nring=len(edges)-1 rings=np.zeros([nring,map.nx,map.ny]) if iswcs: mypix=map.wcs.wcs_world2pix(cent[0],cent[1],1) else: mypix=cent print('mypix is ',mypix) xvec=np.arange(map.nx) yvec=np.arange(map.ny) xmat=np.repeat([xvec],map.ny,axis=0).transpose() ymat=np.repeat([yvec],map.nx,axis=0) srcft=np.fft.fft2(src_map) rmat=np.sqrt( (xmat-mypix[0])**2+(ymat-mypix[1])**2)*pixsize for i in range(nring): #rings[i,(rmat>=edges[i])&(rmat<edges[i+1]=1.0 rings[i,(rmat>=edges[i])]=1.0 rings[i,(rmat>=edges[i+1])]=0.0 rings[i,:,:]=np.real(np.fft.ifft2(np.fft.fft2(rings[i,:,:])*srcft)) return rings def find_jumps(dat,width=10,pad=2,thresh=10,rat=0.5): #find jumps in a block of timestreams, preferably with the common mode removed #width is width in pixels to average over when looking for a jump #pad is the length in units of width to mask at beginning/end of timestream #thresh is threshold in units of filtered data median absolute deviation to qualify as a jump #rat is the ratio of largest neighboring opposite-sign jump to the found jump. If # there is an opposite-sign jump nearby, the jump finder has probably just picked up a spike. n=dat.shape[1] ndet=dat.shape[0] #make a filter template that is a gaussian with sigma with, sign-flipped in the center #so, positive half-gaussian starting from zero, and negative half-gaussian at the end x=np.arange(n) myfilt=np.exp(-0.5*x**2/width**2) myfilt=myfilt-np.exp( (-0.5*(x-n)**2/width**2)) fac=np.abs(myfilt).sum()/2.0 myfilt=myfilt/fac dat_filt=np.fft.rfft(dat,axis=1) myfilt_ft=np.fft.rfft(myfilt) dat_filt=dat_filt*np.repeat([myfilt_ft],ndet,axis=0) dat_filt=np.fft.irfft(dat_filt,axis=1,n=n) dat_filt_org=dat_filt.copy() print(dat_filt.shape) dat_filt[:,0:pad*width]=0 dat_filt[:,-pad*width:]=0 det_thresh=thresh*np.median(np.abs(dat_filt),axis=1) dat_dejump=dat.copy() jumps=[None]*ndet print('have filtered data, now searching for jumps') for i in range(ndet): while np.max(np.abs(dat_filt[i,:]))>det_thresh[i]: ind=np.argmax(np.abs(dat_filt[i,:]))+1 #+1 seems to be the right index to use imin=ind-width if imin<0: imin=0 imax=ind+width if imax>n: imax=n val=dat_filt[i,ind] if val>0: val2=np.min(dat_filt[i,imin:imax]) else: val2=np.max(dat_filt[i,imin:imax]) print('found jump on detector ',i,' at sample ',ind) if np.abs(val2/val)>rat: print('I think this is a spike due to ratio ',np.abs(val2/val)) else: if jumps[i] is None: jumps[i]=[ind] else: jumps[i].append(ind) #independent of if we think it is a spike or a jump, zap that stretch of the data dat_dejump[i,ind:]=dat_dejump[i,ind:]+dat_filt[i,ind] dat_filt[i,ind-pad*width:ind+pad*width]=0 if not(jumps[i] is None): jumps[i]=np.sort(jumps[i]) #return dat_dejump,jumps,dat_filt_org return jumps def fit_jumps_from_cm(dat,jumps,cm,cm_order=1,poly_order=1): jump_vals=jumps[:] ndet=len(jumps) n=dat.shape[1] x=np.linspace(-1,1,n) m1=np.polynomial.legendre.legvander(x,poly_order) m2=np.polynomial.legendre.legvander(x,cm_order-1) for i in range(cm_order): m2[:,i]=m2[:,i]*cm mat=np.append(m1,m2,axis=1) npp=mat.shape[1] dat_dejump=dat.copy() for i in range(ndet): if not(jumps[i] is None): njump=len(jumps[i]) segs=np.append(jumps[i],n) print('working on detector ',i,' who has ', len(jumps[i]),' jumps with segments ',segs) mm=np.zeros([n,npp+njump]) mm[:,:npp]=mat for j in range(njump): mm[segs[j]:segs[j+1],j+npp]=1.0 lhs=np.dot(mm.transpose(),mm) #print lhs rhs=np.dot(mm.transpose(),dat[i,:].transpose()) lhs_inv=np.linalg.inv(lhs) fitp=np.dot(lhs_inv,rhs) jump_vals[i]=fitp[npp:] jump_pred=np.dot(mm[:,npp:],fitp[npp:]) dat_dejump[i,:]=dat_dejump[i,:]-jump_pred return dat_dejump #for i in range(ndet): def gapfill_eig(dat,cuts,tod=None,thresh=5.0, niter_eig=3, niter_inner=3, insert_cuts=False): ndat=dat.shape[1] cuts_empty=cuts.copy() #use this to clear out cut samples cuts_empty.clear() cuts_cur=cuts.copy() cuts_cur.clear() for eig_ctr in range(niter_eig): tmp=dat.copy() cuts_cur.map2tod(tod,tmp,do_add=False) mycov=np.dot(tmp,tmp.T) ee,vv=np.linalg.eig(mycov) mask=ee>thresh*thresh*np.median(ee) neig=np.sum(mask) print('working with ' + repr(neig) + ' eigenvectors.') ee=ee[mask] vv=vv[:,mask] uu=np.dot(vv.T,tmp) lhs=np.dot(uu,uu.T) lhs_inv=np.linalg.inv(lhs) for iter_ctr in range(niter_inner): #in this inner loop, we fit the data rhs=np.dot(tmp,uu.T) fitp=np.dot(lhs_inv,rhs.T) pred=np.dot(fitp.T,uu) cuts_cur.tod2map(tod,pred,do_add=False) cuts_cur.map2tod(tod,tmp,do_add=False) if insert_cuts: cuts_cur.map2tod(dat) return cuts_cur def __gapfill_eig_poly(dat,cuts,tod=None,npoly=2, thresh=5.0, niter_eig=3, niter_inner=3): assert(1==0) #this code is not yet working. regular gapfill_eig should work since the polys could #be described by SVD, so SVD modes should look like polys iff they would have been important ndat=dat.shape[1] if npoly>0: xvec=np.linspace(-1,1,ndat) polymat=np.polynomial.legendre.legvander(x,npoly-1) old_coeffs=None cuts_cur=cuts.copy() cuts_cur.clear() cuts_empty.cuts.copy() cuts_empty.clear() for eig_ctr in range(niter_eig): tmp=dat.copy() cuts_cur.map2tod(tod,tmp,do_add=False) #insert current best-guess solution for the cuts if npoly>1: #if we're fitting polynomials as well as eigenmodes, subtract them off before re-estimating the covariance if not(old_coeffs is None): tmp=tmp-np.dot(polymat,old_coeffs[neig:,:]).T mycov=np.dot(tmp,tmp.T) mycov=0.5*(mycov+mycov.T) ee,vv=np.linalg.eig(mycov) mode_map=ee>thresh*thresh*np.median(ee) neig=mode_map.sum() mat=np.zeros([ndat,neig+npoly]) eigs=vv[:,mode_map] ts_vecs=np.dot(eigs.T,tmp) mat[:,:neig]=ts_vecs.T if npoly>0: mat[:,neig:]=polymat lhs=np.dot(mat.T,mat) lhs_inv=np.linalg.inv(lhs) #now that we have the vectors we expect to describe our data, do a few rounds #of fitting amplitudes to timestream models, subtract that off, assign cuts to zero, #and restore the model. tmp=dat.copy() for inner_ctr in range(niter_inner): cuts_cur.map2tod(tod,tmp) rhs=np.dot(tmp,mat) fitp=np.dot(lhs_inv,rhs.T) pred=np.dot(mat,fitp).T def get_type(nbyte): if nbyte==8: return np.dtype('float64') if nbyte==4: return np.dtype('float32') if nbyte==-4: return np.dtype('int32') if nbyte==-8: return np.dtype('int64') if nbyte==1: return np.dtype('str') print('Unsupported nbyte ' + repr(nbyte) + ' in get_type') return None def read_octave_struct(fname): f=open(fname) nkey=np.fromfile(f,'int32',1)[0] #print 'nkey is ' + repr(nkey) dat={} for i in range(nkey): key=f.readline().strip() #print 'key is ' + key ndim=np.fromfile(f,'int32',1)[0] dims=np.fromfile(f,'int32',ndim) dims=np.flipud(dims) #print 'Dimensions of ' + key + ' are ' + repr(dims) nbyte=np.fromfile(f,'int32',1)[0] #print 'nbyte is ' + repr(nbyte) dtype=get_type(nbyte) tmp=np.fromfile(f,dtype,dims.prod()) dat[key]=np.reshape(tmp,dims) f.close() return dat def nsphere_vol(npp): iseven=(npp%2)==0 if iseven: nn=npp/2 vol=(np.pi**nn)/np.prod(np.arange(1,nn+1)) else: nn=(npp-1)/2 vol=2**(nn+1)*np.pi**nn/np.prod(np.arange(1,npp+1,2)) return vol def _prime_loop(ln,lp,icur,lcur,vals): facs=np.arange(lcur,ln+1e-3,lp[0]) if len(lp)==1: nfac=len(facs) if (nfac>0): vals[icur:(icur+nfac)]=facs icur=icur+nfac #print 2**vals[:icur] else: print('bad facs came from ' + repr([2**lcur,2**ln,2**lp[0]])) #print icur return icur else: facs=np.arange(lcur,ln,lp[0]) for fac in facs: icur=_prime_loop(ln,lp[1:],icur,fac,vals) return icur print('I don''t think I should have gotten here.') return icur def find_good_fft_lens(n,primes=[2,3,5,7]): lmax=np.log(n+0.5) npr=len(primes) vol=nsphere_vol(npr) r=np.log2(n+0.5) lp=np.log2(primes) int_max=(vol/2**npr)*np.prod(r/lp)+30 #add a bit just to make sure we don't act up for small n #print 'int max is ',int max int_max=int(int_max) #vals=np.zeros(int_max,dtype='int') vals=np.zeros(int_max) icur=0 icur=_prime_loop(r,lp,icur,0.0,vals) assert(icur<=int_max) myvals=np.asarray(np.round(2**vals[:icur]),dtype='int') myvals=np.sort(myvals) return myvals def _linfit_2mat(dat,mat1,mat2): np1=mat1.shape[1] np2=mat2.shape[1] mm=np.append(mat1,mat2,axis=1) lhs=np.dot(mm.transpose(),mm) rhs=np.dot(mm.transpose(),dat) lhs_inv=np.linalg.inv(lhs) fitp=np.dot(lhs_inv,rhs) fitp1=fitp[0:np1].copy() fitp2=fitp[np1:].copy() assert(len(fitp2)==np2) return fitp1,fitp2 def fit_mat_vecs_poly_nonoise(dat,mat,order,cm_order=None): if cm_order is None: cm_order=order n=dat.shape[1] x=np.linspace(-1,1,n) polys=np.polynomial.legendre.legvander(x,order).transpose() cm_polys=np.polynomial.legendre.legvander(x,cm_order).transpose() v1=np.sum(dat,axis=0) v2=np.sum(dat*mat,axis=0) rhs1=np.dot(cm_polys,v1) rhs2=np.dot(polys,v2) ndet=dat.shape[0] A1=cm_polys*ndet vv=np.sum(mat,axis=0) A2=polys*np.repeat([vv],order+1,axis=0) A=np.append(A1,A2,axis=0) rhs=np.append(rhs1,rhs2) lhs=np.dot(A,A.transpose()) fitp=np.dot(np.linalg.inv(lhs),rhs) cm_fitp=fitp[:cm_order+1] mat_fitp=fitp[cm_order+1:] assert(len(mat_fitp)==(order+1)) cm_pred=np.dot(cm_fitp,cm_polys) tmp=np.dot(mat_fitp,polys) mat_pred=np.repeat([tmp],ndet,axis=0)*mat pred=cm_pred+mat_pred return pred,cm_fitp,mat_fitp,polys def smooth_spectra(spec,fwhm): nspec=spec.shape[0] n=spec.shape[1] x=np.arange(n) sig=fwhm/np.sqrt(8*np.log(2)) to_conv=np.exp(-0.5*(x/sig)**2) tot=to_conv[0]+to_conv[-1]+2*to_conv[1:-1].sum() #r2r normalization to_conv=to_conv/tot to_conv_ft=mkfftw.fft_r2r(to_conv) xtrans=mkfftw.fft_r2r(spec) for i in range(nspec): xtrans[i,:]=xtrans[i,:]*to_conv_ft #return mkfftw.fft_r2r(xtrans)/(2*(xtrans.shape[1]-1)),to_conv return xtrans,to_conv_ft def smooth_many_vecs(vecs,fwhm=20): n=vecs.shape[1] nvec=vecs.shape[0] x=np.arange(n) sig=fwhm/np.sqrt(8*np.log(2)) to_conv=np.exp(-0.5*(x/sig)**2) tot=to_conv[0]+to_conv[-1]+2*to_conv[1:-1].sum() #r2r normalization to_conv=to_conv/tot to_conv_ft=mkfftw.fft_r2r(to_conv) xtrans=mkfftw.fft_r2r(vecs) for i in range(nvec): xtrans[i,:]=xtrans[i,:]*to_conv_ft back=mkfftw.fft_r2r(xtrans) return back/(2*(n-1)) def smooth_vec(vec,fwhm=20): n=vec.size x=np.arange(n) sig=fwhm/np.sqrt(8*np.log(2)) to_conv=np.exp(-0.5*(x/sig)**2) tot=to_conv[0]+to_conv[-1]+2*to_conv[1:-1].sum() #r2r normalization to_conv=to_conv/tot to_conv_ft=mkfftw.fft_r2r(to_conv) xtrans=mkfftw.fft_r2r(vec) back=mkfftw.fft_r2r(xtrans*to_conv_ft) return back/2.0/(n-1) def fit_cm_plus_poly(dat,ord=2,cm_ord=1,niter=2,medsub=False,full_out=False): n=dat.shape[1] ndet=dat.shape[0] if medsub: med=np.median(dat,axis=1) dat=dat-np.repeat([med],n,axis=0).transpose() xx=np.arange(n)+0.0 xx=xx-xx.mean() xx=xx/xx.max() pmat=np.polynomial.legendre.legvander(xx,ord) cm_pmat=np.polynomial.legendre.legvander(xx,cm_ord-1) calfacs=np.ones(ndet)*1.0 dd=dat.copy() for i in range(1,niter): for j in range(ndet): dd[j,:]/=calfacs[j] cm=np.median(dd,axis=0) cm_mat=np.zeros(cm_pmat.shape) for i in range(cm_mat.shape[1]): cm_mat[:,i]=cm_pmat[:,i]*cm fitp_p,fitp_cm=_linfit_2mat(dat.transpose(),pmat,cm_mat) pred1=np.dot(pmat,fitp_p).transpose() pred2=np.dot(cm_mat,fitp_cm).transpose() pred=pred1+pred2 dd=dat-pred1 if full_out: return dd,pred2,cm #if requested, return the modelled CM as well return dd def run_pcg(b,x0,tods,precon=None,maxiter=25,outroot='map',save_iters=[-1],save_ind=0,save_tail='.fits',plot_iters=[],plot_info=None,plot_ind=0): """ Function which runs preconditioned conjugate gradient on a bundle of tods to generate a map. PCG itteratively approximates the solution to the linear equation Ax = b for A a matrix, x and b vectors. In the map making equation, A = P'N"P and b = P'N"d for d the vector of TODs, N the noise matrix, P the tod to map pointing matrix, i.e. a matrix that specifies which pixel in the map was observed by each TOD data point. Futher x is the map. Arguments: b: The rhs of the equation. In our case this is P'N''d. The tod class has a built in method for computing this. x0: The initial guess. Generally set to for the first itteration and then to the output of the previous itteration. tods: the input tods we want to make into maps. Note the noise has already been estimated and is within the tod object. precon: The preconditioner. A matrix applied to A to ensure faster convergence. 1/hitsmap is a frequent selection. maxiter: Maximum number of iterations to perform. outroot: location at which to save the output map save_iters: The iterations at which to save the result map. Default is to save only the last save_ind: save_tail: Extention for saving the output maps plot_iters: Which iterations to plot plot_info: plot_ind: Outputs: x: best guess for x after the conversion criteria has been reached (either max iter or Ax = b close enough to 0 """ t1=time.time() Ax=tods.dot(x0) try: #compute the remainder r_0 r=b.copy() r.axpy(Ax,-1) except: r=b-Ax if not(precon is None): #print('applying precon') # z_0 = M*r_0 z=precon*r key = tods.tods[0].info['fname'] else: z=r.copy() #Initial p_0 = z_0 = M*r_0 p=z.copy() k=0.0 #compute z*r, which is used for computing alpha zr=r.dot(z) #make a copy of our initial guess x=x0.copy() t2=time.time() nsamp=tods.get_nsamp() tloop=time.time() for iter in range(maxiter): if myrank==0: if iter>0: print(iter,zr,alpha,t2-t1,t3-t2,t3-t1,nsamp/(t2-t1)/1e6) else: print(iter,zr,t2-t1) t1=time.time() #Compute pAp Ap=tods.dot(p) t2=time.time() pAp=p.dot(Ap) #Compute alpha_k alpha=zr/pAp #print('alpha,pAp, and zr are ' + repr(alpha) + ' ' + repr(pAp) + ' ' + repr(zr)) try: #Update guess using alpha x_new=x.copy() x_new.axpy(p,alpha) except: x_new=x+p*alpha try: #Write down next remainder r_k+1 r_new=r.copy() r_new.axpy(Ap,-alpha) except: r_new=r-Ap*alpha if not(precon is None): #print('applying precon') z_new=precon*r_new else: z_new=r_new.copy() #compute new z_k+1 zr_new=r_new.dot(z_new) #compute beta_k, which is used to compute p_k+1 beta=zr_new/zr try: #compute new p_k+1 p_new=z_new.copy() p_new.axpy(p,beta) except: p_new=z_new+p*beta #Update values p=p_new z=z_new r=r_new zr=zr_new x=x_new t3=time.time() if iter in save_iters: if myrank==0: x.maps[save_ind].write(outroot+'_'+repr(iter)+save_tail) if iter in plot_iters: print('plotting on iteration ',iter) x.maps[plot_ind].plot(plot_info) tave=(time.time()-tloop)/maxiter print('average time per iteration was ',tave,' with effective throughput ',nsamp/tave/1e6,' Msamp/s') if iter in plot_iters: print('plotting on iteration ',iter) x.maps[plot_ind].plot(plot_info) else: print('skipping plotting on iter ',iter) return x def run_pcg_wprior(b,x0,tods,prior=None,precon=None,maxiter=25,outroot='map',save_iters=[-1],save_ind=0,save_tail='.fits'): #least squares equations in the presence of a prior - chi^2 = (d-Am)^T N^-1 (d-Am) + (p-m)^T Q^-1 (p-m) #where p is the prior target for parameters, and Q is the variance. The ensuing equations are #(A^T N-1 A + Q^-1)m = A^T N^-1 d + Q^-1 p. For non-zero p, it is assumed you have done this already and that #b=A^T N^-1 d + Q^-1 p #to have a prior then, whenever we call Ax, just a Q^-1 x to Ax. t1=time.time() Ax=tods.dot(x0) if not(prior is None): #print('applying prior') prior.apply_prior(x0,Ax) try: r=b.copy() r.axpy(Ax,-1) except: r=b-Ax if not(precon is None): z=precon*r else: z=r.copy() p=z.copy() k=0.0 zr=r.dot(z) x=x0.copy() t2=time.time() for iter in range(maxiter): if myrank==0: if iter>0: print(iter,zr,alpha,t2-t1,t3-t2,t3-t1) else: print(iter,zr,t2-t1) sys.stdout.flush() t1=time.time() Ap=tods.dot(p) if not(prior is None): #print('applying prior') prior.apply_prior(p,Ap) t2=time.time() pAp=p.dot(Ap) alpha=zr/pAp try: x_new=x.copy() x_new.axpy(p,alpha) except: x_new=x+p*alpha try: r_new=r.copy() r_new.axpy(Ap,-alpha) except: r_new=r-Ap*alpha if not(precon is None): z_new=precon*r_new else: z_new=r_new.copy() zr_new=r_new.dot(z_new) beta=zr_new/zr try: p_new=z_new.copy() p_new.axpy(p,beta) except: p_new=z_new+p*beta p=p_new z=z_new r=r_new zr=zr_new x=x_new t3=time.time() if iter in save_iters: if myrank==0: x.maps[save_ind].write(outroot+'_'+repr(iter)+save_tail) return x def apply_noise(tod,dat=None): if dat is None: #dat=tod['dat_calib'] dat=tod.get_data().copy() dat_rot=np.dot(tod['v'],dat) datft=mkfftw.fft_r2r(dat_rot) nn=datft.shape[1] datft=datft*tod['mywt'][:,0:nn] dat_rot=mkfftw.fft_r2r(datft) dat=np.dot(tod['v'].transpose(),dat_rot) #for fft_r2r, the first/last samples get counted for half of the interior ones, so #divide them by 2 in the post-filtering. Makes symmetry much happier... #print 'hello' dat[:,0]=dat[:,0]*0.5 dat[:,-1]=dat[:,-1]*0.5 return dat def get_grad_mask_2d(map,todvec=None,thresh=4.0,noisemap=None,hitsmap=None): """make a mask that has an estimate of the gradient within a pixel. Look at the rough expected noise to get an idea of which gradients are substantially larger than the map noise.""" if noisemap is None: noisemap=make_hits(todvec,map,do_weights=True) noisemap.invert() noisemap.map=np.sqrt(noisemap.map) if hitsmap is None: hitsmap=make_hits(todvec,map,do_weights=False) mygrad=(map.map-np.roll(map.map,1,axis=0))**2 mygrad=mygrad+(map.map-np.roll(map.map,-1,axis=0))**2 mygrad=mygrad+(map.map-np.roll(map.map,-1,axis=1))**2 mygrad=mygrad+(map.map-np.roll(map.map,1,axis=1))**2 mygrad=np.sqrt(0.25*mygrad) #find the typical timestream noise in a pixel, which should be the noise map times sqrt(hits) hitsmask=hitsmap.map>0 tmp=noisemap.map.copy() tmp[hitsmask]=tmp[hitsmask]*np.sqrt(hitsmap.map[hitsmask]) #return mygrad,tmp mask=(mygrad>(thresh*tmp)) frac=1.0*np.sum(mask)/mask.size print("Cutting " + repr(frac*100) + "% of map pixels in get_grad_mask_2d.") mygrad[np.logical_not(mask)]=0 #return mygrad,tmp,noisemap return mygrad class null_precon: def __init__(self): self.isnull=True def __add__(self,val): return val def __mul__(self,val): return val def scaled_airmass_from_el(mat): airmass=1/np.cos(mat) airmass=airmass-airmass.mean() #airmass=airmass/np.std(airmass) return airmass class tsGeneric: """ Generic timestream model class. Used as a parent for other timestream/map classes. Defines multiple common methods """ def __init__(self,tod=None): #Set file name if tod specified self.fname=tod.info['fname'] def __mul__(self,to_mul): #print('calling mul') #multiplies two timeseries. Returns a a new ts class with the multiplied parameters tt=self.copy() tt.params=self.params*to_mul.params return tt def clear(self): #clears parameters self.params[:]=0 def dot(self,common=None): #Returns the dot product of a ts class. If common is not specified, returns the self dot product, else returns the dot product with common if common is None: return np.sum(self.params*self.params) else: return np.sum(self.params*common.params) def axpy(self,common,a): #returns ts + a*common self.params=self.params+a*common.params def apply_prior(self,x,Ax): Ax.params=Ax.params+self.params*x.params def copy(self): return copy.deepcopy(self) def write(self,fname=None): pass class tsVecs(tsGeneric): """ Generic class for timestreams involving vectors, including tools to go from parameters (maps) to tods and back. Example would be fitting polynomials to timestreams. In this case the self.vecs would be the fit polynomials, self.params are the fit parameters of that polynomial. Then map2tod returns the tods predicted by the polynomials and fit parameters, i.e. self.vec@self.params, while tod2map returns the fit parameters that would generate a given tod. Attributes ---------- fname: str name of the tod tod: tod object tods corresponding to the timestream vecs: n_data x nvec matrix vectors to which fit parameters are applied. ndet: int number of detectors in timestream nvec: int number of fit vectors params: np.array, float, nvec x ndet fit parameters for the vecs. """ def __init__(self,tod,vecs): """ Parameters ---------- tod: tod object tod corresponding to the timestream vecs: n_data x n_predictors matrix vectors to which fit parameters are applied. """ self.vecs=vecs #self.ndet=tod.info['dat_calib'].shape[0] self.ndet=tod.get_data_dims()[0] self.vecs=vecs self.nvec=vecs.shape[0] self.params=np.zeros([self.nvec,self.ndet]) def tod2map(self,tod,mat=None,do_add=True,do_omp=False): """ Computes the parameters of vecs which yield the given tod. In general Am = d for A the vecs, m the parameters, and d the tod. tod2map returns m given A and d. Essentially the inverse of map2tod. Parameters ---------- tod: tod object The tod to convert to parameters mat: tod data object, optional d, the data corresponding to the tod we wish to convert to parameters. If not speicified the data is taken from tod do_add: bool, optional, default = True If true, adds the resulting parameters matrix to the existing parameters matrix. If false, overwrites the existing parameters do_omp: bool, optional, default = False Defines whether to use omp parallelization. Currently not implemented (?) Returns ------- No returns Side effects ------------ Updates params with the values infered from tod or mat. """ if mat is None: #mat=tod.info['dat_calib'] mat=tod.get_data() if do_add: self.params[:]=self.params[:]+np.dot(self.vecs,mat.T) else: self.params[:]=np.dot(self.vecs,mat.T) def map2tod(self,tod,mat=None,do_add=True,do_omp=False): """ Given parameters and vecs, compute the corresponding tod. Given Am = for A the vecs, m the parameters, and d the tod data, return the d corresponding to the specified A and m. Essentially the inverse of tod2map. Parameters ---------- tod: tod object The tod, needed for getting the data if to_add is True mat: tod data object, optional d, the data corresponding to the tod, only used if to_add is True. If not speicified the data is taken from tod do_add: bool, optional, default = True If true, adds the resulting tod data to the existing tod data. If false, overwrites the tod data. do_omp: bool, optional, default = False Defines whether to use omp parallelization. Currently not implemented (?) Returns ------- No returns Side effects ------------ Updates tod data with the values infered from params and vecs. """ if mat is None: #mat=tod.info['dat_calib'] mat=tod.get_data() if do_add: mat[:]=mat[:]+np.dot(self.params.T,self.vecs) else: mat[:]=np.dot(self.params.T,self.vecs) class tsNotch(tsGeneric): def __init__(self,tod,numin,numax): self.fname=tod.info['fname'] tvec=tod.get_tvec() dt=tvec[-1]-tvec[0] bw=numax-numin dnu=1/dt nfreq=int(np.ceil(2*bw/dnu)) #factor of 2 is to account for partial waves ndet=tod.get_ndet() self.freqs=np.linspace(numin,numax,nfreq) self.nfreq=nfreq self.params=np.zeros([2*nfreq,ndet]) def get_vecs(self,tvec): tvec=tvec-tvec[0] vecs=np.zeros([self.nfreq*2,len(tvec)]) for i in range(self.nfreq): vecs[2*i,:]=np.cos(tvec*self.freqs[i]) vecs[2*i+1,:]=np.sin(tvec*self.freqs[i]) return vecs def map2tod(self,tod,mat=None,do_add=True,do_omp=False): tvec=tod.get_tvec() vecs=self.get_vecs(tvec) pred=self.params.T@vecs if mat is None: mat=tod.get_data() if do_add: mat[:]=mat[:]+pred else: mat[:]=pred def tod2map(self,tod,mat=None,do_add=True,do_omp=False): tvec=tod.get_tvec() vecs=self.get_vecs(tvec) if mat is None: mat=tod.get_data() #tmp=mat@(vecs.T) tmp=vecs@mat.T if do_add: self.params[:]=self.params[:]+tmp else: self.params[:]=tmp class tsPoly(tsVecs): """ Class for fitting legandre polynomials to tods. Inheritted from tsVecs. Attributes ---------- fname: str name of the tod tod: tod object tods corresponding to the timestream vecs: n_data x nvec matrix Legandre polynomials to which fit parameters are applied. ndet: int number of detectors in timestream nvec: int number of fit polynomials params: np.array, float, nvec x ndet fit parameters for the vecs. """ def __init__(self,tod,order=10): """Inherits directly from tsVecs. Methods and arguments are the same, the changes are only to how vecs is defined. Parameters ---------- tod: tod object tod corresponding to the timestream order: int order of the legandre polynomial to fit to the tod """ self.fname=tod.info['fname'] #self.ndata=tod.info['dat_calib'].shape[1] dims=tod.get_data_dims() self.ndata=dims[1] self.order=order #self.ndet=tod.info['dat_calib'].shape[0] self.ndet=dims[0] xvec=np.linspace(-1,1,self.ndata) self.vecs=(np.polynomial.legendre.legvander(xvec,order).T).copy() self.nvec=self.vecs.shape[0] self.params=np.zeros([self.nvec,self.ndet]) class tsBowl(tsVecs): """ Class for fitting legandre polynomials to tod elevation. Mustang 2 has observed a consistent problem with gradients in its maps, refered to as bowling. Current thinking is that the ultimate source of the bowling is elevation depended gradients due to the atmosphere. As the sky revolves around a target thru the course of a night, this elevation gradient becomes a radial one. Previous attempts to remove this have fit polynomials to the telescope elevation and subtracted them from the tods, which has been moderately successful. This was done outside the minkasi framework; this class implements that method within the framework. Attributes ---------- fname: str name of the tod tod: tod object tods corresponding to the timestream vecs: n_data x nvec matrix Legandre polynomials to which fit parameters are applied. ndet: int number of detectors in timestream nvec: int number of fit polynomials params: np.array, float, nvec x ndet fit parameters for the vecs. """ def __init__(self, tod, order=3): """ Inherits directly from tsVecs. Methods and arguments are the same, the changes are only to how vecs is defined. Parameters ---------- tod: tod object tod corresponding to the timestream order: int order of the legandre polynomial to fit to the tod elevation """ self.fname=tod.info['fname'] self.order = order dims=tod.get_data_dims() self.ndet=dims[0] self.ndata=dims[1] try: #Apix is the elevation relative to the source. Try to just load it first incase its already computed, otherwise compute it then set it self.apix = tod.info['apix'] except KeyError: tod.set_apix() self.apix = tod.info['apix'] #Normalize apix to run from -1 to 1, to preserve linear independce of leg polys self.apix /= np.max(np.abs(self.apix), axis = 0) #TODO: swap legvander to legval #Array(len(apix), order) of the legander polynomials evaluated at self.apix self.vecs=(np.polynomial.legendre.legvander(self.apix,order)).copy() self.nvec=self.vecs.shape[-1] #Parameters c_ij for the legandre polynomials self.params=np.zeros([self.ndet,self.nvec]) def map2tod(self, tod, mat = None, do_add = True, do_omp = False): """ Given parameters and vecs, compute the corresponding tod. Given Am = for A the vecs, m the parameters, and d the tod data, return the d corresponding to the specified A and m. This will try to use the jit compiled map2todbowl if it is available, else it will fall back to a slower routine. Parameters ---------- tod: tod object The tod, needed for getting the data if to_add is True mat: tod data object, optional d, the data corresponding to the tod, only used if to_add is True. If not speicified the data is taken from tod do_add: bool, optional, default = True If true, adds the resulting tod data to the existing tod data. If false, overwrites the tod data. do_omp : bool, optional Dummy variable to match parameters of other map2tod Returns ------- No returns Side effects ------------ Updates tod data with the values infered from params and vecs. """ if mat is None: mat=tod.get_data() if do_add: mat[:]=mat[:] + map2todbowl(self.vecs, self.params) else: mat[:]=map2todbowl(self.vecs, self.params) def tod2map(self, tod, mat = None, do_add = True, do_omp = False): """ Given legandre vecs and TOD data, computes the corresponding parameters. tod2map is the transpose of the linear transformation map2tod. This will use the jit compiled tod2mapbowl if available, else it will fall back on a slower routine. Parameters ---------- tod : 'minkasi.TOD' minkasi tod object mat : np.array, optional data from tod. If not specified, mat is taken from passed tod do_add : bool If true, adds resulting param to existing param. If false, overwrites it. do_omp : bool Dummy variable to match parameters of other tod2map Returns ------- No returns Side effects ------------ Updates params with the values infered from mat """ if mat is None: mat = tod.get_data() if do_add: self.params = self.params + tod2mapbowl(self.vecs, mat) else: self.paras = tod2mapbowl(self.vecs, mat) def fit_apix(self, tod): if tod.info['fname'] != self.fname: print('Error: bowling fitting can only be performed with the tod used to initialize this timestream; {}'.format(tod.info['fname'])) return for i in range(self.ndet): self.params[i,...] = np.polynomial.legendre.legfit(self.apix[i], tod.info['dat_calib'][i] - self.drift[i], self.order) def partition_interval(start,stop,seg_len=100,round_up=False): #print('partitioning ',start,stop,seg_len) #make sure we behave correctly if the interval is shorter than the desired segment if (stop-start)<=seg_len: return np.asarray([start,stop],dtype='int') nseg=(stop-start)//seg_len if nseg*seg_len<(stop-start): if round_up: nseg=nseg+1 seg_len=(stop-start)//nseg nextra=(stop-start)-seg_len*nseg inds=np.arange(start,stop+1,seg_len) if nextra>0: vec=np.zeros(len(inds),dtype='int') vec[1:nextra+1]=1 vec=np.cumsum(vec) inds=inds+vec return inds def split_partitioned_vec(start,stop,breaks=[],seg_len=100): if len(breaks)==0: return partition_interval(start,stop,seg_len) if breaks[0]==start: breaks=breaks[1:] if len(breaks)==0: return partition_interval(start,stop,seg_len) if breaks[-1]==stop: breaks=breaks[:-1] if len(breaks)==0: return partition_interval(start,stop,seg_len) breaks=np.hstack([start,breaks,stop]) nseg=len(breaks)-1 segs=[None]*(nseg) for i in range(nseg): inds=partition_interval(breaks[i],breaks[i+1],seg_len) if i<(nseg-1): inds=inds[:-1] segs[i]=inds segs=np.hstack(segs) return segs #breaks,stop,start=0,seg_len=100) class tsStripes(tsGeneric): def __init__(self,tod,seg_len=500,do_slope=False,tthresh=10): dims=tod.get_data_dims() tvec=tod.get_tvec() dt=np.median(np.diff(tvec)) splits=np.where(np.abs(np.diff(tvec))>tthresh*dt)[0] dims=tod.get_data_dims() inds=split_partitioned_vec(0,dims[1],splits,seg_len) self.inds=inds self.splits=splits self.nseg=len(self.inds)-1 self.params=np.zeros([dims[0],self.nseg]) def tod2map(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: print('need dat in tod2map destriper') return minkasi_nb.tod2map_destriped(dat,self.params,self.inds,do_add) def map2tod(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: print('need dat in map2tod destriper') return minkasi_nb.map2tod_destriped(dat,self.params,self.inds,do_add) def copy(self): return copy.deepcopy(self) def set_prior_from_corr(self,corrvec,thresh=0.5): assert(corrvec.shape[0]==self.params.shape[0]) n=self.params.shape[1] corrvec=corrvec[:,:n].copy() corrft=mkfftw.fft_r2r(corrvec) if thresh>0: for i in range(corrft.shape[0]): tt=thresh*np.median(corrft[i,:]) ind=corrft[i,:]<tt corrft[i,ind]=tt self.params=1.0/corrft/(2*(n-1)) def apply_prior(self,x,Ax): xft=mkfftw.fft_r2r(x.params) Ax.params=Ax.params+mkfftw.fft_r2r(xft*self.params) class tsBinnedAz(tsGeneric): def __init__(self,tod,lims=[0,2*np.pi],nbin=360): #print('nbin is',nbin) ndet=tod.get_ndet() self.params=np.zeros([ndet,nbin]) self.lims=[lims[0],lims[1]] self.nbin=nbin def map2tod(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: dat=tod.get_data() minkasi_nb.map2tod_binned_det(dat,self.params,tod.info['az'],self.lims,self.nbin,do_add) def tod2map(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: dat=tod.get_data() minkasi_nb.tod2map_binned_det(dat,self.params,tod.info['az'],self.lims,self.nbin,do_add) class tsBinnedAzShared(tsGeneric): #"""class to have az shared amongst TODs (say, if you think the ground is constant for a while)""" def __init__(self,ndet=2,lims=[0,2*np.pi],nbin=360): self.params=np.zeros([ndet,nbin]) self.lims=[lims[0],lims[1]] self.nbin=nbin def map2tod(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: dat=tod.get_data() minkasi_nb.map2tod_binned_det(dat,self.params,tod.info['az'],self.lims,self.nbin,do_add) def tod2map(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: dat=tod.get_data() print('nbin is',self.nbin) print(self.params.dtype) minkasi_nb.tod2map_binned_det(dat,self.params,tod.info['az'],self.lims,self.nbin,do_add) class tsDetAz(tsGeneric): def __init__(self,tod,npoly=4): if isinstance(tod,tsDetAz): #we're starting a new instance from an old one, e.g. from copy self.fname=tod.fname self.az=tod.az self.azmin=tod.azmin self.azmax=tod.azmax self.npoly=tod.npoly self.ndet=tod.ndet else: self.fname=tod.info['fname'] self.az=tod.info['AZ'] self.azmin=np.min(self.az) self.azmax=np.max(self.az) self.npoly=npoly #self.ndet=tod.info['dat_calib'].shape[0] self.ndet=tod.get_ndet() #self.params=np.zeros([self.ndet,self.npoly]) self.params=np.zeros([self.ndet,self.npoly-1]) def _get_polys(self): polys=np.zeros([self.npoly,len(self.az)]) polys[0,:]=1.0 az_scale= (self.az-self.azmin)/(self.azmax-self.azmin)*2.0-1.0 if self.npoly>1: polys[1,:]=az_scale for i in range(2,self.npoly): polys[i,:]=2*az_scale*polys[i-1,:]-polys[i-2,:] polys=polys[1:,:].copy() return polys def map2tod(self,tod,dat=None,do_add=True,do_omp=False): if dat is None: #dat=tod.info['dat_calib'] dat=tod.get_data() if do_add: dat[:]=dat[:]+np.dot(self.params,self._get_polys()) else: dat[:]=np.dot(self.params,self._get_polys()) def tod2map(self,tod,dat=None, do_add=True,do_omp=False): if dat is None: #dat=tod.info['dat_calib'] dat=tod.get_data() if do_add: #print("params shape is ",self.params.shape) #self.params[:]=self.params[:]+np.dot(self._get_polys(),dat) self.params[:]=self.params[:]+np.dot(dat,self._get_polys().T) else: #self.params[:]=np.dot(self._get_polys(),dat) self.params[:]=np.dot(dat,self._get_polys().T) class tsAirmass: def __init__(self,tod=None,order=3): if tod is None: self.sz=np.asarray([0,0],dtype='int') self.params=np.zeros(1) self.fname='' self.order=0 self.airmass=None else: #self.sz=tod.info['dat_calib'].shape self.sz=tod.get_data_dims() self.fname=tod.info['fname'] self.order=order self.params=np.zeros(order) if not('apix' in tod.info.keys()): #tod.info['apix']=scaled_airmass_from_el(tod.info['elev']) self.airmass=scaled_airmass_from_el(tod.info['elev']) else: self.airmass=tod.info['apix'] def copy(self,copyMat=False): cp=tsAirmass() cp.sz=self.sz cp.params=self.params.copy() cp.fname=self.fname cp.order=self.order if copyMat: cp.airmass=self.airmass.copy() else: cp.airmass=self.airmass #since this shouldn't change, use a pointer to not blow up RAM return cp def clear(self): self.params[:]=0.0 def dot(self,ts): return np.sum(self.params*ts.params) def axpy(self,ts,a): self.params=self.params+a*ts.params def _get_current_legmat(self): x=np.linspace(-1,1,self.sz[1]) m1=np.polynomial.legendre.legvander(x,self.order) return m1 def _get_current_model(self): x=np.linspace(-1,1,self.sz[1]) m1=self._get_current_legmat() poly=np.dot(m1,self.params) mat=np.repeat([poly],self.sz[0],axis=0) mat=mat*self.airmass return mat def tod2map(self,tod,dat,do_add=True,do_omp=False): tmp=np.zeros(self.order) for i in range(self.order): #tmp[i]=np.sum(tod.info['apix']**(i+1)*dat) tmp[i]=np.sum(self.airmass**(i+1)*dat) if do_add: self.params[:]=self.params[:]+tmp else: self.params[:]=tmp #poly=self._get_current_legmat() #vec=np.sum(dat*self.airmass,axis=0) #atd=np.dot(vec,poly) #if do_add: # self.params[:]=self.params[:]+atd #else: # self.params[:]=atd def map2tod(self,tod,dat,do_add=True,do_omp=False): mat=0.0 for i in range(self.order): #mat=mat+self.params[i]*tod.info['apix']**(i+1) mat=mat+self.params[i]*self.airmass**(i+1) #mat=self._get_current_model() if do_add: dat[:]=dat[:]+mat else: dat[:]=mat def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params return tt def write(self,fname=None): pass class tsCommon: def __init__(self,tod=None,*args,**kwargs): if tod is None: self.sz=np.asarray([0,0],dtype='int') self.params=np.zeros(1) self.fname='' else: #self.sz=tod.info['dat_calib'].shape self.sz=tod.get_data_dims() self.params=np.zeros(self.sz[1]) self.fname=tod.info['fname'] def copy(self): cp=tsCommon() try: cp.sz=self.sz.copy() except:#if the size doesn't have a copy function, then it's probably a number you can just assign cp.sz=self.sz cp.fname=self.fname cp.params=self.params.copy() return cp def clear(self): self.params[:]=0.0 def dot(self,common=None): if common is None: return np.dot(self.params,self.params) else: return np.dot(self.params,common.params) def axpy(self,common,a): self.params=self.params+a*common.params def tod2map(self,tod,dat,do_add=True,do_omp=False): #assert(self.fname==tod.info['fname'] nm=tod.info['fname'] if do_add==False: self.clear() self.params[:]=self.params[:]+np.sum(dat,axis=0) def map2tod(self,tod,dat,do_add=True,do_omp=True): nm=tod.info['fname'] dat[:]=dat[:]+np.repeat([self.params],dat.shape[0],axis=0) def write(self,fname=None): pass def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params return tt class detOffset: def __init__(self,tod=None): if tod is None: self.sz=1 self.params=np.zeros(1) self.fname='' else: #self.sz=tod.info['dat_calib'].shape[0] self.sz=tod.get_ndet() self.params=np.zeros(self.sz) self.fname=tod.info['fname'] def copy(self): cp=detOffset() cp.sz=self.sz cp.params=self.params.copy() cp.fname=self.fname return cp def clear(self): self.params[:]=0 def dot(self,other=None): if other is None: return np.dot(self.params,self.params) else: return np.dot(self.params,other.params) def axpy(self,common,a): self.params=self.params+a*common.params def tod2map(self,tod,dat,do_add=True,do_omp=False): if do_add==False: self.clear() self.params[:]=self.params[:]+np.sum(dat,axis=1) def map2tod(self,tod,dat,do_add=True,do_omp=False): if do_add==False: dat[:]=0 dat[:]=dat[:]+np.repeat([self.params],dat.shape[1],axis=0).transpose() def write(self,fname=None): pass def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params return tt class tsCalib: def __init__(self,tod=None,model=None): if tod is None: self.sz=1 self.params=np.zeros(1) self.fname='' self.pred=None else: #self.sz=tod.info['dat_calib'].shape[0] self.sz=tod.get_ndet() self.params=np.zeros(self.sz) self.pred=model[tod.info['fname']].copy() self.fname=tod.info['fname'] def copy(self): cp=tsCalib() cp.sz=self.sz cp.params=self.params.copy() cp.fname=self.fname cp.pred=self.pred return cp def clear(self): self.params[:]=0 def dot(self,other=None): if other is None: return np.dot(self.params,self.params) else: return np.dot(self.params,other.params) def axpy(self,common,a): self.params=self.params+a*common.params def tod2map(self,tod,dat,do_add=True,do_omp=False): if do_add==False: self.clear() if self.pred.ndim==1: self.params[:]=self.params[:]+np.dot(dat,self.pred) else: self.params[:]=self.params[:]+np.sum(dat*self.pred,axis=1) def map2tod(self,tod,dat,do_add=True,do_omp=False): if do_add==False: dat[:]=0 if self.pred.ndim==1: dat[:]=dat[:]+np.outer(self.params,self.pred) else: dat[:]=dat[:]+(self.pred.transpose()*self.params).transpose() def write(self,fname=None): pass def __mul__(self,to_mul): tt=self.copy() tt.params=self.params*to_mul.params return tt class tsModel: def __init__(self,todvec=None,modelclass=None,*args,**kwargs): self.data={} if todvec is None: return for tod in todvec.tods: nm=tod.info['fname'] self.data[nm]=modelclass(tod,*args,**kwargs) def copy(self): new_tsModel=tsModel() for nm in self.data.keys(): new_tsModel.data[nm]=self.data[nm].copy() return new_tsModel def tod2map(self,tod,dat,do_add=True,do_omp=False): nm=tod.info['fname'] if do_add==False: self.clear() self.data[nm].tod2map(tod,dat,do_add,do_omp) def map2tod(self,tod,dat,do_add=True,do_omp=True): nm=tod.info['fname'] if do_add==False: dat[:]=0.0 self.data[nm].map2tod(tod,dat,do_add,do_omp) def apply_prior(self,x,Ax): for nm in self.data.keys(): self.data[nm].apply_prior(x.data[nm],Ax.data[nm]) def dot(self,tsmodels=None): tot=0.0 for nm in self.data.keys(): if tsmodels is None: tot=tot+self.data[nm].dot(self.data[nm]) else: #if tsmodels.data.has_key(nm): if nm in tsmodels.data: tot=tot+self.data[nm].dot(tsmodels.data[nm]) else: print('error in tsModel.dot - missing key ',nm) assert(1==0) #pretty sure we want to crash if missing names if have_mpi: tot=comm.allreduce(tot) return tot def clear(self): for nm in self.data.keys(): self.data[nm].clear() def axpy(self,tsmodel,a): for nm in self.data.keys(): self.data[nm].axpy(tsmodel.data[nm],a) def __mul__(self,tsmodel): #this is used in preconditioning - need to fix if ts-based preconditioning is desired tt=self.copy() for nm in self.data.keys(): tt.data[nm]=self.data[nm]*tsmodel.data[nm] return tt #for nm in tt.data.keys(): # tt.params[nm]=tt.params[nm]*tsmodel.params[nm] def mpi_reduce(self): pass def get_caches(self): for nm in self.data.keys(): try: self.data[nm].get_caches() except: pass def clear_caches(self): for nm in self.data.keys(): try: self.data[nm].clear_caches() except: pass def mpi_reduce(self): pass class tsMultiModel(tsModel): """A class to hold timestream models that are shared between groups of TODs.""" def __init__(self,todvec=None,todtags=None,modelclass=None,tag='ts_multi_model',*args,**kwargs): self.data={} self.tag=tag if not(todtags is None): alltags=comm.allgather(todtags) alltags=np.hstack(alltags) alltags=np.unique(alltags) if not(modelclass is None): for mytag in alltags: self.data[mytag]=modelclass(*args,**kwargs) if not(todvec is None): for i,tod in enumerate(todvec.tods): tod.info[tag]=todtags[i] def copy(self): return copy.deepcopy(self) def tod2map(self,tod,dat,do_add=True,do_omp=False): self.data[tod.info[self.tag]].tod2map(tod,dat,do_add,do_omp) def map2tod(self,tod,dat,do_add=True,do_omp=False): self.data[tod.info[self.tag]].map2tod(tod,dat,do_add,do_omp) def dot(self,tsmodels=None): tot=0.0 for nm in self.data.keys(): if tsmodels is None: tot=tot+self.data[nm].dot(self.data[nm]) else: if nm in tsmodels.data: tot=tot+self.data[nm].dot(tsmodels.data[nm]) else: print('error in tsMultiModel.dot - missing key ',nm) assert(1==0) return tot class Mapset: def __init__(self): self.nmap=0 self.maps=[] def add_map(self,map): self.maps.append(map.copy()) self.nmap=self.nmap+1 def clear(self): for i in range(self.nmap): self.maps[i].clear() def copy(self): new_mapset=Mapset() for i in range(self.nmap): new_mapset.add_map(self.maps[i].copy()) return new_mapset def dot(self,mapset): tot=0.0 for i in range(self.nmap): tot=tot+self.maps[i].dot(mapset.maps[i]) return tot def axpy(self,mapset,a): for i in range(self.nmap): self.maps[i].axpy(mapset.maps[i],a) def __add__(self,mapset): mm=self.copy() mm.axpy(mapset,1.0) return mm def __sub__(self,mapset): mm=self.copy() mm.axpy(mapset,-1.0) return mm def __mul__(self,mapset): #mm=self.copy() mm=mapset.copy() #return mm for i in range(self.nmap): #print('callin mul on map ',i) mm.maps[i]=self.maps[i]*mapset.maps[i] return mm def get_caches(self): for i in range(self.nmap): self.maps[i].get_caches() def clear_caches(self): for i in range(self.nmap): self.maps[i].clear_caches() def apply_prior(self,x,Ax): for i in range(self.nmap): if not(self.maps[i] is None): try: if self.maps[i].isglobal_prior: #print('applying global prior') self.maps[i].apply_prior(x,Ax) else: self.maps[i].apply_prior(x.maps[i],Ax.maps[i]) except: #print('going through exception') self.maps[i].apply_prior(x.maps[i],Ax.maps[i]) def mpi_reduce(self): if have_mpi: for map in self.maps: map.mpi_reduce() class SkyMap: def __init__(self,lims,pixsize=0,proj='CAR',pad=2,primes=None,cosdec=None,nx=None,ny=None,mywcs=None,tag='ipix',purge_pixellization=False,ref_equ=False): if mywcs is None: assert(pixsize!=0) #we had better have a pixel size if we don't have an incoming WCS that contains it self.wcs=get_wcs(lims,pixsize,proj,cosdec,ref_equ) else: self.wcs=mywcs pixsize_use=mywcs.wcs.cdelt[1]*np.pi/180 #print('pixel size from wcs and requested are ',pixsize_use,pixsize,100*(pixsize_use-pixsize)/pixsize) pixsize=pixsize_use corners=np.zeros([4,2]) corners[0,:]=[lims[0],lims[2]] corners[1,:]=[lims[0],lims[3]] corners[2,:]=[lims[1],lims[2]] corners[3,:]=[lims[1],lims[3]] pix_corners=self.wcs.wcs_world2pix(corners*180/np.pi,1) pix_corners=np.round(pix_corners) #print pix_corners #print type(pix_corners) #if pix_corners.min()<0.5: if pix_corners.min()<-0.5: print('corners seem to have gone negative in SkyMap projection. not good, you may want to check this.') if True: #try a patch to fix the wcs xxx if nx is None: nx=(pix_corners[:,0].max()+pad) if ny is None: ny=(pix_corners[:,1].max()+pad) else: nx=(pix_corners[:,0].max()+pad) ny=(pix_corners[:,1].max()+pad) #print nx,ny nx=int(nx) ny=int(ny) if not(primes is None): lens=find_good_fft_lens(2*(nx+ny),primes) #print 'nx and ny initially are ',nx,ny nx=lens[lens>=nx].min() ny=lens[lens>=ny].min() #print 'small prime nx and ny are now ',nx,ny self.primes=primes[:] else: self.primes=None self.nx=nx self.ny=ny self.lims=lims self.pixsize=pixsize self.map=np.zeros([nx,ny]) self.proj=proj self.pad=pad self.tag=tag self.purge_pixellization=purge_pixellization self.caches=None self.cosdec=cosdec self.tod2map_method=None def get_caches(self): npix=self.nx*self.ny nthread=get_nthread() self.caches=np.zeros([nthread,npix]) def clear_caches(self): self.map[:]=np.reshape(np.sum(self.caches,axis=0),self.map.shape) self.caches=None def set_tod2map(self,method=None,todvec=None): """Select which method of tod2map to use. options include simple (1 proc), omp (everyone makes a map copy), everyone (everyone loops through all the data but assigns only to their own piece), atomic (no map copy, accumulate via atomic adds), and cached (every thread has a sticky copy of the map).""" if method is None: if nproc==1: self.tod2map_method=self.tod2map_simple else: self.tod2map_method=self.tod2map_omp return if method=='omp': self.tod2map_method=self.tod2map_omp return if method=='simple': self.tod2map_method=self.tod2map_simple if method=='everyone': if todvec is None: print('need tods when setting to everyone so we can know which pieces belong to which threads') for tod in todvec.tods: ipix=self.get_pix(tod,False) ipix=ipix.copy() ipix=np.ravel(ipix) ipix.sort() inds=len(ipix)*np.arange(nproc+1)//nproc inds=np.asarray(inds,dtype='int32') tod.save_pixellization(self.tag+'_edges',inds) self.tod2map_method=self.tod2map_everyone if method=='cached': self.get_caches() self.tod2map_method=self.tod2map_cached if method=='atomic': self.tod2map_method=self.todmap_atomic def tod2map_atomic(self,tod,dat): ipix=self.get_pix(tod) tod2map_omp(self.map,dat,ipix,True) def todmap_omp(self,tod,dat): ipix=self.get_pix(tod) tod2map_omp(self.map,dat,ipix,False) def tod2map_simple(self,tod,dat): ipix=self.get_pix(tod) tod2map_simple(self.map,dat,ipix) #def tod2map_cached(self.map,dat,ipix): # ipix=self.get_pix(tod) # tod2map_cached(map,dat,ipix) def copy(self): if False: newmap=SkyMap(self.lims,self.pixsize,self.proj,self.pad,self.primes,cosdec=self.cosdec,nx=self.nx,ny=self.ny,mywcs=self.wcs,tag=self.tag) newmap.map[:]=self.map[:] return newmap else: return copy.deepcopy(self) def clear(self): self.map[:]=0 def axpy(self,map,a): self.map[:]=self.map[:]+a*map.map[:] def assign(self,arr): assert(arr.shape[0]==self.nx) assert(arr.shape[1]==self.ny) #self.map[:,:]=arr self.map[:]=arr def pix_from_radec(self,ra,dec): ndet=ra.shape[0] nsamp=ra.shape[1] nn=ndet*nsamp coords=np.zeros([nn,2]) #coords[:,0]=np.reshape(tod.info['dx']*180/np.pi,nn) #coords[:,1]=np.reshape(tod.info['dy']*180/np.pi,nn) coords[:,0]=np.reshape(ra*180/np.pi,nn) coords[:,1]=np.reshape(dec*180/np.pi,nn) #print coords.shape pix=self.wcs.wcs_world2pix(coords,1) #print pix.shape xpix=np.reshape(pix[:,0],[ndet,nsamp])-1 #-1 is to go between unit offset in FITS and zero offset in python ypix=np.reshape(pix[:,1],[ndet,nsamp])-1 xpix=np.round(xpix) ypix=np.round(ypix) ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') return ipix def get_pix(self,tod,savepix=True): if not(self.tag is None): ipix=tod.get_saved_pix(self.tag) if not(ipix is None): return ipix ra,dec=tod.get_radec() #ndet=tod.info['dx'].shape[0] #nsamp=tod.info['dx'].shape[1] if False: ndet=ra.shape[0] nsamp=ra.shape[1] nn=ndet*nsamp coords=np.zeros([nn,2]) #coords[:,0]=np.reshape(tod.info['dx']*180/np.pi,nn) #coords[:,1]=np.reshape(tod.info['dy']*180/np.pi,nn) coords[:,0]=np.reshape(ra*180/np.pi,nn) coords[:,1]=np.reshape(dec*180/np.pi,nn) #print coords.shape pix=self.wcs.wcs_world2pix(coords,1) #print pix.shape xpix=np.reshape(pix[:,0],[ndet,nsamp])-1 #-1 is to go between unit offset in FITS and zero offset in python ypix=np.reshape(pix[:,1],[ndet,nsamp])-1 xpix=np.round(xpix) ypix=np.round(ypix) ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') else: ipix=self.pix_from_radec(ra,dec) if savepix: if not(self.tag is None): tod.save_pixellization(self.tag,ipix) return ipix def map2tod(self,tod,dat,do_add=True,do_omp=True): ipix=self.get_pix(tod) #map2tod(dat,self.map,tod.info['ipix'],do_add,do_omp) map2tod(dat,self.map,ipix,do_add,do_omp) def tod2map(self,tod,dat=None,do_add=True,do_omp=True): if dat is None: dat=tod.get_data() if do_add==False: self.clear() ipix=self.get_pix(tod) if not(self.caches is None): #tod2map_cached(self.caches,dat,tod.info['ipix']) tod2map_cached(self.caches,dat,ipix) else: if do_omp: #tod2map_omp(self.map,dat,tod.info['ipix']) tod2map_omp(self.map,dat,ipix) else: #tod2map_simple(self.map,dat,tod.info['ipix']) tod2map_simple(self.map,dat,ipix) if self.purge_pixellization: tod.clear_saved_pix(self.tag) def tod2map_old(self,tod,dat=None,do_add=True,do_omp=True): if dat is None: dat=tod.get_data() if do_add==False: self.clear() ipix=self.get_pix(tod) if not(self.caches is None): #tod2map_cached(self.caches,dat,tod.info['ipix']) tod2map_cached(self.caches,dat,ipix) else: if do_omp: #tod2map_omp(self.map,dat,tod.info['ipix']) tod2map_omp(self.map,dat,ipix) else: #tod2map_simple(self.map,dat,tod.info['ipix']) tod2map_simple(self.map,dat,ipix) if self.purge_pixellization: tod.clear_saved_pix(self.tag) def r_th_maps(self): xvec=np.arange(self.nx) xvec=xvec-xvec.mean() yvec=np.arange(self.ny) yvec=yvec-yvec.mean() ymat,xmat=np.meshgrid(yvec,xvec) rmat=np.sqrt(xmat**2+ymat**2) th=np.arctan2(xmat,ymat) return rmat,th def dot(self,map): tot=np.sum(self.map*map.map) return tot def plot(self,plot_info=None): vmin=self.map.min() vmax=self.map.max() clf=True pause=True pause_len=0.001 if not(plot_info is None): if 'vmin' in plot_info.keys(): vmin=plot_info['vmin'] if 'vmax' in plot_info.keys(): vmax=plot_info['vmax'] if 'clf' in plot_info.keys(): clf=plot_info['clf'] if 'pause' in plot_info.keys(): pause=plot_info['pause'] if pause_len in plot_info.keys(): pause_len=plot_info['pause_len'] from matplotlib import pyplot as plt if clf: plt.clf() plt.imshow(self.map,vmin=vmin,vmax=vmax) if pause: plt.pause(pause_len) def write(self,fname='map.fits'): header=self.wcs.to_header() if True: #try a patch to fix the wcs xxx tmp=self.map.transpose().copy() hdu=fits.PrimaryHDU(tmp,header=header) else: hdu=fits.PrimaryHDU(self.map,header=header) try: hdu.writeto(fname,overwrite=True) except: hdu.writeto(fname,clobber=True) def __mul__(self,map): new_map=map.copy() new_map.map[:]=self.map[:]*map.map[:] return new_map def mpi_reduce(self,chunksize=1e5): #chunksize is added since at least on my laptop mpi4py barfs if it #tries to reduce an nside=512 healpix map, so need to break it into pieces. if have_mpi: #print("reducing map") if chunksize>0: nchunk=(1.0*self.nx*self.ny)/chunksize nchunk=int(np.ceil(nchunk)) else: nchunk=1 #print('nchunk is ',nchunk) if nchunk==1: self.map=comm.allreduce(self.map) else: inds=np.asarray(np.linspace(0,self.nx*self.ny,nchunk+1),dtype='int') if len(self.map.shape)>1: tmp=np.zeros(self.map.size) tmp[:]=np.reshape(self.map,len(tmp)) else: tmp=self.map for i in range(len(inds)-1): tmp[inds[i]:inds[i+1]]=comm.allreduce(tmp[inds[i]:inds[i+1]]) #self.map[inds[i]:inds[i+1]]=comm.allreduce(self.map[inds[i]:inds[i+1]]) #tmp=np.zeros(inds[i+1]-inds[i]) #tmp[:]=self.map[inds[i]:inds[i+1]] #tmp=comm.allreduce(tmp) #self.map[inds[i]:inds[i+1]]=tmp if len(self.map.shape)>1: self.map[:]=np.reshape(tmp,self.map.shape) #print("reduced") def invert(self): mask=np.abs(self.map)>0 self.map[mask]=1.0/self.map[mask] class MapNoiseWhite: def __init__(self,ivar_map,isinv=True,nfac=1.0): self.ivar=read_fits_map(ivar_map) if not(isinv): mask=self.ivar>0 self.ivar[mask]=1.0/self.ivar[mask] self.ivar=self.ivar*nfac def apply_noise(self,map): return map*self.ivar class SkyMapCoarse(SkyMap): def __init__(self,map): self.nx=map.shape[0] try: self.ny=map.shape[1] except: self.ny=1 self.map=map.copy() def get_caches(self): return def clear_caches(self): return def copy(self): cp=copy.copy(self) cp.map=self.map.copy() return cp def get_pix(self): return def map2tod(self,*args,**kwargs): return def tod2map(self,*args,**kwargs): return class SkyMapTwoRes: """A pair of maps to serve as a prior for multi-experiment mapping. This would e.g. be the ACT map that e.g. Mustang should agree with on large scales.""" def __init__(self,map_lowres,lims,osamp=1,smooth_fac=0.0): small_wcs,lims_use,map_corner=get_aligned_map_subregion_car(lims,map_lowres,osamp=osamp) self.small_lims=lims_use self.small_wcs=small_wcs self.map=read_fits_map(map_lowres) self.osamp=osamp self.map_corner=map_corner self.beamft=None self.mask=None self.map_deconvolved=None self.noise=None self.fine_prior=None self.nx_coarse=None self.ny_coarse=None self.grid_facs=None self.isglobal_prior=True self.smooth_fac=smooth_fac def copy(self): return copy.copy(self) def get_map_deconvolved(self,map_deconvolved): self.map_deconvolved=read_fits_map(map_deconvolved) def set_beam_gauss(self,fwhm_pix): tmp=0*self.map xvec=get_ft_vec(tmp.shape[0]) yvec=get_ft_vec(tmp.shape[1]) xx,yy=np.meshgrid(yvec,xvec) rsqr=xx**2+yy**2 sig_pix=fwhm_pix/np.sqrt(8*np.log(2)) beam=np.exp(-0.5*rsqr/(sig_pix**2)) beam=beam/np.sum(beam) self.beamft=np.fft.rfft2(beam) def set_beam_1d(self,prof,pixsize): tmp=0*self.map xvec=get_ft_vec(tmp.shape[0]) yvec=get_ft_vec(tmp.shape[1]) xx,yy=np.meshgrid(yvec,xvec) rsqr=xx**2+yy**2 rr=np.sqrt(rsqr)*pixsize beam=interp(rr,prof[:,0],prof[:,1]) beam=beam/np.sum(beam) self.beamft=np.fft.rfft2(beam) def set_noise_white(self,ivar_map,isinv=True,nfac=1.0): self.noise=MapNoiseWhite(ivar_map,isinv,nfac) def maps2fine(self,fine,coarse): out=fine.copy() for i in range(self.nx_coarse): for j in range(self.ny_coarse): out[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]=coarse[i+self.map_corner[0],j+self.map_corner[1]] out[self.mask]=fine[self.mask] return out def maps2coarse(self,fine,coarse): out=coarse.copy() for i in range(self.nx_coarse): for j in range(self.ny_coarse): out[i+self.map_corner[0],j+self.map_corner[1]]=(1-self.grid_facs[i,j])*coarse[i+self.map_corner[0],j+self.map_corner[1]]+np.sum(fine[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)])/self.osamp**2 return out def coarse2maps(self,inmap): coarse=1.0*inmap fine=np.zeros([self.nx_coarse*self.osamp,self.ny_coarse*self.osamp]) for i in range(self.nx_coarse): for j in range(self.ny_coarse): coarse[i+self.map_corner[0],j+self.map_corner[1]]=(1-self.grid_facs[i,j])*inmap[i+self.map_corner[0],j+self.map_corner[1]] fine[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]=inmap[i+self.map_corner[0],j+self.map_corner[1]]/self.osamp**2 fine=fine*self.mask return coarse,fine def set_mask(self,hits,thresh=0): self.mask=hits>thresh self.fine_prior=0*hits self.nx_coarse=int(np.round(hits.shape[0]/self.osamp)) self.ny_coarse=int(np.round(hits.shape[1]/self.osamp)) self.grid_facs=np.zeros([self.nx_coarse,self.ny_coarse]) for i in range(self.nx_coarse): for j in range(self.ny_coarse): self.grid_facs[i,j]=np.mean(self.mask[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]) self.fine_prior[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]=self.map_deconvolved[self.map_corner[0]+i,self.map_corner[1]+j] def apply_Qinv(self,map): tmp=self.fine_prior.copy() tmp[self.mask]=map[self.mask] tmp2=0*self.map_deconvolved.copy() for i in range(self.nx_coarse): for j in range(self.nx_coarse): tmp2[self.map_corner[0]+i,self.map_corner[1]+j]=np.mean(tmp[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]) tmp2_conv=np.fft.irfft2(np.fft.rfft2(tmp2)*self.beamft) tmp2_conv_filt=self.noise.apply_noise(tmp2_conv) tmp2_reconv=np.fft.irfft2(np.fft.rfft2(tmp2_conv_filt)*self.beamft) #tmp2_reconv=np.fft.irfft2(np.fft.rfft2(tmp2_conv)*self.beamft) #tmp2_reconv=tmp2.copy() fac=1.0/self.osamp**2 for i in range(self.nx_coarse): for j in range(self.ny_coarse): tmp[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]=fac*tmp2_reconv[i+self.map_corner[0],j+self.map_corner[1]] ans=0.0*tmp ans[self.mask]=tmp[self.mask] return ans def apply_H(self,coarse,fine): mm=self.maps2coarse(coarse,fine) mm=self.beam_convolve(mm) return mm def apply_HT(self,mm): mm=self.beam_convolve(mm) coarse,fine=self.coarse2maps(mm) return coarse,fine def get_rhs(self,mapset): #if map is None: # map=self.map #map_filt=self.noise.apply_noise(map) #map_filt_conv=np.fft.irfft2(np.fft.rfft2(map_filt)*self.beamft) #tmp=0.0*self.mask #fac=1.0/self.osamp**2 #for i in range(self.nx_coarse): # for j in range(self.ny_coarse): # tmp[(i*self.osamp):((i+1)*self.osamp),(j*self.osamp):((j+1)*self.osamp)]=fac*map_filt_conv[i+self.map_corner[0],j+self.map_corner[1]] #ans=0*tmp #ans[self.mask]=tmp[self.mask] #return ans coarse_ind=None fine_ind=None for i in range(mapset.nmap): if isinstance(mapset.maps[i],SkyMapCoarse): coarse_ind=i else: if isinstance(mapset.maps[i],SkyMap): fine_ind=i if (coarse_ind is None)|(fine_ind is None): print("Errror in twolevel prior: either fine or coarse skymap not found.") return mm=self.noise.apply_noise(self.map) if True: coarse,fine=self.apply_HT(mm) mapset.maps[coarse_ind].map[:]=mapset.maps[coarse_ind].map[:]+coarse mapset.maps[fine_ind].map[:]=mapset.maps[fine_ind].map[:]+fine else: mm=self.beam_convolve(mm) coarse,fine=self.coarse2maps(mm) i1=self.map_corner[0] i2=i1+self.nx_coarse j1=self.map_corner[1] j2=j1+self.ny_coarse coarse[i1:i2,j1:j2]=coarse[i1:i2,j1:j2]*(1-self.grid_facs) mapset.maps[coarse_ind].map[:]=mapset.maps[coarse_ind].map[:]+coarse mapset.maps[fine_ind].map[self.mask]=mapset.maps[fine_ind].map[self.mask]+fine[self.mask]/self.osamp**2 def beam_convolve(self,map): mapft=np.fft.rfft2(map) mapft=mapft*self.beamft return np.fft.irfft2(mapft) def apply_prior(self,mapset,outmapset): coarse_ind=None fine_ind=None for i in range(mapset.nmap): if isinstance(mapset.maps[i],SkyMapCoarse): coarse_ind=i else: if isinstance(mapset.maps[i],SkyMap): fine_ind=i if (coarse_ind is None)|(fine_ind is None): print("Errror in twolevel prior: either fine or coarse skymap not found.") return if True: mm=self.apply_H(mapset.maps[fine_ind].map,mapset.maps[coarse_ind].map) mm_filt=self.noise.apply_noise(mm) coarse,fine=self.apply_HT(mm_filt) else: summed=self.maps2coarse(mapset.maps[fine_ind].map,mapset.maps[coarse_ind].map) summed=self.beam_convolve(summed) summed=self.noise.apply_noise(summed) summed=self.beam_convolve(summed) coarse,fine=self.coarse2maps(summed) outmapset.maps[fine_ind].map[self.mask]=outmapset.maps[fine_ind].map[self.mask]+fine[self.mask] outmapset.maps[coarse_ind].map[:]=outmapset.maps[coarse_ind].map[:]+coarse if self.smooth_fac>0: summed=self.maps2coarse(mapset.maps[fine_ind].map,mapset.maps[coarse_ind].map) summed_smooth=self.beam_convolve(summed) delt=summed-summed_smooth delt_filt=self.noise.apply_noise(delt)*self.smooth_fac delt_filt=delt_filt-self.beam_convolve(delt_filt) coarse,fine=self.coarse2maps(delt_filt) outmapset.maps[fine_ind].map[self.mask]=outmapset.maps[fine_ind].map[self.mask]+fine[self.mask] outmapset.maps[coarse_ind].map[:]=outmapset.maps[coarse_ind].map[:]+coarse def __bust_apply_prior(self,map,outmap): outmap.map[:]=outmap.map[:]+self.apply_Qinv(map.map) def poltag2pols(poltag): if poltag=='I': return ['I'] if poltag=='IQU': return ['I','Q','U'] if poltag=='QU': return ['Q','U'] if poltag=='IQU_PRECON': return ['I','Q','U','QQ','QU','UU'] if poltag=='QU_PRECON': return ['QQ','UU','QU'] return None class PolMap: def __init__(self,lims,pixsize,poltag='I',proj='CAR',pad=2,primes=None,cosdec=None,nx=None,ny=None,mywcs=None,tag='ipix',purge_pixellization=False,ref_equ=False): pols=poltag2pols(poltag) if pols is None: print('Unrecognized polarization state ' + poltag + ' in PolMap.__init__') return npol=len(pols) if mywcs is None: self.wcs=get_wcs(lims,pixsize,proj,cosdec,ref_equ) else: self.wcs=mywcs corners=np.zeros([4,2]) corners[0,:]=[lims[0],lims[2]] corners[1,:]=[lims[0],lims[3]] corners[2,:]=[lims[1],lims[2]] corners[3,:]=[lims[1],lims[3]] pix_corners=self.wcs.wcs_world2pix(corners*180/np.pi,1) pix_corners=np.round(pix_corners) #print pix_corners #print type(pix_corners) #if pix_corners.min()<0.5: if pix_corners.min()<-0.5: print('corners seem to have gone negative in SkyMap projection. not good, you may want to check this.') if True: #try a patch to fix the wcs xxx if nx is None: nx=(pix_corners[:,0].max()+pad) if ny is None: ny=(pix_corners[:,1].max()+pad) else: nx=(pix_corners[:,0].max()+pad) ny=(pix_corners[:,1].max()+pad) #print nx,ny nx=int(nx) ny=int(ny) if not(primes is None): lens=find_good_fft_lens(2*(nx+ny),primes) #print 'nx and ny initially are ',nx,ny nx=lens[lens>=nx].min() ny=lens[lens>=ny].min() #print 'small prime nx and ny are now ',nx,ny self.primes=primes[:] else: self.primes=None self.nx=nx self.ny=ny self.npol=npol self.poltag=poltag self.pols=pols self.lims=lims self.tag=tag self.purge_pixellization=purge_pixellization self.pixsize=pixsize if npol>1: self.map=np.zeros([nx,ny,npol]) else: self.map=np.zeros([nx,ny]) self.proj=proj self.pad=pad self.caches=None self.cosdec=cosdec def get_caches(self): npix=self.nx*self.ny*self.npol nthread=get_nthread() self.caches=np.zeros([nthread,npix]) def clear_caches(self): self.map[:]=np.reshape(np.sum(self.caches,axis=0),self.map.shape) self.caches=None def copy(self): if False: newmap=PolMap(self.lims,self.pixsize,self.poltag,self.proj,self.pad,self.primes,cosdec=self.cosdec,nx=self.nx,ny=self.ny,mywcs=self.wcs) newmap.map[:]=self.map[:] return newmap else: return copy.deepcopy(self) def clear(self): self.map[:]=0 def axpy(self,map,a): self.map[:]=self.map[:]+a*map.map[:] def assign(self,arr): assert(arr.shape[0]==self.nx) assert(arr.shape[1]==self.ny) if self.npol>1: assert(arr.shape[2]==self.npol) #self.map[:,:]=arr self.map[:]=arr def set_polstate(self,poltag): pols=poltag2pols(poltag) if pols is None: print('Unrecognized polarization state ' + poltag + ' in PolMap.set_polstate.') return npol=len(pols) self.npol=npol self.poltag=poltag self.pols=pols if npol>1: self.map=np.zeros([self.nx,self.ny,npol]) else: self.map=np.zeros([self.nx,self.ny]) def invert(self,thresh=1e-6): #We can use np.linalg.pinv to reasonably efficiently invert a bunch of tiny matrices with an #eigenvalue cut. It's more efficient to do this in C, but it only has to be done once per run if self.npol>1: if self.poltag=='QU_PRECON': tmp=np.zeros([self.nx*self.ny,2,2]) tmp[:,0,0]=np.ravel(self.map[:,:,0]) tmp[:,1,1]=np.ravel(self.map[:,:,1]) tmp[:,0,1]=np.ravel(self.map[:,:,2]) tmp[:,1,0]=np.ravel(self.map[:,:,2]) tmp=np.linalg.pinv(tmp,thresh) self.map[:,:,0]=np.reshape(tmp[:,0,0],[self.map.shape[0],self.map.shape[1]]) self.map[:,:,1]=np.reshape(tmp[:,1,1],[self.map.shape[0],self.map.shape[1]]) self.map[:,:,2]=np.reshape(tmp[:,0,1],[self.map.shape[0],self.map.shape[1]]) if self.poltag=='IQU_PRECON': #the mapping may seem a bit abstruse here. The preconditioner matrix has entries # [I Q U ] # [Q QQ QU ] # [U QU UU ] #so the unpacking needs to match the ordering in the C file before we can use pinv n=self.nx*self.ny nx=self.nx ny=self.ny tmp=np.zeros([self.nx*self.ny,3,3]) tmp[:,0,0]=np.reshape(self.map[:,:,0],n) tmp[:,0,1]=np.reshape(self.map[:,:,1],n) tmp[:,1,0]=tmp[:,0,1] tmp[:,0,2]=np.reshape(self.map[:,:,2],n) tmp[:,2,0]=tmp[:,0,2] tmp[:,1,1]=np.reshape(self.map[:,:,3],n) tmp[:,1,2]=np.reshape(self.map[:,:,4],n) tmp[:,2,1]=tmp[:,1,2] tmp[:,2,2]=np.reshape(self.map[:,:,5],n) alldets=np.linalg.det(tmp) isbad=alldets<thresh*alldets.max() ispos=tmp[:,0,0]>0 inds=isbad&ispos vec=tmp[inds,0,0] print('determinant range is ' + repr(alldets.max())+ ' ' + repr(alldets.min())) tmp=np.linalg.pinv(tmp,thresh) if True: print('Warning! zeroing out bits like this is super janky. Be warned...') tmp[isbad,:,:]=0 inds=isbad&ispos tmp[inds,0,0]=1.0/vec alldets=np.linalg.det(tmp) print('determinant range is now ' + repr(alldets.max())+ ' ' + repr(alldets.min())) self.map[:,:,0]=np.reshape(tmp[:,0,0],[nx,ny]) self.map[:,:,1]=np.reshape(tmp[:,0,1],[nx,ny]) self.map[:,:,2]=np.reshape(tmp[:,0,2],[nx,ny]) self.map[:,:,3]=np.reshape(tmp[:,1,1],[nx,ny]) self.map[:,:,4]=np.reshape(tmp[:,1,2],[nx,ny]) self.map[:,:,5]=np.reshape(tmp[:,2,2],[nx,ny]) else: mask=self.map!=0 self.map[mask]=1.0/self.map[mask] def pix_from_radec(self,ra,dec): ndet=ra.shape[0] nsamp=ra.shape[1] nn=ndet*nsamp coords=np.zeros([nn,2]) coords[:,0]=np.reshape(ra*180/np.pi,nn) coords[:,1]=np.reshape(dec*180/np.pi,nn) #print coords.shape pix=self.wcs.wcs_world2pix(coords,1) #print pix.shape xpix=np.reshape(pix[:,0],[ndet,nsamp])-1 #-1 is to go between unit offset in FITS and zero offset in python ypix=np.reshape(pix[:,1],[ndet,nsamp])-1 xpix=np.round(xpix) ypix=np.round(ypix) ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') return ipix def get_pix(self,tod,savepix=True): if not(self.tag is None): ipix=tod.get_saved_pix(self.tag) if not(ipix is None): return ipix if False: ndet=tod.info['dx'].shape[0] nsamp=tod.info['dx'].shape[1] nn=ndet*nsamp coords=np.zeros([nn,2]) coords[:,0]=np.reshape(tod.info['dx']*180/np.pi,nn) coords[:,1]=np.reshape(tod.info['dy']*180/np.pi,nn) #print coords.shape pix=self.wcs.wcs_world2pix(coords,1) #print pix.shape xpix=np.reshape(pix[:,0],[ndet,nsamp])-1 #-1 is to go between unit offset in FITS and zero offset in python ypix=np.reshape(pix[:,1],[ndet,nsamp])-1 xpix=np.round(xpix) ypix=np.round(ypix) ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') else: ra,dec=tod.get_radec() ipix=self.pix_from_radec(ra,dec) if savepix: if not(self.tag is None): tod.save_pixellization(self.tag,ipix) return ipix def map2tod(self,tod,dat,do_add=True,do_omp=True): ipix=self.get_pix(tod) if self.npol>1: #polmap2tod(dat,self.map,self.poltag,tod.info['twogamma_saved'],tod.info['ipix'],do_add,do_omp) polmap2tod(dat,self.map,self.poltag,tod.info['twogamma_saved'],ipix,do_add,do_omp) else: #map2tod(dat,self.map,tod.info['ipix'],do_add,do_omp) map2tod(dat,self.map,ipix,do_add,do_omp) def tod2map(self,tod,dat,do_add=True,do_omp=True): if do_add==False: self.clear() ipix=self.get_pix(tod) #print('ipix start is ',ipix[0,0:500:100]) if self.npol>1: #tod2polmap(self.map,dat,self.poltag,tod.info['twogamma_saved'],tod.info['ipix']) tod2polmap(self.map,dat,self.poltag,tod.info['twogamma_saved'],ipix) if self.purge_pixellization: tod.clear_saved_pix(self.tag) return #print("working on nonpolarized bit") if not(self.caches is None): #tod2map_cached(self.caches,dat,tod.info['ipix']) tod2map_cached(self.caches,dat,ipix) else: if do_omp: #tod2map_omp(self.map,dat,tod.info['ipix']) tod2map_omp(self.map,dat,ipix) else: #tod2map_simple(self.map,dat,tod.info['ipix']) tod2map_simple(self.map,dat,ipix) if self.purge_pixellization: tod.clear_saved_pix(self.tag) def r_th_maps(self): xvec=np.arange(self.nx) xvec=xvec-xvec.mean() yvec=np.arange(self.ny) yvec=yvec-yvec.mean() ymat,xmat=np.meshgrid(yvec,xvec) rmat=np.sqrt(xmat**2+ymat**2) th=np.arctan2(xmat,ymat) return rmat,th def dot(self,map): tot=np.sum(self.map*map.map) return tot def write(self,fname='map.fits'): header=self.wcs.to_header() if self.npol>1: ind=fname.rfind('.') if ind>0: if fname[ind+1:]=='fits': head=fname[:ind] tail=fname[ind:] else: head=fname tail='.fits' else: head=fname tail='.fits' tmp=np.zeros([self.ny,self.nx]) for i in range(self.npol): tmp[:]=np.squeeze(self.map[:,:,i]).T hdu=fits.PrimaryHDU(tmp,header=header) try: hdu.writeto(head+'_'+self.pols[i]+tail,overwrite=True) except: hdu.writeto(head+'_'+self.pols[i]+tail,clobber=True) return if True: #try a patch to fix the wcs xxx tmp=self.map.transpose().copy() hdu=fits.PrimaryHDU(tmp,header=header) else: hdu=fits.PrimaryHDU(self.map,header=header) try: hdu.writeto(fname,overwrite=True) except: hdu.writeto(fname,clobber=True) def __mul__(self,map): if self.npol==1: new_map=self.copy() new_map.map[:]=self.map[:]*map.map[:] return new_map else: assert(map.poltag+'_PRECON'==self.poltag) new_map=map.copy() if self.poltag=='QU_PRECON': new_map.map[:,:,0]=self.map[:,:,0]*map.map[:,:,0]+self.map[:,:,2]*map.map[:,:,1] new_map.map[:,:,1]=self.map[:,:,2]*map.map[:,:,0]+self.map[:,:,1]*map.map[:,:,1] return new_map if self.poltag=='IQU_PRECON': #the indices are set such that the preconditioner matrix [I Q U; Q QQ QU; U QU UU] match the C code. #once we've inverted, the output should be the product of that matrix times [I Q U] new_map.map[:,:,0]=self.map[:,:,0]*map.map[:,:,0]+self.map[:,:,1]*map.map[:,:,1]+self.map[:,:,2]*map.map[:,:,2] new_map.map[:,:,1]=self.map[:,:,1]*map.map[:,:,0]+self.map[:,:,3]*map.map[:,:,1]+self.map[:,:,4]*map.map[:,:,2] new_map.map[:,:,2]=self.map[:,:,2]*map.map[:,:,0]+self.map[:,:,4]*map.map[:,:,1]+self.map[:,:,5]*map.map[:,:,2] return new_map print('unrecognized tag in PolMap.__mul__: ' + repr(self.poltag)) assert(1==0) def mpi_reduce(self,chunksize=1e5): #chunksize is added since at least on my laptop mpi4py barfs if it #tries to reduce an nside=512 healpix map, so need to break it into pieces. if have_mpi: #print("reducing map") if chunksize>0: nchunk=(1.0*self.nx*self.ny*self.npol)/chunksize nchunk=int(np.ceil(nchunk)) else: nchunk=1 #print('nchunk is ',nchunk) if nchunk==1: self.map=comm.allreduce(self.map) else: inds=np.asarray(np.linspace(0,self.nx*self.ny*self.npol,nchunk+1),dtype='int') if len(self.map.shape)>1: tmp=np.zeros(self.map.size) tmp[:]=np.reshape(self.map,len(tmp)) else: tmp=self.map for i in range(len(inds)-1): tmp[inds[i]:inds[i+1]]=comm.allreduce(tmp[inds[i]:inds[i+1]]) #self.map[inds[i]:inds[i+1]]=comm.allreduce(self.map[inds[i]:inds[i+1]]) #tmp=np.zeros(inds[i+1]-inds[i]) #tmp[:]=self.map[inds[i]:inds[i+1]] #tmp=comm.allreduce(tmp) #self.map[inds[i]:inds[i+1]]=tmp if len(self.map.shape)>1: self.map[:]=np.reshape(tmp,self.map.shape) #print("reduced") class HealMap(SkyMap): def __init__(self,proj='RING',nside=512,tag='ipix'): if not(have_healpy): printf("Healpix map requested, but healpy not found.") return self.proj=proj self.nside=nside self.nx=healpy.nside2npix(self.nside) self.ny=1 self.caches=None self.tag=tag self.map=np.zeros([self.nx,self.ny]) def copy(self): newmap=HealMap(self.proj,self.nside,self.tag) newmap.map[:]=self.map[:] return newmap def pix_from_radec(self,ra,dec): ipix=healpy.ang2pix(self.nside,np.pi/2-dec,ra,self.proj=='NEST') return np.asarray(ipix,dtype='int32') #def get_pix(self,tod,savepix=True): # if not(self.tag is None): # ipix=tod.get_saved_pix(self.tag) # if not(ipix is None): # return ipix # ra,dec=tod.get_radec() # #ipix=healpy.ang2pix(self.nside,np.pi/2-tod.info['dy'],tod.info['dx'],self.proj=='NEST') # ipix=healpy.ang2pix(self.nside,np.pi/2-dec,ra,self.proj=='NEST') # if savepix: # tod.save_pixellization(self.tag,ipix) # return ipix def write(self,fname='map.fits',overwrite=True): if self.map.shape[1]<=1: healpy.write_map(fname,self.map[:,0],nest=(self.proj=='NEST'),overwrite=overwrite) class HealPolMap(PolMap): def __init__(self,poltag='I',proj='RING',nside=512,tag='ipix',purge_pixellization=False): if not(have_healpy): printf("Healpix map requested, but healpy not found.") return pols=poltag2pols(poltag) if pols is None: print('Unrecognized polarization state ' + poltag + ' in PolMap.__init__') return npol=len(pols) self.proj=proj self.nside=nside self.nx=healpy.nside2npix(self.nside) self.ny=1 self.npol=npol self.poltag=poltag self.pols=pols self.caches=None self.tag=tag self.purge_pixellization=purge_pixellization if self.npol>1: self.map=np.zeros([self.nx,self.ny,self.npol]) else: self.map=np.zeros([self.nx,self.ny]) def copy(self): if False: newmap=HealPolMap(self.poltag,self.proj,self.nside,self.tag) newmap.map[:]=self.map[:] return newmap else: return copy.deepcopy(self) #def get_pix(self,tod): # ipix=healpy.ang2pix(self.nside,np.pi/2-tod.info['dy'],tod.info['dx'],self.proj=='NEST') # return ipix def pix_from_radec(self,ra,dec): ipix=healpy.ang2pix(self.nside,np.pi/2-dec,ra,self.proj=='NEST') return np.asarray(ipix,dtype='int32') #def get_pix(self,tod,savepix=True): # if not(self.tag is None): # ipix=tod.get_saved_pix(self.tag) # if not(ipix is None): # return ipix # ra,dec=tod.get_radec() # ipix=self.pix_from_radec(ra,dec) # if savepix: # if not(self.tag is None): # tod.save_pixellization(self.tag,ipix) # return ipix def write(self,fname='map.fits',overwrite=True): if self.map.shape[1]<=1: if self.npol==1: healpy.write_map(fname,self.map[:,0],nest=(self.proj=='NEST'),overwrite=overwrite) else: ind=fname.rfind('.') if ind>0: if fname[ind+1:]=='fits': head=fname[:ind] tail=fname[ind:] else: head=fname tail='.fits' else: head=fname tail='.fits' #tmp=np.zeros([self.ny,self.nx]) tmp=np.zeros(self.nx) for i in range(self.npol): tmp[:]=np.squeeze(self.map[:,:,i]).T #print('tmp shape is ',tmp.shape) fname=head+'_'+self.pols[i]+tail healpy.write_map(fname,tmp,nest=(self.proj=='NEST'),overwrite=overwrite) #healpy.write_map(fname,tmp[:,0],nest=(self.proj=='NEST'),overwrite=overwrite) class Cuts: def __init__(self,tod,do_add=True): #if class(tod)==Cuts: #for use in copy if isinstance(tod,Cuts): self.map=tod.map.copy() self.bad_inds=tod.bad_inds.copy() self.namps=tod.nsamp self.do_add=tod.do_add return bad_inds=np.where(tod.info['bad_samples']) #dims=tod.info['dat_calib'].shape dims=tod.get_data_dims() bad_inds=np.ravel_multi_index(bad_inds,dims) self.nsamp=len(bad_inds) self.inds=bad_inds self.map=np.zeros(self.nsamp) self.do_add=do_add def clear(self): self.map[:]=0 def axpy(self,cuts,a): self.map[:]=self.map[:]+a*cuts.map[:] def map2tod(self,tod,dat): dd=np.ravel(dat) if self.do_add: dd[self.inds]=self.map else: dd[self.inds]+=self.map def tod2map(self,tod,dat): dd=np.ravel(dat) self.map[:]=dd[self.inds] def dot(self,cuts): tot=np.dot(self.map,cuts.map) return tot def copy(self): return Cuts(self) class CutsCompact: def __init__(self,tod): if isinstance(tod,CutsCompact): self.ndet=tod.ndet self.nseg=tod.nseg self.istart=tod.istart self.istop=tod.istop else: #ndet=tod.info['dat_calib'].shape[0] ndet=tod.get_ndet() self.ndet=ndet self.nseg=np.zeros(ndet,dtype='int') self.istart=[None]*ndet self.istop=[None]*ndet #self.imax=tod.info['dat_calib'].shape[1] self.imax=tod.get_ndata() self.imap=None self.map=None def copy(self,deep=True): copy=CutsCompact(self) if deep: if not(self.imap is None): copy.imap=self.imap.copy() if not(self.map is None): copy.map=self.map.copy() else: copy.imap=self.imap copy.map=self.map return copy def add_cut(self,det,istart,istop): if istart>=self.imax: #this is asking to add a cut past the end of the data. return if istop>self.imax: #don't have a cut run past the end of the timestream istop=self.imax self.nseg[det]=self.nseg[det]+1 if self.istart[det] is None: self.istart[det]=[istart] else: self.istart[det].append(istart) if self.istop[det] is None: self.istop[det]=[istop] else: self.istop[det].append(istop) def get_imap(self): ncut=0 for det in range(self.ndet): for i in range(self.nseg[det]): ncut=ncut+(self.istop[det][i]-self.istart[det][i]) print('ncut is ' + repr(ncut)) self.imap=np.zeros(ncut,dtype='int64') icur=0 for det in range(self.ndet): for i in range(self.nseg[det]): istart=det*self.imax+self.istart[det][i] istop=det*self.imax+self.istop[det][i] nn=istop-istart self.imap[icur:icur+nn]=np.arange(istart,istop) icur=icur+nn self.map=np.zeros(len(self.imap)) def cuts_from_array(self,cutmat): for det in range(cutmat.shape[0]): nseg,istart,istop=segs_from_vec(cutmat[det,:]) self.nseg[det]=nseg self.istart[det]=istart self.istop[det]=istop def merge_cuts(self): tmp=np.ones(self.imax+2,dtype='bool') for det in range(self.ndet): if self.nseg[det]>1: #if we only have one segment, don't have to worry about strange overlaps tmp[:]=True for i in range(self.nseg[det]): tmp[(self.istart[det][i]+1):(self.istop[det][i]+1)]=False nseg,istart,istop=segs_from_vec(tmp,pad=False) self.nseg[det]=nseg self.istart[det]=istart self.istop[det]=istop def tod2map(self,tod,mat=None,do_add=True,do_omp=False): if mat is None: #mat=tod.info['dat_calib'] mat=tod.get_data() tod2cuts_c(self.map.ctypes.data,mat.ctypes.data,self.imap.ctypes.data,len(self.imap),do_add) def map2tod(self,tod,mat=None,do_add=True,do_omp=False): if mat is None: #mat=tod.info['dat_calib'] mat=tod.get_data() #print('first element is ' + repr(mat[0,self.imap[0]])) cuts2tod_c(mat.ctypes.data,self.map.ctypes.data,self.imap.ctypes.data,len(self.imap),do_add) #print('first element is now ' + repr(mat[0,self.imap[0]])) #return mat def clear(self): if not(self.map is None): self.map[:]=0 def dot(self,other=None): if self.map is None: return None if other is None: return np.dot(self.map,self.map) else: if other.map is None: return None return np.dot(self.map,other.map) def axpy(self,common,a): self.map=self.map+a*common.map def write(self,fname=None): pass def apply_prior(self,x,Ax): Ax.map=Ax.map+self.map*x.map def __mul__(self,to_mul): tt=self.copy() tt.map=self.map*to_mul.map return tt class SkyMapCar(SkyMap): def pix_from_radec(self,ra,dec): xpix=np.round((ra-self.lims[0])*self.cosdec/self.pixsize) #ypix=np.round((dec-self.lims[2])/self.pixsize) ypix=((dec-self.lims[2])/self.pixsize)+0.5 ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') return ipix class SkyMapCarOld: def __init__(self,lims,pixsize): try: self.lims=lims.copy() except: self.lims=lims[:] self.pixsize=pixsize self.cosdec=np.cos(0.5*(lims[2]+lims[3])) nx=int(np.ceil((lims[1]-lims[0])/pixsize*self.cosdec)) ny=int(np.ceil((lims[3]-lims[2])/pixsize)) self.nx=nx self.ny=ny self.npix=nx*ny self.map=np.zeros([nx,ny]) def copy(self): mycopy=SkyMapCar(self.lims,self.pixsize) mycopy.map[:]=self.map[:] return mycopy def clear(self): self.map[:,:]=0 def axpy(self,map,a): self.map[:]=self.map[:]+a*map.map[:] def assign(self,arr): assert(arr.shape[0]==self.nx) assert(arr.shape[1]==self.ny) self.map[:,:]=arr def get_pix(self,tod): xpix=np.round((tod.info['dx']-self.lims[0])*self.cosdec/self.pixsize) ypix=np.round((tod.info['dy']-self.lims[2])/self.pixsize) #ipix=np.asarray(ypix*self.nx+xpix,dtype='int32') ipix=np.asarray(xpix*self.ny+ypix,dtype='int32') return ipix def map2tod(self,tod,dat,do_add=True,do_omp=True): map2tod(dat,self.map,tod.info['ipix'],do_add,do_omp) def tod2map(self,tod,dat,do_add=True,do_omp=True): if do_add==False: self.clear() if do_omp: tod2map_omp(self.map,dat,tod.info['ipix']) else: tod2map_simple(self.map,dat,tod.info['ipix']) def r_th_maps(self): xvec=np.arange(self.nx) xvec=xvec-xvec.mean() yvec=np.arange(self.ny) yvec=yvec-yvec.mean() ymat,xmat=np.meshgrid(yvec,xvec) rmat=np.sqrt(xmat**2+ymat**2) th=np.arctan2(xmat,ymat) return rmat,th def dot(self,map): tot=np.sum(self.map*map.map) return tot def find_bad_skew_kurt(dat,skew_thresh=6.0,kurt_thresh=5.0): ndet=dat.shape[0] isgood=np.ones(ndet,dtype='bool') skew=np.mean(dat**3,axis=1) mystd=np.std(dat,axis=1) skew=skew/mystd**1.5 mykurt=np.mean(dat**4,axis=1) kurt=mykurt/mystd**4-3 isgood[np.abs(skew)>skew_thresh*np.median(np.abs(skew))]=False isgood[np.abs(kurt)>kurt_thresh*np.median(np.abs(kurt))]=False return skew,kurt,isgood def timestreams_from_gauss(ra,dec,fwhm,tod,pred=None): if pred is None: #pred=np.zeros(tod.info['dat_calib'].shape) pred=tod.get_empty(True) #n=tod.info['dat_calib'].size n=np.product(tod.get_data_dims()) assert(pred.size==n) npar_src=4 #x,y,sig,amp dx=tod.info['dx'] dy=tod.info['dy'] pp=np.zeros(npar_src) pp[0]=ra pp[1]=dec pp[2]=fwhm/np.sqrt(8*np.log(2))*np.pi/180/3600 pp[3]=1 fill_gauss_src_c(pp.ctypes.data,dx.ctypes.data,dy.ctypes.data,pred.ctypes.data,n) return pred def timestreams_from_isobeta_c(params,tod,pred=None): if pred is None: #pred=np.zeros(tod.info['dat_calib'].shape) pred=tod.get_empty(True) #n=tod.info['dat_calib'].size n=np.product(tod.get_data_dims()) assert(pred.size==n) dx=tod.info['dx'] dy=tod.info['dy'] fill_isobeta_c(params.ctypes.data,dx.ctypes.data,dy.ctypes.data,pred.ctypes.data,n) npar_beta=5 #x,y,theta,beta,amp npar_src=4 #x,y,sig,amp nsrc=(params.size-npar_beta)//npar_src for i in range(nsrc): pp=np.zeros(npar_src) ioff=i*npar_src+npar_beta pp[:]=params[ioff:(ioff+npar_src)] fill_gauss_src_c(pp.ctypes.data,dx.ctypes.data,dy.ctypes.data,pred.ctypes.data,n) return pred def derivs_from_elliptical_isobeta(params,tod,*args,**kwargs): npar=len(params) assert(npar==7) pred=tod.get_empty() dims=np.hstack([npar,pred.shape]) derivs=np.empty(dims) dx=tod.info['dx'] dy=tod.info['dy'] minkasi_nb.fill_elliptical_isobeta_derivs(params,dx,dy,pred,derivs) return derivs,pred def derivs_from_elliptical_gauss(params,tod,*args,**kwargs): npar=len(params) assert(npar==6) pred=tod.get_empty() dims=np.hstack([npar,pred.shape]) derivs=np.empty(dims) dx=tod.info['dx'] dy=tod.info['dy'] minkasi_nb.fill_elliptical_gauss_derivs(params,dx,dy,pred,derivs) return derivs,pred def derivs_from_isobeta_c(params,tod,*args,**kwargs): npar=5; #n=tod.info['dat_calib'].size dims=tod.get_data_dims() n=np.product(dims) #sz_deriv=np.append(npar,tod.info['dat_calib'].shape) sz_deriv=np.append(npar,dims) #pred=np.zeros(tod.info['dat_calib'].shape) pred=tod.get_empty(True) derivs=np.zeros(sz_deriv) dx=tod.info['dx'] dy=tod.info['dy'] fill_isobeta_derivs_c(params.ctypes.data,dx.ctypes.data,dy.ctypes.data,pred.ctypes.data,derivs.ctypes.data,n) return derivs,pred def derivs_from_gauss_c(params,tod,*args,**kwargs): npar=4 #n=tod.info['dat_calib'].size n=tod.get_nsamp() #sz_deriv=np.append(npar,tod.info['dat_calib'].shape) sz_deriv=np.append(npar,tod.get_data_dims()) #pred=np.zeros(tod.info['dat_calib'].shape) pred=tod.get_empty(True) derivs=np.zeros(sz_deriv) dx=tod.info['dx'] dy=tod.info['dy'] fill_gauss_derivs_c(params.ctypes.data,dx.ctypes.data,dy.ctypes.data,pred.ctypes.data,derivs.ctypes.data,n) return derivs,pred def derivs_from_map(pars,tod,fun,map,dpar,do_symm=False,*args,**kwargs): #print('do_symm is ',do_symm) pred=tod.get_empty() fun(map,pars,*args,**kwargs) map.map2tod(tod,pred,False) npar=len(pars) tmp=tod.get_empty() if do_symm: tmp2=tod.get_empty() derivs=np.empty([npar,pred.shape[0],pred.shape[1]]) for i in range(npar): pp=pars.copy() pp[i]=pp[i]+dpar[i] fun(map,pp,*args,**kwargs) tmp[:]=0 #strictly speaking, we shouldn't need this, but it makes us more robust to bugs elsewhere map.map2tod(tod,tmp,False) if do_symm: pp=pars.copy() pp[i]=pp[i]-dpar[i] fun(map,pp,*args,**kwargs) tmp2[:]=0 map.map2tod(tod,tmp2,False) derivs[i,:,:]=(tmp-tmp2)/(2*dpar[i]) else: derivs[i,:,:]=(tmp-pred)/(dpar[i]) #pred=np.reshape(pred,pred.size) #derivs=np.reshape(derivs,[derivs.shape[0],derivs.shape[1]*derivs.shape[2]]) return derivs,pred def timestreams_from_isobeta(params,tod): npar_beta=5 #x,y,theta,beta,amp npar_src=4 #x,y,sig,amp nsrc=(params.size-npar_beta)//npar_src assert(params.size==nsrc*npar_src+npar_beta) x0=params[0] y0=params[1] theta=params[2] beta=params[3] amp=params[4] cosdec=np.cos(y0) dx=(tod.info['dx']-x0)*cosdec dy=tod.info['dy']-y0 rsqr=dx*dx+dy*dy rsqr=rsqr/theta**2 #print rsqr.max() pred=amp*(1+rsqr)**(0.5-1.5*beta) for i in range(nsrc): src_x=params[i*npar_src+npar_beta+0] src_y=params[i*npar_src+npar_beta+1] src_sig=params[i*npar_src+npar_beta+2] src_amp=params[i*npar_src+npar_beta+3] dx=tod.info['dx']-src_x dy=tod.info['dy']-src_y rsqr=( (dx*np.cos(src_y))**2+dy**2) pred=pred+src_amp*np.exp(-0.5*rsqr/src_sig**2) return pred def isobeta_src_chisq(params,tods): chisq=0.0 for tod in tods.tods: pred=timestreams_from_isobeta_c(params,tod) #chisq=chisq+tod.timestream_chisq(tod.info['dat_calib']-pred) chisq=chisq+tod.timestream_chisq(tod.get_data()-pred) return chisq npar_beta=5 #x,y,theta,beta,amp npar_src=4 #x,y,sig,amp nsrc=(params.size-npar_beta)//npar_src assert(params.size==nsrc*npar_src+npar_beta) x0=params[0] y0=params[1] theta=params[2] beta=params[3] amp=params[4] cosdec=np.cos(y0) chisq=0.0 for tod in tods.tods: dx=tod.info['dx']-x0 dy=tod.info['dy']-y0 rsqr=(dx*cosdec)**2+dy**2 pred=amp*(1+rsqr/theta**2)**(0.5-1.5*beta) for i in range(nsrc): src_x=params[i*npar_src+npar_beta+0] src_y=params[i*npar_src+npar_beta+1] src_sig=params[i*npar_src+npar_beta+2] src_amp=params[i*npar_src+npar_beta+3] dx=tod.info['dx']-src_x dy=tod.info['dy']-src_y rsqr=( (dx*np.cos(src_y))**2+dy**2) pred=pred+src_amp*np.exp(-0.5*rsqr/src_sig**2) #chisq=chisq+tod.timestream_chisq(tod.info['dat_calib']-pred) chisq=chisq+tod.timestream_chisq(tod.get_data()-pred) return chisq class NoiseBinnedDet: def __init__(self,dat,dt,freqs=None,scale_facs=None): ndet=dat.shape[0] ndata=dat.shape[1] nn=2*(ndata-1) dnu=1/(nn*dt) bins=np.asarray(freqs/dnu,dtype='int') bins=bins[bins<ndata] bins=np.hstack([bins,ndata]) if bins[0]>0: bins=np.hstack([0,bins]) if bins[0]<0: bins[0]=0 self.bins=bins nbin=len(bins)-1 self.nbin=nbin det_ps=np.zeros([ndet,nbin]) datft=mkfftw.fft_r2r(dat) for i in range(nbin): det_ps[:,i]=1.0/np.mean(datft[:,bins[i]:bins[i+1]]**2,axis=1) self.det_ps=det_ps self.ndata=ndata self.ndet=ndet self.nn=nn def apply_noise(self,dat): datft=mkfftw.fft_r2r(dat) for i in range(self.nbin): #datft[:,self.bins[i]:self.bins[i+1]]=datft[:,self.bins[i]:self.bins[i+1]]*np.outer(self.det_ps[:,i],self.bins[i+1]-self.bins[i]) datft[:,self.bins[i]:self.bins[i+1]]=datft[:,self.bins[i]:self.bins[i+1]]*np.outer(self.det_ps[:,i],np.ones(self.bins[i+1]-self.bins[i])) dd=mkfftw.fft_r2r(datft) dd[:,0]=0.5*dd[:,0] dd[:,-1]=0.5*dd[:,-1] return dd class NoiseWhite: def __init__(self,dat): #this is the ratio between the median absolute #deviation of the diff and sigma fac=scipy.special.erfinv(0.5)*2 sigs=np.median(np.abs(np.diff(dat,axis=1)),axis=1)/fac self.sigs=sigs self.weights=1/sigs**2 def apply_noise(self,dat): assert(dat.shape[0]==len(self.weights)) ndet=dat.shape[0] for i in range(ndet): dat[i,:]=dat[i,:]*self.weights[i] return dat class NoiseWhiteNotch: def __init__(self,dat,numin,numax,tod): fac=scipy.special.erfinv(0.5)*2 sigs=np.median(np.abs(np.diff(dat,axis=1)),axis=1)/fac self.sigs=sigs self.weights=1/sigs**2 self.weights=self.weights/(2*(dat.shape[1]-1)) #fold in fft normalization to the weights tvec=tod.get_tvec() dt=np.median(np.diff(tvec)) tlen=tvec[-1]-tvec[0] dnu=1.0/(2*tlen-dt) self.istart=int(np.floor(numin/dnu)) self.istop=int(np.ceil(numax/dnu))+1 def apply_noise(self,dat): assert(dat.shape[0]==len(self.weights)) datft=mkfftw.fft_r2r(dat) datft[:,self.istart:self.istop]=0 dat=mkfftw.fft_r2r(datft) ndet=dat.shape[0] for i in range(ndet): dat[i,:]=dat[i,:]*self.weights[i] return dat class NoiseBinnedEig: def __init__(self,dat,dt,freqs=None,scale_facs=None,thresh=5.0): ndet=dat.shape[0] ndata=dat.shape[1] nn=2*(ndata-1) mycov=np.dot(dat,dat.T) mycov=0.5*(mycov+mycov.T) ee,vv=np.linalg.eig(mycov) mask=ee>thresh*thresh*np.median(ee) vecs=vv[:,mask] ts=np.dot(vecs.T,dat) resid=dat-np.dot(vv[:,mask],ts) dnu=1/(nn*dt) print('dnu is ' + repr(dnu)) bins=np.asarray(freqs/dnu,dtype='int') bins=bins[bins<ndata] bins=np.hstack([bins,ndata]) if bins[0]>0: bins=np.hstack([0,bins]) if bins[0]<0: bins[0]=0 self.bins=bins nbin=len(bins)-1 self.nbin=nbin nmode=ts.shape[0] det_ps=np.zeros([ndet,nbin]) mode_ps=np.zeros([nmode,nbin]) residft=mkfftw.fft_r2r(resid) modeft=mkfftw.fft_r2r(ts) for i in range(nbin): det_ps[:,i]=1.0/np.mean(residft[:,bins[i]:bins[i+1]]**2,axis=1) mode_ps[:,i]=1.0/np.mean(modeft[:,bins[i]:bins[i+1]]**2,axis=1) self.modes=vecs.copy() if not(np.all(np.isfinite(det_ps))): print("warning - have non-finite numbers in noise model. This should not be unexpected.") det_ps[~np.isfinite(det_ps)]=0.0 self.det_ps=det_ps self.mode_ps=mode_ps self.ndata=ndata self.ndet=ndet self.nn=nn def apply_noise(self,dat): assert(dat.shape[0]==self.ndet) assert(dat.shape[1]==self.ndata) datft=mkfftw.fft_r2r(dat) for i in range(self.nbin): n=self.bins[i+1]-self.bins[i] #print('bins are ',self.bins[i],self.bins[i+1],n,datft.shape[1]) tmp=self.modes*np.outer(self.det_ps[:,i],np.ones(self.modes.shape[1])) mat=np.dot(self.modes.T,tmp) mat=mat+np.diag(self.mode_ps[:,i]) mat_inv=np.linalg.inv(mat) Ax=datft[:,self.bins[i]:self.bins[i+1]]*np.outer(self.det_ps[:,i],np.ones(n)) tmp=np.dot(self.modes.T,Ax) tmp=np.dot(mat_inv,tmp) tmp=np.dot(self.modes,tmp) tmp=Ax-tmp*np.outer(self.det_ps[:,i],np.ones(n)) datft[:,self.bins[i]:self.bins[i+1]]=tmp #print(tmp.shape,mat.shape) dd=mkfftw.fft_r2r(datft) dd[:,0]=0.5*dd[:,0] dd[:,-1]=0.5*dd[:,-1] return dd class NoiseCMWhite: def __init__(self,dat): print('setting up noise cm white') u,s,v=np.linalg.svd(dat,0) self.ndet=len(s) ind=np.argmax(s) self.v=np.zeros(self.ndet) self.v[:]=u[:,ind] pred=np.outer(self.v*s[ind],v[ind,:]) dat_clean=dat-pred myvar=np.std(dat_clean,1)**2 self.mywt=1.0/myvar def apply_noise(self,dat,dd=None): t1=time.time() mat=np.dot(self.v,np.diag(self.mywt)) lhs=np.dot(self.v,mat.T) rhs=np.dot(mat,dat) if isinstance(lhs,np.ndarray): cm=np.dot(np.linalg.inv(lhs),rhs) else: cm=rhs/lhs t2=time.time() if dd is None: dd=np.empty(dat.shape) if have_numba: np.outer(-self.v,cm,dd) t3=time.time() #dd[:]=dd[:]+dat minkasi_nb.axpy_in_place(dd,dat) minkasi_nb.scale_matrix_by_vector(dd,self.mywt) else: dd=dat-np.outer(self.v,cm) #print(dd[:4,:4]) t3=time.time() tmp=np.repeat([self.mywt],len(cm),axis=0).T dd=dd*tmp t4=time.time() #print(t2-t1,t3-t2,t4-t3) return dd def get_det_weights(self): return self.mywt.copy() class NoiseSmoothedSVD: def __init__(self,dat_use,fwhm=50,prewhiten=False,fit_powlaw=False,u_in=None): if prewhiten: noisevec=np.median(np.abs(np.diff(dat_use,axis=1)),axis=1) dat_use=dat_use/(np.repeat([noisevec],dat_use.shape[1],axis=0).transpose()) if u_in is None: u,s,v=np.linalg.svd(dat_use,0) ndet=s.size else: u=u_in assert(u.shape[0]==u.shape[1]) ndet=u.shape[0] #print(u.shape,s.shape,v.shape) print('got svd') n=dat_use.shape[1] self.v=np.zeros([ndet,ndet]) self.v[:]=u.transpose() if u_in is None: self.vT=self.v.T else: self.vT=np.linalg.inv(self.v) dat_rot=np.dot(self.v,dat_use) if fit_powlaw: spec_smooth=0*dat_rot for ind in range(ndet): fitp,datsqr,C=fit_ts_ps(dat_rot[ind,:]); spec_smooth[ind,1:]=C else: dat_trans=mkfftw.fft_r2r(dat_rot) spec_smooth=smooth_many_vecs(dat_trans**2,fwhm) spec_smooth[:,1:]=1.0/spec_smooth[:,1:] spec_smooth[:,0]=0 if prewhiten: self.noisevec=noisevec.copy() else: self.noisevec=None self.mywt=spec_smooth def apply_noise(self,dat): if not(self.noisevec is None): noisemat=np.repeat([self.noisevec],dat.shape[1],axis=0).transpose() dat=dat/noisemat dat_rot=np.dot(self.v,dat) datft=mkfftw.fft_r2r(dat_rot) nn=datft.shape[1] datft=datft*self.mywt[:,:nn] dat_rot=mkfftw.fft_r2r(datft) #dat=np.dot(self.v.T,dat_rot) dat=np.dot(self.vT,dat_rot) dat[:,0]=0.5*dat[:,0] dat[:,-1]=0.5*dat[:,-1] if not(self.noisevec is None): #noisemat=np.repeat([self.noisevec],dat.shape[1],axis=0).transpose() dat=dat/noisemat return dat def apply_noise_wscratch(self,dat,tmp,tmp2): if not(self.noisevec is None): noisemat=np.repeat([self.noisevec],dat.shape[1],axis=0).transpose() dat=dat/noisemat dat_rot=tmp dat_rot=np.dot(self.v,dat,dat_rot) dat=tmp2 datft=dat datft=mkfftw.fft_r2r(dat_rot,datft) nn=datft.shape[1] datft[:]=datft*self.mywt[:,:nn] dat_rot=tmp dat_rot=mkfftw.fft_r2r(datft,dat_rot) #dat=np.dot(self.v.T,dat_rot) dat=np.dot(self.vT,dat_rot,dat) dat[:,0]=0.5*dat[:,0] dat[:,-1]=0.5*dat[:,-1] if not(self.noisevec is None): #noisemat=np.repeat([self.noisevec],dat.shape[1],axis=0).transpose() dat=dat/noisemat return dat def get_det_weights(self): """Find the per-detector weights for use in making actual noise maps.""" mode_wt=np.sum(self.mywt,axis=1) #tmp=np.dot(self.v.T,np.dot(np.diag(mode_wt),self.v)) tmp=np.dot(self.vT,np.dot(np.diag(mode_wt),self.v)) return np.diag(tmp).copy()*2.0 class Tod: def __init__(self,info): self.info=info.copy() self.jumps=None self.cuts=None self.noise=None self.noise_delayed=False def lims(self): xmin=self.info['dx'].min() xmax=self.info['dx'].max() ymin=self.info['dy'].min() ymax=self.info['dy'].max() return xmin,xmax,ymin,ymax def set_apix(self): '''calculates dxel normalized to +-1 from elevation''' #TBD pass in and calculate scan center's elevation vs time elev=np.mean(self.info['elev'],axis=0) x=np.arange(elev.shape[0])/elev.shape[0] a=np.polyfit(x,elev,2) ndet=self.info['elev'].shape[0] track_elev,xel=np.meshgrid(a[2]+a[1]*x+a[0]*x**2,np.ones(ndet)) delev=self.info['elev'] - track_elev ml=np.max(np.abs(delev)) self.info['apix']=delev/ml def get_ndet(self): return self.info['dat_calib'].shape[0] def get_ndata(self): return self.info['dat_calib'].shape[1] def get_nsamp(self): #get total number of timestream samples, not samples per detector #return np.product(self.info['dat_calib'].shape) return self.get_ndet()*self.get_ndata() def get_saved_pix(self,tag=None): if tag is None: return None if tag in self.info.keys(): return self.info[tag] else: return None def clear_saved_pix(self,tag=None): if tag is None: return if tag in self.info.keys(): del(self.info[tag]) def save_pixellization(self,tag,ipix): if tag in self.info.keys(): print('warning - overwriting key ',tag,' in tod.save_pixellization.') self.info[tag]=ipix def get_data_dims(self): return (self.get_ndet(),self.get_ndata()) #dims=self.info['dat_calib'].shape #if len(dims)==1: # dims=np.asarray([1,dims[0]],dtype='int') #return dims #return self.info['dat_calib'].shape def get_data(self): return self.info['dat_calib'] def get_tvec(self): return self.info['ctime'] def get_radec(self): return self.info['dx'],self.info['dy'] def get_empty(self,clear=False): if 'dtype' in self.info.keys(): dtype=self.info['dtype'] elif 'dat_calib' in self.info.keys(): dtype=self.info['dat_calib'].dtype else: dtype='float' if clear: #return np.zeros(self.info['dat_calib'].shape,dtype=self.info['dat_calib'].dtype) return np.zeros([self.get_ndet(),self.get_ndata()],dtype=dtype) else: #return np.empty(self.info['dat_calib'].shape,dtype=self.info['dat_calib'].dtype) return np.empty([self.get_ndet(),self.get_ndata()],dtype=dtype) def set_tag(self,tag): self.info['tag']=tag def set_pix(self,map): ipix=map.get_pix(self) #self.info['ipix']=ipix self.info[map.tag]=ipix def copy(self,copy_info=False): if copy_info: myinfo=self.info.copy() for key in myinfo.keys(): try: myinfo[key]=self.info[key].copy() except: pass tod=Tod(myinfo) else: tod=Tod(self.info) if not(self.jumps is None): try: tod.jumps=self.jumps.copy() except: tod.jumps=self.jumps[:] if not(self.cuts is None): try: tod.cuts=self.cuts.copy() except: tod.cuts=self.cuts[:] tod.cuts=self.cuts[:] tod.noise=self.noise return tod def set_noise(self,modelclass=NoiseSmoothedSVD,dat=None,delayed=False,*args,**kwargs): if delayed: self.noise_args=copy.deepcopy(args) self.noise_kwargs=copy.deepcopy(kwargs) self.noise_delayed=True self.noise_modelclass=modelclass else: self.noise_delayed=False if dat is None: dat=self.info['dat_calib'] self.noise=modelclass(dat,*args,**kwargs) def get_det_weights(self): if self.noise is None: print("noise model not set in get_det_weights.") return None try: return self.noise.get_det_weights() except: print("noise model does not support detector weights in get_det_weights.") return None def set_noise_white_masked(self): self.info['noise']='white_masked' self.info['mywt']=np.ones(self.info['dat_calib'].shape[0]) def apply_noise_white_masked(self,dat=None): if dat is None: dat=self.info['dat_calib'] dd=self.info['mask']*dat*np.repeat([self.info['mywt']],self.info['dat_calib'].shape[1],axis=0).transpose() return dd def set_noise_cm_white(self): print('deprecated usage - please switch to tod.set_noise(minkasi.NoiseCMWhite)') self.set_noise(NoiseCMWhite) return u,s,v=np.linalg.svd(self.info['dat_calib'],0) ndet=len(s) ind=np.argmax(s) mode=np.zeros(ndet) #mode[:]=u[:,0] #bug fixes pointed out by Fernando Zago. 19 Nov 2019 #pred=np.outer(mode,v[0,:]) mode[:]=u[:,ind] pred=np.outer(mode*s[ind],v[ind,:]) dat_clean=self.info['dat_calib']-pred myvar=np.std(dat_clean,1)**2 self.info['v']=mode self.info['mywt']=1.0/myvar self.info['noise']='cm_white' def apply_noise_cm_white(self,dat=None): print("I'm not sure how you got here (tod.apply_noise_cm_white), but you should not have been able to. Please complain to someone.") if dat is None: dat=self.info['dat_calib'] mat=np.dot(self.info['v'],np.diag(self.info['mywt'])) lhs=np.dot(self.info['v'],mat.transpose()) rhs=np.dot(mat,dat) #if len(lhs)>1: if isinstance(lhs,np.ndarray): cm=np.dot(np.linalg.inv(lhs),rhs) else: cm=rhs/lhs dd=dat-np.outer(self.info['v'],cm) tmp=np.repeat([self.info['mywt']],len(cm),axis=0).transpose() dd=dd*tmp return dd def set_noise_binned_eig(self,dat=None,freqs=None,scale_facs=None,thresh=5.0): if dat is None: dat=self.info['dat_calib'] mycov=np.dot(dat,dat.T) mycov=0.5*(mycov+mycov.T) ee,vv=np.linalg.eig(mycov) mask=ee>thresh*thresh*np.median(ee) vecs=vv[:,mask] ts=np.dot(vecs.T,dat) resid=dat-np.dot(vv[:,mask],ts) return resid def set_noise_smoothed_svd(self,fwhm=50,func=None,pars=None,prewhiten=False,fit_powlaw=False): '''If func comes in as not empty, assume we can call func(pars,tod) to get a predicted model for the tod that we subtract off before estimating the noise.''' print('deprecated usage - please switch to tod.set_noise(minkasi.NoiseSmoothedSVD)') if func is None: self.set_noise(NoiseSmoothedSVD,self.info['dat_calib']) else: dat_use=func(pars,self) dat_use=self.info['dat_calib']-dat_use self.set_noise(NoiseSmoothedSVD,dat_use) return if func is None: dat_use=self.info['dat_calib'] else: dat_use=func(pars,self) dat_use=self.info['dat_calib']-dat_use #u,s,v=numpy.linalg.svd(self.info['dat_calib']-tmp,0) if prewhiten: noisevec=np.median(np.abs(np.diff(dat_use,axis=1)),axis=1) dat_use=dat_use/(np.repeat([noisevec],dat_use.shape[1],axis=0).transpose()) u,s,v=np.linalg.svd(dat_use,0) print('got svd') ndet=s.size n=self.info['dat_calib'].shape[1] self.info['v']=np.zeros([ndet,ndet]) self.info['v'][:]=u.transpose() dat_rot=np.dot(self.info['v'],self.info['dat_calib']) if fit_powlaw: spec_smooth=0*dat_rot for ind in range(ndet): fitp,datsqr,C=fit_ts_ps(dat_rot[ind,:]); spec_smooth[ind,1:]=C else: dat_trans=mkfftw.fft_r2r(dat_rot) spec_smooth=smooth_many_vecs(dat_trans**2,fwhm) spec_smooth[:,1:]=1.0/spec_smooth[:,1:] spec_smooth[:,0]=0 if prewhiten: self.info['noisevec']=noisevec.copy() self.info['mywt']=spec_smooth self.info['noise']='smoothed_svd' #return dat_rot def apply_noise(self,dat=None): if dat is None: #dat=self.info['dat_calib'] dat=self.get_data().copy() #the .copy() is here so you don't #overwrite data stored in the TOD. if self.noise_delayed: self.noise=self.noise_modelclass(dat,*(self.noise_args), **(self.noise_kwargs)) self.noise_delayed=False try: return self.noise.apply_noise(dat) except: print("unable to use class-based noised, falling back onto hardwired.") if self.info['noise']=='cm_white': #print 'calling cm_white' return self.apply_noise_cm_white(dat) if self.info['noise']=='white_masked': return self.apply_noise_white_masked(dat) #if self.info.has_key('noisevec'): if 'noisevec' in self.info: noisemat=np.repeat([self.info['noisevec']],dat.shape[1],axis=0).transpose() dat=dat/noisemat dat_rot=np.dot(self.info['v'],dat) datft=mkfftw.fft_r2r(dat_rot) nn=datft.shape[1] datft=datft*self.info['mywt'][:,0:nn] dat_rot=mkfftw.fft_r2r(datft) dat=np.dot(self.info['v'].transpose(),dat_rot) #if self.info.has_key('noisevec'): if 'noisevec' in self.info: dat=dat/noisemat dat[:,0]=0.5*dat[:,0] dat[:,-1]=0.5*dat[:,-1] return dat def mapset2tod(self,mapset,dat=None): if dat is None: #dat=0*self.info['dat_calib'] dat=self.get_empty(True) for map in mapset.maps: map.map2tod(self,dat) return dat def tod2mapset(self,mapset,dat=None): if dat is None: #dat=self.info['dat_calib'] dat=self.get_data() for map in mapset.maps: map.tod2map(self,dat) def dot(self,mapset,mapset_out,times=False): #tmp=0.0*self.info['dat_calib'] #for map in mapset.maps: # map.map2tod(self,tmp) t1=time.time() tmp=self.mapset2tod(mapset) t2=time.time() tmp=self.apply_noise(tmp) t3=time.time() self.tod2mapset(mapset_out,tmp) t4=time.time() #for map in mapset_out.maps: # map.tod2map(self,tmp) if times: return(np.asarray([t2-t1,t3-t2,t4-t3])) def set_jumps(self,jumps): self.jumps=jumps def cut_detectors(self,isgood): #cut all detectors not in boolean array isgood isbad=np.asarray(1-isgood,dtype='bool') bad_inds=np.where(isbad) bad_inds=np.fliplr(bad_inds) bad_inds=bad_inds[0] print(bad_inds) nkeep=np.sum(isgood) for key in self.info.keys(): if isinstance(self.info[key],np.ndarray): self.info[key]=slice_with_copy(self.info[key],isgood) if not(self.jumps is None): for i in bad_inds: print('i in bad_inds is ',i) del(self.jumps[i]) if not(self.cuts is None): for i in bad_inds: del(self.cuts[i]) def timestream_chisq(self,dat=None): if dat is None: dat=self.info['dat_calib'] dat_filt=self.apply_noise(dat) chisq=np.sum(dat_filt*dat) return chisq def prior_from_skymap(self,skymap): """stuff. prior_from_skymap(self,skymap): Given e.g. the gradient of a map that has been zeroed under some threshold, return a CutsCompact object that can be used as a prior for solving for per-sample deviations due to strong map gradients. This is to reduce X's around bright sources. The input map should be a SkyMap that is non-zero where one wishes to solve for the per-sample deviations, and the non-zero values should be the standard deviations expected in those pixel. The returned CutsCompact object will have the weight (i.e. 1/input squared) in its map. """ tmp=np.zeros(self.info['dat_calib'].shape) skymap.map2tod(self,tmp) mask=(tmp==0) prior=CutsCompact(self) prior.cuts_from_array(mask) prior.get_imap() prior.tod2map(self,tmp) prior.map=1.0/prior.map**2 return prior def slice_with_copy(arr,ind): if isinstance(arr,np.ndarray): myshape=arr.shape if len(myshape)==1: ans=np.zeros(ind.sum(),dtype=arr.dtype) print(ans.shape) print(ind.sum()) ans[:]=arr[ind] else: mydims=np.append(np.sum(ind),myshape[1:]) print(mydims,mydims.dtype) ans=np.zeros(mydims,dtype=arr.dtype) ans[:,:]=arr[ind,:].copy() return ans return None #should not get here class TodVec: def __init__(self): self.tods=[] self.ntod=0 def add_tod(self,tod): self.tods.append(tod.copy()) self.tods[-1].set_tag(self.ntod) self.ntod=self.ntod+1 def lims(self): if self.ntod==0: return None xmin,xmax,ymin,ymax=self.tods[0].lims() for i in range(1,self.ntod): x1,x2,y1,y2=self.tods[i].lims() xmin=min(x1,xmin) xmax=max(x2,xmax) ymin=min(y1,ymin) ymax=max(y2,ymax) if have_mpi: print('before reduction lims are ',[xmin,xmax,ymin,ymax]) xmin=comm.allreduce(xmin,op=MPI.MIN) xmax=comm.allreduce(xmax,op=MPI.MAX) ymin=comm.allreduce(ymin,op=MPI.MIN) ymax=comm.allreduce(ymax,op=MPI.MAX) print('after reduction lims are ',[xmin,xmax,ymin,ymax]) return [xmin,xmax,ymin,ymax] def set_pix(self,map): for tod in self.tods: #ipix=map.get_pix(tod) #tod.info['ipix']=ipix tod.set_pix(map) def set_apix(self): for tod in self.tods: tod.set_apix() def dot_cached(self,mapset,mapset2=None): nthread=get_nthread() mapset2.get_caches() for i in range(self.ntod): tod=self.tods[i] tod.dot(mapset,mapset2) mapset2.clear_caches() if have_mpi: mapset2.mpi_reduce() return mapset2 def get_nsamp(self,reduce=True): tot=0 for tod in self.tods: tot=tot+tod.get_nsamp() if reduce: if have_mpi: tot=comm.allreduce(tot) return tot def dot(self,mapset,mapset2=None,report_times=False,cache_maps=False): if mapset2 is None: mapset2=mapset.copy() mapset2.clear() if cache_maps: mapset2=self.dot_cached(mapset,mapset2) return mapset2 times=np.zeros(self.ntod) tot_times=0 #for tod in self.tods: for i in range(self.ntod): tod=self.tods[i] t1=time.time() mytimes=tod.dot(mapset,mapset2,True) t2=time.time() tot_times=tot_times+mytimes times[i]=t2-t1 if have_mpi: mapset2.mpi_reduce() print(tot_times) if report_times: return mapset2,times else: return mapset2 def make_rhs(self,mapset,do_clear=False): if do_clear: mapset.clear() for tod in self.tods: dat_filt=tod.apply_noise() for map in mapset.maps: map.tod2map(tod,dat_filt) if have_mpi: mapset.mpi_reduce() def read_tod_from_fits_cbass(fname,dopol=False,lat=37.2314,lon=-118.2941,v34=True,nm20=False): f=pyfits.open(fname) raw=f[1].data ra=raw['RA'] dec=raw['DEC'] flag=raw['FLAG'] I=0.5*(raw['I1']+raw['I2']) mjd=raw['MJD'] tvec=(mjd-2455977.5+2400000.5)*86400+1329696000 #(mjd-2455977.5)*86400+1329696000; dt=np.median(np.diff(tvec)) dat={} dat['dx']=np.reshape(np.asarray(ra,dtype='float64'),[1,len(ra)]) dat['dy']=np.reshape(np.asarray(dec,dtype='float64'),[1,len(dec)]) dat['dt']=dt dat['ctime']=tvec if dopol: dat['dx']=np.vstack([dat['dx'],dat['dx']]) dat['dy']=np.vstack([dat['dy'],dat['dy']]) Q=0.5*(raw['Q1']+raw['Q2']) U=0.5*(raw['U1']+raw['U2']) dat['dat_calib']=np.zeros([2,len(Q)]) if v34: #We believe this is the correct sign convention for V34 dat['dat_calib'][0,:]=-U dat['dat_calib'][1,:]=Q else: dat['dat_calib'][0,:]=Q dat['dat_calib'][1,:]=U az=raw['AZ'] el=raw['EL'] #JLS- changing default az/el to radians and not degrees in TOD dat['az']=az*np.pi/180 dat['el']=el*np.pi/180 #dat['AZ']=az #dat['EL']=el #dat['ctime']=tvec dat['mask']=np.zeros([2,len(Q)],dtype='int8') dat['mask'][0,:]=1-raw['FLAG'] dat['mask'][1,:]=1-raw['FLAG'] if have_qp: Q = qp.QPoint(accuracy='low', fast_math=True, mean_aber=True,num_threads=4) #q_bore = Q.azel2bore(dat['AZ'], dat['EL'], 0*dat['AZ'], 0*dat['AZ'], lon*np.pi/180, lat*np.pi/180, dat['ctime']) q_bore = Q.azel2bore(az,el, 0*az, 0*az, lon, lat, dat['ctime']) q_off = Q.det_offset(0.0,0.0,0.0) #ra, dec, sin2psi, cos2psi = Q.bore2radec(q_off, ctime, q_bore) ra, dec, sin2psi, cos2psi = Q.bore2radec(q_off, tvec, q_bore) tmp=np.arctan2(sin2psi,cos2psi) tmp=tmp-np.pi/2 #this seems to be needed to get these coordinates to line up with #the expected, in IAU convention I believe. JLS Nov 12 2020 #dat['twogamma_saved']=np.arctan2(sin2psi,cos2psi) dat['twogamma_saved']=np.vstack([tmp,tmp+np.pi/2]) #print('pointing rms is ',np.std(ra*np.pi/180-dat['dx']),np.std(dec*np.pi/180-dat['dy'])) dat['ra']=ra*np.pi/180 dat['dec']=dec*np.pi/180 else: dat['dat_calib']=np.zeros([1,len(I)]) dat['dat_calib'][:]=I dat['mask']=np.zeros([1,len(I)],dtype='int8') dat['mask'][:]=1-raw['FLAG'] dat['pixid']=[0] dat['fname']=fname if nm20: try: #kludget to read in bonus cuts, which should be in f[3] raw=f[3].data dat['nm20_start']=raw['START'] dat['nm20_stop']=raw['END'] #nm20=0*dat['flag'] print(dat.keys()) nm20=0*dat['mask'] start=dat['nm20_start'] stop=dat['nm20_stop'] for i in range(len(start)): nm20[:,start[i]:stop[i]]=1 #nm20[:,start[i]:stop[i]]=0 dat['mask']=dat['mask']*nm20 except: print('missing nm20 for ',fname) f.close() return dat def read_tod_from_fits(fname,hdu=1,branch=None): f=pyfits.open(fname) raw=f[hdu].data #print 'sum of cut elements is ',np.sum(raw['UFNU']<9e5) try : #read in calinfo (per-scan beam volumes etc) if present calinfo={'calinfo':True} kwds=('scan','bunit','azimuth','elevatio','beameff','apereff','antgain','gainunc','bmaj','bmin','bpa','parang','beamvol','beamvunc')#for now just hardwired ones we want for kwd in kwds: calinfo[kwd]=f[hdu].header[kwd] except KeyError : print('WARNING - calinfo information not found in fits file header - to track JytoK etc you may need to reprocess the fits files using mustangidl > revision 932') calinfo['calinfo']=False pixid=raw['PIXID'] dets=np.unique(pixid) ndet=len(dets) nsamp=len(pixid)/len(dets) if True: ff=180/np.pi xmin=raw['DX'].min()*ff xmax=raw['DX'].max()*ff ymin=raw['DY'].min()*ff ymax=raw['DY'].max()*ff print('nsamp and ndet are ',ndet,nsamp,len(pixid),' on ',fname, 'with lims ',xmin,xmax,ymin,ymax) else: print('nsamp and ndet are ',ndet,nsamp,len(pixid),' on ',fname) #print raw.names dat={} #this bit of odd gymnastics is because a straightforward reshape doesn't seem to leave the data in #memory-contiguous order, which causes problems down the road #also, float32 is a bit on the edge for pointing, so cast to float64 dx=raw['DX'] if not(branch is None): bb=branch*np.pi/180.0 dx[dx>bb]=dx[dx>bb]-2*np.pi #dat['dx']=np.zeros([ndet,nsamp],dtype=type(dx[0])) ndet=int(ndet) nsamp=int(nsamp) dat['dx']=np.zeros([ndet,nsamp],dtype='float64') dat['dx'][:]=np.reshape(dx,[ndet,nsamp])[:] dy=raw['DY'] #dat['dy']=np.zeros([ndet,nsamp],dtype=type(dy[0])) dat['dy']=np.zeros([ndet,nsamp],dtype='float64') dat['dy'][:]=np.reshape(dy,[ndet,nsamp])[:] if 'ELEV' in raw.names: elev=raw['ELEV']*np.pi/180 dat['elev']=np.zeros([ndet,nsamp],dtype='float64') dat['elev'][:]=np.reshape(elev,[ndet,nsamp])[:] tt=np.reshape(raw['TIME'],[ndet,nsamp]) tt=tt[0,:] dt=np.median(np.diff(tt)) dat['dt']=dt pixid=np.reshape(pixid,[ndet,nsamp]) pixid=pixid[:,0] dat['pixid']=pixid dat_calib=raw['FNU'] #print 'shapes are ',raw['FNU'].shape,raw['UFNU'].shape,np.mean(raw['UFNU']>9e5) #dat_calib[raw['UFNU']>9e5]=0.0 #dat['dat_calib']=np.zeros([ndet,nsamp],dtype=type(dat_calib[0])) dat['dat_calib']=np.zeros([ndet,nsamp],dtype='float64') #go to double because why not dat_calib=np.reshape(dat_calib,[ndet,nsamp]) dat['dat_calib'][:]=dat_calib[:] if np.sum(raw['UFNU']>9e5)>0: dat['mask']=np.reshape(raw['UFNU']<9e5,dat['dat_calib'].shape) dat['mask_sum']=np.sum(dat['mask'],axis=0) #print 'cut frac is now ',np.mean(dat_calib==0) #print 'cut frac is now ',np.mean(dat['dat_calib']==0),dat['dat_calib'][0,0] dat['fname']=fname dat['calinfo']=calinfo f.close() return dat def downsample_array_r2r(arr,fac): n=arr.shape[1] nn=int(n/fac) arr_ft=mkfftw.fft_r2r(arr) arr_ft=arr_ft[:,0:nn].copy() arr=mkfftw.fft_r2r(arr_ft)/(2*(n-1)) return arr def downsample_vec_r2r(vec,fac): n=len(vec) nn=int(n/fac) vec_ft=mkfftw.fft_r2r(vec) vec_ft=vec_ft[0:nn].copy() vec=mkfftw.fft_r2r(vec_ft)/(2*(n-1)) return vec def downsample_tod(dat,fac=10): ndata=dat['dat_calib'].shape[1] keys=dat.keys() for key in dat.keys(): try: if len(dat[key].shape)==1: #print('working on downsampling ' + key) #print('shape is ' + repr(dat[key].shape[0])+' '+repr(n)) if len(dat[key]): #print('working on downsampling ' + key) dat[key]=downsample_vec_r2r(dat[key],fac) else: if dat[key].shape[1]==ndata: #print 'downsampling ' + key dat[key]=downsample_array_r2r(dat[key],fac) except: #print 'not downsampling ' + key pass def truncate_tod(dat,primes=[2,3,5,7,11]): n=dat['dat_calib'].shape[1] lens=find_good_fft_lens(n-1,primes) n_new=lens.max()+1 if n_new<n: print('truncating from ',n,' to ',n_new) for key in dat.keys(): try: #print('working on key ' + key) if len(dat[key].shape)==1: if dat[key].shape[0]==n: dat[key]=dat[key][:n_new].copy() else: if dat[key].shape[1]==n: dat[key]=dat[key][:,0:n_new].copy() except: #print('skipping key ' + key) pass def todvec_from_files_octave(fnames): todvec=TodVec() for fname in fnames: info=read_octave_struct(fname) tod=Tod(info) todvec.add_tod(tod) return todvec def make_hits(todvec,map,do_weights=False): hits=map.copy() try: if map.npol>1: hits.set_polstate(map.poltag+'_PRECON') except: pass hits.clear() for tod in todvec.tods: if do_weights: try: weights=tod.get_det_weights() #sz=tod.info['dat_calib'].shape sz=tod.get_data_dims() tmp=np.outer(weights,np.ones(sz[1])) #tmp=np.outer(weights,np.ones(tod.info['dat_calb'].shape[1])) except: print("error in making weight map. Detector weights requested, but do not appear to be present. Do you have a noise model?") else: #tmp=np.ones(tod.info['dat_calib'].shape) tmp=np.ones(tod.get_data_dims()) #if tod.info.has_key('mask'): if 'mask' in tod.info: tmp=tmp*tod.info['mask'] hits.tod2map(tod,tmp) if have_mpi: print('reducing hits') tot=hits.map.sum() print('starting with total hitcount ' + repr(tot)) hits.mpi_reduce() tot=hits.map.sum() print('ending with total hitcount ' + repr(tot)) return hits def decimate(vec,nrep=1): for i in range(nrep): if len(vec)%2: vec=vec[:-1] vec=0.5*(vec[0::2]+vec[1::2]) return vec def plot_ps(vec,downsamp=0): vecft=mkfftw.fft_r2r(vec) def get_wcs(lims,pixsize,proj='CAR',cosdec=None,ref_equ=False): w=wcs.WCS(naxis=2) dec=0.5*(lims[2]+lims[3]) if cosdec is None: cosdec=np.cos(dec) if proj=='CAR': #CAR in FITS seems to already correct for cosin(dec), which has me confused, but whatever... cosdec=1.0 if ref_equ: w.wcs.crval=[0.0,0.0] #this seems to be a needed hack if you want the position sent #in for the corner to actually be the corner. w.wcs.crpix=[lims[1]/pixsize+1,-lims[2]/pixsize+1] #w.wcs.crpix=[lims[1]/pixsize,-lims[2]/pixsize] #print 'crpix is ',w.wcs.crpix else: w.wcs.crpix=[1.0,1.0] w.wcs.crval=[lims[1]*180/np.pi,lims[2]*180/np.pi] w.wcs.cdelt=[-pixsize/cosdec*180/np.pi,pixsize*180/np.pi] w.wcs.ctype=['RA---CAR','DEC--CAR'] return w print('unknown projection type ',proj,' in get_wcs.') return None def get_aligned_map_subregion_car(lims,fname=None,big_wcs=None,osamp=1): """Get a wcs for a subregion of a map, with optionally finer pixellization. Designed for use in e.g. combining ACT maps and Mustang data. Input arguments are RA/Dec limits for the subregion (which will be tweaked as-needed) and either a WCS structure or the name of a FITS file containing the WCS info the sub-region will be aligned with.""" if big_wcs is None: if fname is None: print("Error in get_aligned_map_subregion_car. Must send in either a file or a WCS.") big_wcs=wcs.WCS(fname) ll=np.asarray(lims) ll=ll*180/np.pi #get the ra/dec limits in big pixel coordinates corner1=big_wcs.wcs_world2pix(ll[0],ll[2],0) corner2=big_wcs.wcs_world2pix(ll[1],ll[3],0) #get the pixel edges for the corners. FITS works in #pixel centers, so edges are a half-pixel off corner1[0]=np.ceil(corner1[0])+0.5 corner1[1]=np.floor(corner1[1])-0.5 corner2[0]=np.floor(corner2[0])-0.5 corner2[1]=np.ceil(corner2[1])+0.5 corner1_radec=big_wcs.wcs_pix2world(corner1[0],corner1[1],0) corner2_radec=big_wcs.wcs_pix2world(corner2[0],corner2[1],0) dra=(corner1_radec[0]-corner2_radec[0])/(corner1[0]-corner2[0]) ddec=(corner1_radec[1]-corner2_radec[1])/(corner1[1]-corner2[1]) assert(np.abs(dra/ddec)-1<1e-5) #we are not currently smart enough to deal with rectangular pixels lims_use=np.asarray([corner1_radec[0],corner2_radec[0],corner1_radec[1],corner2_radec[1]]) pixsize=ddec/osamp lims_use=lims_use+np.asarray([0.5,-0.5,0.5,-0.5])*pixsize small_wcs=get_wcs(lims_use*np.pi/180,pixsize*np.pi/180,ref_equ=True) imin=int(np.round(corner2[0]+0.5)) jmin=int(np.round(corner1[1]+0.5)) map_corner=np.asarray([imin,jmin],dtype='int') lims_use=lims_use*np.pi/180 return small_wcs,lims_use,map_corner def fit_linear_ps_uncorr(dat,vecs,tol=1e-3,guess=None,max_iter=15): if guess is None: lhs=np.dot(vecs,vecs.transpose()) rhs=np.dot(vecs,dat**2) guess=np.dot(np.linalg.inv(lhs),rhs) guess=0.5*guess #scale down since we're less likely to run into convergence issues if we start low #print guess fitp=guess.copy() converged=False npp=len(fitp) iter=0 grad_tr=np.zeros(npp) grad_chi=np.zeros(npp) curve=np.zeros([npp,npp]) datsqr=dat*dat while (converged==False): iter=iter+1 C=np.dot(fitp,vecs) Cinv=1.0/C for i in range(npp): grad_chi[i]=0.5*np.sum(datsqr*vecs[i,:]*Cinv*Cinv) grad_tr[i]=-0.5*np.sum(vecs[i,:]*Cinv) for j in range(i,npp): #curve[i,j]=-0.5*np.sum(datsqr*Cinv*Cinv*Cinv*vecs[i,:]*vecs[j,:]) #data-only curvature #curve[i,j]=-0.5*np.sum(Cinv*Cinv*vecs[i,:]*vecs[j,:]) #Fisher curvature curve[i,j]=0.5*np.sum(Cinv*Cinv*vecs[i,:]*vecs[j,:])-np.sum(datsqr*Cinv*Cinv*Cinv*vecs[i,:]*vecs[j,:]) #exact curve[j,i]=curve[i,j] grad=grad_chi+grad_tr curve_inv=np.linalg.inv(curve) errs=np.diag(curve_inv) dp=np.dot(grad,curve_inv) fitp=fitp-dp frac_shift=dp/errs #print dp,errs,frac_shift if np.max(np.abs(frac_shift))<tol: print('successful convergence after ',iter,' iterations with error estimate ',np.max(np.abs(frac_shift))) converged=True print(C[0],C[-1]) if iter==max_iter: print('not converging after ',iter,' iterations in fit_linear_ps_uncorr with current convergence parameter ',np.max(np.abs(frac_shift))) converged=True return fitp def get_curve_deriv_powspec(fitp,nu_scale,lognu,datsqr,vecs): vec=nu_scale**fitp[2] C=fitp[0]+fitp[1]*vec Cinv=1.0/C vecs[1,:]=vec vecs[2,:]=fitp[1]*lognu*vec grad_chi=0.5*np.dot(vecs,datsqr*Cinv*Cinv) grad_tr=-0.5*np.dot(vecs,Cinv) grad=grad_chi+grad_tr np=len(grad_chi) curve=np.zeros([np,np]) for i in range(np): for j in range(i,np): curve[i,j]=0.5*np.sum(Cinv*Cinv*vecs[i,:]*vecs[j,:])-np.sum(datsqr*Cinv*Cinv*Cinv*vecs[i,:]*vecs[j,:]) curve[j,i]=curve[i,j] like=-0.5*sum(datsqr*Cinv)-0.5*sum(np.log(C)) return like,grad,curve,C def fit_ts_ps(dat,dt=1.0,ind=-1.5,nu_min=0.0,nu_max=np.inf,scale_fac=1.0,tol=0.01): datft=mkfftw.fft_r2r(dat) n=len(datft) dnu=0.5/(len(dat)*dt) #coefficient should reflect the type of fft you did... nu=dnu*np.arange(n) isgood=(nu>nu_min)&(nu<nu_max) datft=datft[isgood] nu=nu[isgood] n=len(nu) vecs=np.zeros([2,n]) vecs[0,:]=1.0 #white noise vecs[1,:]=nu**ind guess=fit_linear_ps_uncorr(datft,vecs) pred=np.dot(guess,vecs) #pred=guess[0]*vecs[0]+guess[1]*vecs[1] #return pred rat=vecs[1,:]*guess[1]/(vecs[0,:]*guess[0]) #print 'rat lims are ',rat.max(),rat.min() my_ind=np.max(np.where(rat>1)[0]) nu_ref=np.sqrt(nu[my_ind]*nu[0]) #WAG as to a sensible frequency pivot point #nu_ref=0.2*nu[my_ind] #WAG as to a sensible frequency pivot point #print 'knee is roughly at ',nu[my_ind],nu_ref #model = guess[1]*nu^ind+guess[0] # = guess[1]*(nu/nu_ref*nu_ref)^ind+guess[0] # = guess[1]*(nu_ref)^in*(nu/nu_ref)^ind+guess[0] nu_scale=nu/nu_ref guess_scale=guess.copy() guess_scale[1]=guess[1]*(nu_ref**ind) #print 'guess is ',guess #print 'guess_scale is ',guess_scale C_scale=guess_scale[0]+guess_scale[1]*(nu_scale**ind) fitp=np.zeros(3) fitp[0:2]=guess_scale fitp[2]=ind npp=3 vecs=np.zeros([npp,n]) vecs[0,:]=1.0 lognu=np.log(nu_scale) curve=np.zeros([npp,npp]) grad_chi=np.zeros(npp) grad_tr=np.zeros(npp) datsqr=datft**2 #for robustness, start with downscaling 1/f part fitp[1]=0.5*fitp[1] like,grad,curve,C=get_curve_deriv_powspec(fitp,nu_scale,lognu,datsqr,vecs) lamda=0.0 #print 'starting likelihood is',like for iter in range(50): tmp=curve+lamda*np.diag(np.diag(curve)) curve_inv=np.linalg.inv(tmp) dp=np.dot(grad,curve_inv) trial_fitp=fitp-dp errs=np.sqrt(-np.diag(curve_inv)) frac=dp/errs new_like,new_grad,new_curve,C=get_curve_deriv_powspec(trial_fitp,nu_scale,lognu,datsqr,vecs) if (new_like>like): #if True: like=new_like grad=new_grad curve=new_curve fitp=trial_fitp lamda=update_lamda(lamda,True) else: lamda=update_lamda(lamda,False) if (lamda==0)&(np.max(np.abs(frac))<tol): converged=True else: converged=False if False: vec=nu_scale**fitp[2] C=fitp[0]+fitp[1]*vec Cinv=1.0/C vecs[1,:]=vec vecs[2,:]=fitp[1]*lognu*vec like=-0.5*np.sum(datsqr*Cinv)-0.5*np.sum(np.log(C)) for i in range(np): grad_chi[i]=0.5*np.sum(datsqr*vecs[i,:]*Cinv*Cinv) grad_tr[i]=-0.5*np.sum(vecs[i,:]*Cinv) for j in range(i,np): curve[i,j]=0.5*np.sum(Cinv*Cinv*vecs[i,:]*vecs[j,:])-np.sum(datsqr*Cinv*Cinv*Cinv*vecs[i,:]*vecs[j,:]) curve[j,i]=curve[i,j] grad=grad_chi+grad_tr curve_inv=np.linalg.inv(curve) errs=np.diag(curve_inv) dp=np.dot(grad,curve_inv) fitp=fitp-dp*scale_fac frac_shift=dp/errs #print fitp,errs,frac_shift,np.mean(np.abs(new_grad-grad)) #print fitp,grad,frac,lamda if converged: print('converged after ',iter,' iterations') break #C=np.dot(guess,vecs) print('mean diff is ',np.mean(np.abs(C_scale-C))) #return datft,vecs,nu,C return fitp,datsqr,C def get_derivs_tod_isosrc(pars,tod,niso=None): np_src=4 np_iso=5 #nsrc=(len(pars)-np_iso)/np_src npp=len(pars) if niso is None: niso=(npp%np_src)/(np_iso-np_src) nsrc=(npp-niso*np_iso)/np_src #print nsrc,niso fitp_iso=np.zeros(np_iso) fitp_iso[:]=pars[:np_iso] #print 'fitp_iso is ',fitp_iso derivs_iso,f_iso=derivs_from_isobeta_c(fitp_iso,tod) #nn=tod.info['dat_calib'].size nn=tod.get_nsamp() derivs=np.reshape(derivs_iso,[np_iso,nn]) pred=f_iso for ii in range(nsrc): fitp_src=np.zeros(np_src) istart=np_iso+ii*np_src fitp_src[:]=pars[istart:istart+np_src] derivs_src,f_src=derivs_from_gauss_c(fitp_src,tod) pred=pred+f_src derivs_src_tmp=np.reshape(derivs_src,[np_src,nn]) derivs=np.append(derivs,derivs_src_tmp,axis=0) return derivs,pred def get_curve_deriv_tod_manygauss(pars,tod,return_vecs=False): npp=4 nsrc=len(pars)//npp fitp_gauss=np.zeros(npp) #dat=tod.info['dat_calib'] dat=tod.get_data() big_derivs=np.zeros([npp*nsrc,dat.shape[0],dat.shape[1]]) pred=0 curve=np.zeros([npp*nsrc,npp*nsrc]) deriv=np.zeros([npp*nsrc]) for i in range(nsrc): fitp_gauss[:]=pars[i*npp:(i+1)*npp] derivs,src_pred=derivs_from_gauss_c(fitp_gauss,tod) pred=pred+src_pred big_derivs[i*npp:(i+1)*npp,:,:]=derivs delt=dat-pred delt_filt=tod.apply_noise(delt) chisq=0.5*np.sum(delt[:,0]*delt_filt[:,0]) chisq=chisq+0.5*np.sum(delt[:,-1]*delt_filt[:,-1]) chisq=chisq+np.sum(delt[:,1:-1]*delt_filt[:,1:-1]) for i in range(npp*nsrc): deriv_filt=tod.apply_noise(big_derivs[i,:,:]) for j in range(i,npp*nsrc): curve[i,j]=curve[i,j]+0.5*np.sum(deriv_filt[:,0]*big_derivs[j,:,0]) curve[i,j]=curve[i,j]+0.5*np.sum(deriv_filt[:,-1]*big_derivs[j,:,-1]) curve[i,j]=curve[i,j]+np.sum(deriv_filt[:,1:-1]*big_derivs[j,:,1:-1]) curve[j,i]=curve[i,j] #print i,j,curve[i,j] deriv[i]=deriv[i]+0.5*np.sum(deriv_filt[:,0]*delt[:,0]) deriv[i]=deriv[i]+0.5*np.sum(deriv_filt[:,-1]*delt[:,-1]) deriv[i]=deriv[i]+np.sum(deriv_filt[:,1:-1]*delt[:,1:-1]) return curve,deriv,chisq def get_curve_deriv_tod_isosrc(pars,tod,return_vecs=False): np_src=4 np_iso=5 nsrc=(len(pars)-np_iso)/np_src #print 'nsrc is ',nsrc fitp_iso=np.zeros(np_iso) fitp_iso[:]=pars[:np_iso] #print 'fitp_iso is ',fitp_iso derivs_iso,f_iso=derivs_from_isobeta_c(fitp_iso,tod) derivs_iso_filt=0*derivs_iso #tmp=0*tod.info['dat_calib'] tmp=tod.get_empty(True) #nn=tod.info['dat_calib'].size nn=tod.get_nsamp for i in range(np_iso): tmp[:,:]=derivs_iso[i,:,:] derivs_iso_filt[i,:,:]=tod.apply_noise(tmp) derivs=np.reshape(derivs_iso,[np_iso,nn]) derivs_filt=np.reshape(derivs_iso_filt,[np_iso,nn]) pred=f_iso for ii in range(nsrc): fitp_src=np.zeros(np_src) istart=np_iso+ii*np_src fitp_src[:]=pars[istart:istart+np_src] #print 'fitp_src is ',fitp_src derivs_src,f_src=derivs_from_gauss_c(fitp_src,tod) pred=pred+f_src derivs_src_filt=0*derivs_src for i in range(np_src): tmp[:,:]=derivs_src[i,:,:] derivs_src_filt[i,:,:]=tod.apply_noise(tmp) derivs_src_tmp=np.reshape(derivs_src,[np_src,nn]) derivs=np.append(derivs,derivs_src_tmp,axis=0) derivs_src_tmp=np.reshape(derivs_src_filt,[np_src,nn]) derivs_filt=np.append(derivs_filt,derivs_src_tmp,axis=0) #delt_filt=tod.apply_noise(tod.info['dat_calib']-pred) delt_filt=tod.apply_noise(tod.get_data()-pred) delt_filt=np.reshape(delt_filt,nn) #dvec=np.reshape(tod.info['dat_calib'],nn) dvec=np.ravel(tod.get_data()) #predvec=np.reshape(pred,nn) predvec=np.ravel(pred) delt=dvec-predvec grad=np.dot(derivs_filt,delt) grad2=np.dot(derivs,delt_filt) curve=np.dot(derivs_filt,derivs.transpose()) #return pred if return_vecs: return grad,grad2,curve,derivs,derivs_filt,delt,delt_filt else: return grad,grad2,curve def get_timestream_chisq_from_func(func,pars,tods): chisq=0.0 for tod in tods.tods: derivs,pred=func(pars,tod) #delt=tod.info['dat_calib']-pred delt=tod.get_data()-pred delt_filt=tod.apply_noise(delt) delt_filt[:,0]=delt_filt[:,0]*0.5 delt_filt[:,-1]=delt_filt[:,-1]*0.5 chisq=chisq+np.sum(delt*delt_filt) return chisq def get_timestream_chisq_curve_deriv_from_func(func,pars,tods,rotmat=None,*args,**kwargs): chisq=0.0 grad=0.0 curve=0.0 #print 'inside func, len(tods) is ',len(tods.tods),len(pars) for tod in tods.tods: #print 'type of tod is ',type(tod) derivs,pred=func(pars,tod,*args,**kwargs) if not(rotmat is None): derivs=np.dot(rotmat.transpose(),derivs) derivs=np.reshape(derivs,[derivs.shape[0],np.product(derivs.shape[1:])]) derivs_filt=0*derivs #print('derivs_filt shape is ',derivs_filt.shape) #derivs_filt=np.reshape(derivs_filt,[derivs_filt.shape[0],np.product(derivs_filt.shape[1:])]) #sz=tod.info['dat_calib'].shape sz=tod.get_data_dims() tmp=np.zeros(sz) npp=derivs.shape[0] nn=np.product(derivs.shape[1:]) #delt=tod.info['dat_calib']-pred delt=tod.get_data()-pred delt_filt=tod.apply_noise(delt) for i in range(npp): tmp[:,:]=np.reshape(derivs[i,:],sz) tmp_filt=tod.apply_noise(tmp) #tmp_filt[:,1:-1]=tmp_filt[:,1:-1]*2 tmp_filt[:,0]=tmp_filt[:,0]*0.5 tmp_filt[:,-1]=tmp_filt[:,-1]*0.5 derivs_filt[i,:]=np.reshape(tmp_filt,nn) delt=np.reshape(delt,nn) delt_filt=np.reshape(delt_filt,nn) grad1=np.dot(derivs,delt_filt) grad2=np.dot(derivs_filt,delt) #print 'grad error is ',np.mean(np.abs((grad1-grad2)/(0.5*(np.abs(grad1)+np.abs(grad2))))) grad=grad+0.5*(grad1+grad2) curve=curve+np.dot(derivs,derivs_filt.transpose()) chisq=chisq+np.dot(delt,delt_filt) curve=0.5*(curve+curve.transpose()) if have_mpi: curve=comm.allreduce(curve) grad=comm.allreduce(grad) chisq=comm.allreduce(chisq) return chisq,grad,curve def get_ts_derivs_many_funcs(tod,pars,npar_fun,funcs,func_args=None,*args,**kwargs): #ndet=tod.info['dat_calib'].shape[0] #ndat=tod.info['dat_calib'].shape[1] ndet=tod.get_ndet() ndat=tod.get_ndata() npar=np.sum(np.asarray(npar_fun),dtype='int') #vals=np.zeros([ndet,ndat]) pred=0 derivs=np.zeros([npar,ndet,ndat]) icur=0 for i in range(len(funcs)): tmp=pars[icur:icur+npar_fun[i]].copy() myderivs,mypred=funcs[i](tmp,tod,*args,**kwargs) pred=pred+mypred derivs[icur:icur+npar_fun[i],:,:]=myderivs icur=icur+npar_fun[i] return derivs,pred #derivs,pred=funcs[i](pars,tod) def get_ts_curve_derivs_many_funcs(todvec,pars,npar_fun,funcs,driver=get_ts_derivs_many_funcs,*args,**kwargs): curve=0 grad=0 chisq=0 for tod in todvec.tods: derivs,pred=driver(tod,pars,npar_fun,funcs,*args,**kwargs) npar=derivs.shape[0] ndet=derivs.shape[1] ndat=derivs.shape[2] #pred_filt=tod.apply_noise(pred) derivs_filt=np.empty(derivs.shape) for i in range(npar): derivs_filt[i,:,:]=tod.apply_noise(derivs[i,:,:]) derivs=np.reshape(derivs,[npar,ndet*ndat]) derivs_filt=np.reshape(derivs_filt,[npar,ndet*ndat]) #delt=tod.info['dat_calib']-pred delt=tod.get_data()-pred delt_filt=tod.apply_noise(delt) chisq=chisq+np.sum(delt*delt_filt) delt=np.reshape(delt,ndet*ndat) #delt_filt=np.reshape(delt_filt,[1,ndet*ndat]) grad=grad+np.dot(derivs_filt,delt.T) #grad2=grad2+np.dot(derivs,delt_filt.T) curve=curve+np.dot(derivs_filt,derivs.T) if have_mpi: chisq=comm.allreduce(chisq) grad=comm.allreduce(grad) curve=comm.allreduce(curve) return chisq,grad,curve def update_lamda(lamda,success): if success: if lamda<0.2: return 0 else: return lamda/np.sqrt(2) else: if lamda==0.0: return 1.0 else: return 2.0*lamda def invscale(mat,do_invsafe=False): vec=1/np.sqrt(abs(np.diag(mat))) vec[np.where(vec == np.inf)[0]] = 1e-10 mm=np.outer(vec,vec) mat=mm*mat #ee,vv=np.linalg.eig(mat) #print 'rcond is ',ee.max()/ee.min(),vv[:,np.argmin(ee)] if do_invsafe: return mm*invsafe(mat) else: try: return mm*np.linalg.inv(mat) except: return mm*np.linalg.pinv(mat) def _par_step(grad,curve,to_fit,lamda,flat_priors=None,return_full=False): curve_use=curve+lamda*np.diag(np.diag(curve)) if to_fit is None: step=np.dot(invscale(curve_use,True),grad) errs=np.sqrt(np.diag(invscale(curve_use,True))) else: curve_use=curve_use[to_fit,:] curve_use=curve_use[:,to_fit] grad_use=grad[to_fit] step=np.dot(invscale(curve_use),grad_use) step_use=np.zeros(len(to_fit)) step_use[to_fit]=step errs_tmp=np.sqrt(np.diag(invscale(curve_use,True))) errs=np.zeros(len(to_fit)) errs[to_fit]=errs_tmp step=step_use #print('step shape ',step.shape,step) if return_full: return step,errs else: return step def fit_timestreams_with_derivs_manyfun(funcs,pars,npar_fun,tods,to_fit=None,to_scale=None,tol=1e-2,chitol=1e-4,maxiter=10,scale_facs=None,driver=get_ts_derivs_many_funcs, priors=None, prior_vals=None): lamda=0 t1=time.time() chisq,grad,curve=get_ts_curve_derivs_many_funcs(tods,pars,npar_fun,funcs,driver=driver) t2=time.time() if myrank==0: print('starting chisq is ',chisq,' with ',t2-t1,' seconds to get curvature') if to_fit is None: #If to_fit is not already defined, define it an intialize it to true #we're going to use it to handle not stepping for flat priors to_fit = np.ones(len(pars),dtype='bool') for iter in range(maxiter): temp_to_fit = np.copy(to_fit) #Make a copy of to fit, so we can temporarily set values to false if np.any(priors): #first build a mask that will identify parameters with flat priors flat_mask = np.where((priors == 'flat'))[0] for flat_id in flat_mask: print(pars[flat_id]) if (pars[flat_id] == prior_vals[flat_id][0]) or (pars[flat_id] == prior_vals[flat_id][1]): #Check to see if we're at the boundry values, if so don't fit for this iter temp_to_fit[flat_id] = False #Make the new step pars_new = pars + _par_step(grad, curve, temp_to_fit, lamda) #check to see if we're outside the range for the flat priors: if so, peg them print('old gamma: ', pars_new[flat_id]) for flat_id in flat_mask: if (pars_new[flat_id] < prior_vals[flat_id][0]): pars_new[flat_id] = prior_vals[flat_id][0] elif (pars_new[flat_id] > prior_vals[flat_id][1]): pars_new[flat_id] = prior_vals[flat_id][1] print('new gamma: ',pars_new[flat_id]) else: pars_new=pars+_par_step(grad,curve,to_fit,lamda) chisq_new,grad_new,curve_new=get_ts_curve_derivs_many_funcs(tods,pars_new,npar_fun,funcs,driver=driver) if chisq_new<chisq: if myrank==0: print('accepting with delta_chisq ',chisq_new-chisq,' and lamda ',lamda,pars_new.shape) print(repr(pars_new)) pars=pars_new curve=curve_new grad=grad_new lamda=update_lamda(lamda,True) if (chisq-chisq_new<chitol)&(lamda==0): step,errs=_par_step(grad,curve,temp_to_fit,lamda,return_full=True) return pars,chisq_new,curve_new,errs else: chisq=chisq_new else: if myrank==0: print('rejecting with delta_chisq ',chisq_new-chisq,' and lamda ',lamda) lamda=update_lamda(lamda,False) sys.stdout.flush() if myrank==0: print("fit_timestreams_with_derivs_manyfun failed to converge after ",maxiter," iterations.") step,errs=_par_step(grad,curve,temp_to_fit,lamda,return_full=True) return pars,chisq,curve,errs def fit_timestreams_with_derivs(func,pars,tods,to_fit=None,to_scale=None,tol=1e-2,chitol=1e-4,maxiter=10,scale_facs=None): if not(to_fit is None): #print 'working on creating rotmat' to_fit=np.asarray(to_fit,dtype='int64') inds=np.unique(to_fit) nfloat=np.sum(to_fit==1) ncovary=np.sum(inds>1) nfit=nfloat+ncovary rotmat=np.zeros([len(pars),nfit]) solo_inds=np.where(to_fit==1)[0] icur=0 for ind in solo_inds: rotmat[ind,icur]=1.0 icur=icur+1 if ncovary>0: group_inds=inds[inds>1] for ind in group_inds: ii=np.where(to_fit==ind)[0] rotmat[ii,icur]=1.0 icur=icur+1 else: rotmat=None iter=0 converged=False pp=pars.copy() lamda=0.0 chi_ref,grad,curve=get_timestream_chisq_curve_deriv_from_func(func,pp,tods,rotmat) chi_cur=chi_ref iter=0 while (converged==False) and (iter<maxiter): iter=iter+1 curve_tmp=curve+lamda*np.diag(np.diag(curve)) #curve_inv=np.linalg.inv(curve_tmp) curve_inv=invscale(curve_tmp) shifts=np.dot(curve_inv,grad) if not(rotmat is None): shifts_use=np.dot(rotmat,shifts) else: shifts_use=shifts pp_tmp=pp+shifts_use chi_new=get_timestream_chisq_from_func(func,pp_tmp,tods) if chi_new<=chi_cur+chitol: #add in a bit of extra tolerance in chi^2 in case we're bopping about the minimum success=True else: success=False if success: pp=pp_tmp chi_cur=chi_new chi_tmp,grad,curve=get_timestream_chisq_curve_deriv_from_func(func,pp,tods,rotmat) lamda=update_lamda(lamda,success) if (lamda==0)&success: errs=np.sqrt(np.diag(curve_inv)) conv_fac=np.max(np.abs(shifts/errs)) if (conv_fac<tol): print('we have converged') converged=True else: conv_fac=None to_print=np.asarray([3600*180.0/np.pi,3600*180.0/np.pi,3600*180.0/np.pi,1.0,1.0,3600*180.0/np.pi,3600*180.0/np.pi,3600*180.0/np.pi*np.sqrt(8*np.log(2)),1.0])*(pp-pars) print('iter',iter,' max_shift is ',conv_fac,' with lamda ',lamda,chi_ref-chi_cur,chi_ref-chi_new) return pp,chi_cur def split_dict(mydict,vec,thresh): #split a dictionary into sub-dictionaries wherever a gap in vec is larger than thresh. #useful for e.g. splitting TODs where there's a large time gap due to cuts. inds=np.where(np.diff(vec)>thresh)[0] #print(inds,len(inds)) if len(inds)==0: return [mydict] ndict=len(inds)+1 inds=np.hstack([[0],inds+1,[len(vec)]]) #print(inds) out=[None]*ndict for i in range(ndict): out[i]={} for key in mydict.keys(): tmp=mydict[key] for i in range(ndict): out[i][key]=tmp try: dims=tmp.shape ndim=len(dims) if ndim==1: if dims[0]==len(vec): for i in range(ndict): out[i][key]=tmp[inds[i]:inds[i+1]].copy() if ndim==2: if dims[1]==len(vec): for i in range(ndict): out[i][key]=tmp[:,inds[i]:inds[i+1]].copy() elif dims[0]==len(vec): for i in range(ndict): out[i][key]=tmp[inds[i]:inds[i+1],:].copy() except: continue #print('copying ',key,' unchanged') #don't need below as it's already copied by default #for i in range(ndict): # out[i][key]=mydict[key] return out def mask_dict(mydict,mask): for key in mydict.keys(): tmp=mydict[key] try: dims=tmp.shape ndim=len(dims) if ndim==1: if dims[0]==len(mask): tmp=tmp[mask] mydict[key]=tmp if ndim==2: if dims[0]==len(mask): tmp=tmp[mask,:] if dims[1]==len(mask): tmp=tmp[:,mask] mydict[key]=tmp if ndim==3: if dims[0]==len(mask): tmp=tmp[mask,:,:] if dims[1]==len(mask): tmp=tmp[:,mask,:] if dims[2]==len(maks): tmp=tmp[:,:,mask] mydict[key]=tmp except: continue
187,717
34.674268
578
py
minkasi
minkasi-master/minkasi/mkfftw.py
import os import numpy import ctypes import time try: mylib=ctypes.cdll.LoadLibrary("libmkfftw.so") except OSError: mylib=ctypes.cdll.LoadLibrary(os.path.join(os.path.dirname(os.path.abspath(__file__)), "libmkfftw.so")) many_fft_r2c_1d_c=mylib.many_fft_r2c_1d many_fft_r2c_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int,ctypes.c_int] many_fftf_r2c_1d_c=mylib.many_fftf_r2c_1d many_fftf_r2c_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int,ctypes.c_int] many_fft_c2r_1d_c=mylib.many_fft_c2r_1d many_fft_c2r_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int,ctypes.c_int] many_fftf_c2r_1d_c=mylib.many_fftf_c2r_1d many_fftf_c2r_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int,ctypes.c_int] fft_r2r_1d_c=mylib.fft_r2r_1d fft_r2r_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int] many_fft_r2r_1d_c=mylib.many_fft_r2r_1d many_fft_r2r_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int] many_fftf_r2r_1d_c=mylib.many_fftf_r2r_1d many_fftf_r2r_1d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_int,ctypes.c_int] fft_r2c_n_c=mylib.fft_r2c_n fft_r2c_n_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_void_p] fft_c2r_n_c=mylib.fft_c2r_n fft_c2r_n_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_int,ctypes.c_void_p] fft_r2c_3d_c=mylib.fft_r2c_3d fft_r2c_3d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p] fft_c2r_3d_c=mylib.fft_c2r_3d fft_c2r_3d_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p,ctypes.c_void_p] set_threaded_c=mylib.set_threaded set_threaded_c.argtypes=[ctypes.c_int] read_wisdom_c=mylib.read_wisdom read_wisdom_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p] write_wisdom_c=mylib.write_wisdom write_wisdom_c.argtypes=[ctypes.c_void_p,ctypes.c_void_p] def set_threaded(n=-1): set_threaded_c(n) def rfftn(dat): myshape=dat.shape myshape=numpy.asarray(myshape,dtype='int32') myshape2=myshape.copy() myshape2[-1]=(myshape2[-1]//2+1) datft=numpy.zeros(myshape2,dtype='complex') fft_r2c_n_c(dat.ctypes.data,datft.ctypes.data,len(myshape),myshape.ctypes.data) return datft def irfftn(datft,iseven=True,preserve_input=True): #the c2r transforms destroy input. if you want to keep the input #around, then we need to copy the incoming data. if preserve_input: datft=datft.copy() myshape=datft.shape myshape=numpy.asarray(myshape,dtype='int32') myshape2=myshape.copy() if iseven: myshape2[-1]=2*(myshape2[-1]-1) else: myshape2[-1]=2*myshape2[-1]-1 #print(myshape2) dat=numpy.empty(myshape2,dtype='float64') fft_c2r_n_c(datft.ctypes.data,dat.ctypes.data,len(myshape2),myshape2.ctypes.data) return dat def fft_r2c_3d(dat): myshape=dat.shape assert(len(myshape)==3) myshape=numpy.asarray(myshape,dtype='int') myshape2=myshape.copy() myshape2[-1]=(myshape2[-1]//2+1) datft=numpy.zeros(myshape2,dtype='complex') fft_r2c_3d_c(dat.ctypes.data,datft.ctypes.data,myshape.ctypes.data) return datft def fft_c2r_3d(datft,iseven=True,preserve_input=True): #the c2r transforms destroy input. if you want to keep the input #around, then we need to copy the incoming data. if preserve_input: datft=datft.copy() myshape=datft.shape assert(len(myshape)==3) myshape=numpy.asarray(myshape,dtype='int') myshape2=myshape.copy() if iseven: myshape2[-1]=2*(myshape2[-1]-1) else: myshape2[-1]=2*myshape2[-1]-1 #print(myshape2) dat=numpy.empty(myshape2,dtype='float64') fft_c2r_3d_c(datft.ctypes.data,dat.ctypes.data,myshape2.ctypes.data) return dat def fft_r2c(dat): ndat=dat.shape[1] ntrans=dat.shape[0] if dat.dtype==numpy.dtype('float64'): #datft=numpy.zeros(dat.shape,dtype=complex) datft=numpy.empty(dat.shape,dtype=complex) many_fft_r2c_1d_c(dat.ctypes.data,datft.ctypes.data,ntrans,ndat,ndat,ndat) else: assert(dat.dtype==numpy.dtype('float32')) datft=numpy.empty(dat.shape,dtype='complex64') many_fftf_r2c_1d_c(dat.ctypes.data,datft.ctypes.data,ntrans,ndat,ndat,ndat) return datft def fft_c2r(datft): ndat=datft.shape[1] ntrans=datft.shape[0] if datft.dtype==numpy.dtype('complex128'): dat=numpy.zeros(datft.shape) many_fft_c2r_1d_c(datft.ctypes.data,dat.ctypes.data,ntrans,ndat,ndat,ndat) dat=dat/ndat else: assert(datft.dtype==numpy.dtype('complex64')) dat=numpy.zeros(datft.shape,dtype='float32') many_fftf_c2r_1d_c(datft.ctypes.data,dat.ctypes.data,ntrans,ndat,ndat,ndat) dat=dat/numpy.float32(ndat) return dat def fft_r2r_1d(dat,kind=1): nn=dat.size trans=numpy.zeros(nn) fft_r2r_1d_c(dat.ctypes.data,trans.ctypes.data,nn,kind) return trans def fft_r2r(dat,trans=None,kind=1): if len(dat.shape)==1: return fft_r2r_1d(dat,kind) ntrans=dat.shape[0] n=dat.shape[1] #trans=numpy.zeros([ntrans,n],dtype=type(dat[0,0])) if trans is None: trans=numpy.empty([ntrans,n],dtype=type(dat[0,0])) if type(dat[0,0])==numpy.dtype('float32'): #print 'first two element in python are ',dat[0,0],dat[0,1] many_fftf_r2r_1d_c(dat.ctypes.data,trans.ctypes.data,n,kind,ntrans) else: many_fft_r2r_1d_c(dat.ctypes.data,trans.ctypes.data,n,kind,ntrans) return trans def read_wisdom(double_file='.fftw_wisdom',single_file='.fftwf_wisdom'): df=numpy.zeros(len(double_file)+1,dtype='int8') df[0:-1]=[ord(c) for c in double_file] sf=numpy.zeros(len(single_file)+1,dtype='int8') sf[0:-1]=[ord(c) for c in single_file] read_wisdom_c(df.ctypes.data,sf.ctypes.data) def write_wisdom(double_file='.fftw_wisdom',single_file='.fftwf_wisdom'): df=numpy.zeros(len(double_file)+1,dtype='int8') df[0:-1]=[ord(c) for c in double_file] sf=numpy.zeros(len(single_file)+1,dtype='int8') sf[0:-1]=[ord(c) for c in single_file] write_wisdom_c(df.ctypes.data,sf.ctypes.data)
6,217
31.385417
113
py
minkasi
minkasi-master/minkasi/pyregion_tools.py
import pyregion from astropy.io import fits from astropy.wcs import WCS import numpy as np import copy __all__ = ["region_binner", "bootstrap"] def bootstrap(data, n = 10000): """ Bootstraps data. Given an input data vector, the bootstrapping proceedure selects a random subsample of data, with replacement, and computes a statistic on that subsample. TODO: currently only mean is supported, should add others. The variance of the resulting statistics is a good measure of the true variance of that statistic. In other words, if we have some data, and we compute mean(data) and want to know what var(data) is, bootstrapping is a good way to do that. Parameters ---------- data : numpy.array array of data which we wish to bootstrap n : int, optional number of instances of bootstrapping to perform Returns ------- stats : numpy.array array of average of the bootstrapped samples """ stats = np.zeros(n) data = np.array(data) for i in range(n): flags = np.random.randint(len(data), size = len(data)) stats[i] = np.mean(data[flags]) return stats def region_binner(shapeList, hdu, plot = False, return_rs = True): """ Helper function for making binned profiles from pyregion and map. This function computes the binned mean and variance in regions of a map as specified by a pyregion ShapeList. This function handles the various different types of pyregions and sepcifications, i.e. specifying multiple regions explicitly or using the pyregion defualt radial or angular bins. Parameters ---------- shapeList : 'pyregion.core.ShapeList' pyregion ShapeList specifying the regions overwhich to compute the profile bins. hdu : 'astropy.io.fits.hdu.hdulist.HDUList' astropy HDUList containing the relevant map at index 0. Must contain the map data at hdu[0].data plot : Bool, optional specifies whether to plot the bin regions Returns ------- means : np.array(float) simple average of data within each binning region var : np.array(float) variance within the same regions as computed by bootstrapping the pixels """ if len(shapeList) > 1: #Handles multiple explicitly defined regions of any type. Since they are explicitly defined one function can handle all return _region_binner_explicit_regions(shapeList, hdu, plot = plot, return_rs = return_rs) elif shapeList[0].name == 'epanda': #Handles epanda ShapeLists where n radial bins has been specified return _region_binner_epanda_rbins(shapeList, hdu, plot = plot) else: print('Error: region type {} not currently supported'.format(shapeList[0].name)) return def _region_binner_explicit_regions(shapeList, hdu, plot = False, return_rs = True): """ Anulizing function for shapeList where each region has been explicilty defined. This function takes a list of regions and compute the average and variance within each of those regions. It assumes that each region is explicitly defined, meaning there are no radial bins, angular bins, etc. within a region. Further this function assumes that you know what youre doing. For example you can hand it a list of overlapping regions and it will return their means/vars without warning. Currently only supported method for computing variance is bootstrapping of pixels but TODO add more methods. Since the regions are all explicitly defined we don't need to compute any bins/regions ourselves and this function can handle all types of regions. Parameters ---------- shapeList : 'pyregion.core.ShapeList' pyregion ShapeList specifying the regions overwhich to compute the profile bins. hdu : 'astropy.io.fits.hdu.hdulist.HDUList' astropy HDUList containing the relevant map at index 0. Must contain the map data at hdu[0].data plot : Bool, optional specifies whether to plot the bin regions return_rs : Bool, optional specifies whether to generate and return radii associated with means/vars Returns ------- means : np.array(float) simple average of data within each binning region var : np.array(float) variance within the same regions as computed by bootstrapping the pixels rs : np.array(float) region radii associated with means/var """ means = np.zeros(len(shapeList)) var = np.zeros(len(shapeList)) filters = shapeList.get_filter() for i in range(len(var)): cur_filter = filters[i] tempmask = cur_filter.mask(hdu[0].data.shape) #Sets values in the map outside the mask to zero. This may be cleaner with masked arrays masked_data = hdu[0].data*tempmask if plot: plt.imshow(masked_data,origin = 'lower') plt.show() plt.close() #Since the masked data points are still in the map just set to zero, we have to extract those with non-zero value #to compute the mean/var. This is the part that would be cleaner with masked arrays. vals = masked_data[np.abs(masked_data)>1e-10] means[i] = np.mean(vals) var[i] = np.var(bootstrap(vals)) if return_rs: rs = np.zeros(len(means)) header = hdu[0].header cdelt = header.get('CDELT2', 1.0) for i in range(len(rs)): if shapeList[i].name == 'epanda': #If region is epanda, return the average of r_semimajor at the endpoints of the region rs[i] = np.mean([shapeList[i].coord_list[5], shapeList[i].coord_list[7]])*cdelt*3600 elif shapeList[i].name == 'panda': #If region is panda, return average of r at endpoints of region rs[i] = np.mean(shapeList[i].coord_list[4:5])*cdelt*3600 else: print("Error: region type {} not supported for automatic radii generation".format(shapeList[i].name)) return means, var return means, var, rs return means, var def _region_binner_epanda_rbins(shapeList, hdu, plot=False, return_rs = True): # 0 1 2 3 4 5 6 7 8 9 10 #Epanda has format (ra, dec, ang_start, ang_end, n_ang_bins, inner a, Inner b, outer a, outer b, number of radial bins, PA) coords = shapeList[0].coord_list step_a = (coords[7]-coords[5])/coords[-2] step_b = (coords[8]-coords[6])/coords[-2] #N bins located at second to last means = np.zeros(coords[-2]) var = np.zeros(coords[-2]) if return_rs: rs = np.zeros(coords[-2]) for i in range(coords[-2]): temp_r = copy.deepcopy(shapeList) #note that the inner_a and outer_a in the coord_list are for the region as a whole inner_a = coords[5] + i*step_a outer_a = coords[5] + (i)*step_a+step_a inner_b = coords[6] + i*step_b outer_b = coords[6] + (i)*step_b+step_b temp_r[0].coord_list[5] = inner_a temp_r[0].coord_list[7] = outer_a temp_r[0].coord_list[6] = inner_b temp_r[0].coord_list[8] = outer_b temp_r[0].coord_list[-2] = 1 tempfilter = temp_r.get_filter() tempmask = tempfilter.mask(hdu[0].data.shape) if plot: plt.imshow(hdu[0].data*tempmask,origin = 'lower') plt.show() plt.close() masked_data = hdu[0].data*tempmask vals = masked_data[np.abs(masked_data)>1e-8] means[i] = np.mean(vals) var[i] = np.var(bootstrap(vals)) if return_rs: rs[i] = inner_a + i*step_a/2 if return_rs: return means, var, rs return means, var
7,945
36.658768
140
py
minkasi
minkasi-master/minkasi/__init__.py
from .minkasi import *
23
11
22
py
minkasi
minkasi-master/minkasi/zernike.py
import numpy def zernike_column(m,nmax,rmat): """Generate the radial part of zernike polynomials for all n from m up to nmax""" if ((m-nmax)%2!=0): #print 'm an n must have same parity' #return None #if parity is wrong, then drop nmax by one. makes external loop to generate all zns much simpler nmax=nmax-1 if (m>nmax): print 'm may not be larger than n' return None nm=(nmax-m)/2+1 mask=rmat>1 zn=[None]*nm nn=numpy.zeros(nm,dtype='int') zn[0]=rmat**m zn[0][mask]=0 nn[0]=m if nm==1: return zn,nn rsqr=rmat**2 zn[1]=((m+2)*rsqr-m-1)*zn[0] zn[1][mask]=0 nn[1]=m+2 if nm==2: return zn,nn ii=2 for n in range(m+4,nmax+1,2): f1=2*(n-1)*(2*n*(n-2)*rsqr-m*m-n*(n-2))*zn[ii-1] f2=n*(n+m-2)*(n-m-2)*zn[ii-2] f3=1.0/((n+m)*(n-m)*(n-2)) zn[ii]=(f1-f2)*f3 nn[ii]=n ii=ii+1 return zn,nn def all_zernike(n,r,th): znvec=[None]*(n+1) nvec=[None]*(n+1) mvec=[None]*(n+1) nzer=0 for m in range(0,n+1): znvec[m],nvec[m]=zernike_column(m,n,r) mvec[m]=0*nvec[m]+m if m==0: nzer=nzer+len(znvec[m]) else: nzer=nzer+2*len(znvec[m]) #print nzer shp=r.shape shp=numpy.append(nzer,shp) zns=numpy.zeros(shp) #print shp icur=0 #print n,len(znvec) for m in range(0,n+1): #print icur ss=numpy.sin(m*th) cc=numpy.cos(m*th) zz=znvec[m] nn=len(zz) if m==0: for i in range(nn): zns[icur,:]=zz[i] icur=icur+1 else: for i in range(nn): zns[icur,:]=zz[i]*cc icur=icur+1 zns[icur,:]=zz[i]*ss icur=icur+1 return zns,znvec
1,894
20.534091
105
py
tbsm
tbsm-main/tbsm_pytorch.py
# Copyright (c) Facebook, Inc. and its affiliates. # # This source code is licensed under the MIT license found in the # LICENSE file in the root directory of this source tree. ### import packages ### from __future__ import absolute_import, division, print_function, unicode_literals # miscellaneous import time import os from os import path import random # numpy and scikit-learn import numpy as np from sklearn.metrics import roc_auc_score # pytorch import torch import torch.nn as nn import torch.nn.functional as Functional from torch.nn.parameter import Parameter from torch.utils.tensorboard import SummaryWriter # tbsm data import tbsm_data_pytorch as tp # set python, numpy and torch random seeds def set_seed(seed, use_gpu): random.seed(seed) os.environ['PYTHONHASHSEED'] = str(seed) np.random.seed(seed) if use_gpu: torch.manual_seed(seed) torch.cuda.manual_seed(seed) torch.cuda.manual_seed_all(seed) # if using multi-GPU. torch.backends.cudnn.benchmark = False torch.backends.cudnn.deterministic = True ### define time series layer (TSL) ### class TSL_Net(nn.Module): def __init__( self, arch_interaction_op='dot', arch_attention_mechanism='mlp', ln=None, model_type="tsl", tsl_inner="def", mha_num_heads=8, ln_top="" ): super(TSL_Net, self).__init__() # save arguments self.arch_interaction_op = arch_interaction_op self.arch_attention_mechanism = arch_attention_mechanism self.model_type = model_type self.tsl_inner = tsl_inner # setup for mechanism type if self.arch_attention_mechanism == 'mlp': self.mlp = dlrm.DLRM_Net().create_mlp(ln, len(ln) - 2) # setup extra parameters for some of the models if self.model_type == "tsl" and self.tsl_inner in ["def", "ind"]: m = ln_top[-1] # dim of dlrm output mean = 0.0 std_dev = np.sqrt(2 / (m + m)) W = np.random.normal(mean, std_dev, size=(1, m, m)).astype(np.float32) self.A = Parameter(torch.tensor(W), requires_grad=True) elif self.model_type == "mha": m = ln_top[-1] # dlrm output dim self.nheads = mha_num_heads self.emb_m = self.nheads * m # mha emb dim mean = 0.0 std_dev = np.sqrt(2 / (m + m)) # np.sqrt(1 / m) # np.sqrt(1 / n) qm = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.Q = Parameter(torch.tensor(qm), requires_grad=True) km = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.K = Parameter(torch.tensor(km), requires_grad=True) vm = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.V = Parameter(torch.tensor(vm), requires_grad=True) def forward(self, x=None, H=None): # adjust input shape (batchSize, vector_dim) = x.shape x = torch.reshape(x, (batchSize, 1, -1)) x = torch.transpose(x, 1, 2) # debug prints # print("shapes: ", self.A.shape, x.shape) # perform mode operation if self.model_type == "tsl": if self.tsl_inner == "def": ax = torch.matmul(self.A, x) x = torch.matmul(self.A.permute(0, 2, 1), ax) # debug prints # print("shapes: ", H.shape, ax.shape, x.shape) elif self.tsl_inner == "ind": x = torch.matmul(self.A, x) # perform interaction operation if self.arch_interaction_op == 'dot': if self.arch_attention_mechanism == 'mul': # coefficients a = torch.transpose(torch.bmm(H, x), 1, 2) # context c = torch.bmm(a, H) elif self.arch_attention_mechanism == 'mlp': # coefficients a = torch.transpose(torch.bmm(H, x), 1, 2) # MLP first/last layer dims are automatically adjusted to ts_length y = dlrm.DLRM_Net().apply_mlp(a, self.mlp) # context, y = mlp(a) c = torch.bmm(torch.reshape(y, (batchSize, 1, -1)), H) else: sys.exit('ERROR: --arch-attention-mechanism=' + self.arch_attention_mechanism + ' is not supported') else: sys.exit('ERROR: --arch-interaction-op=' + self.arch_interaction_op + ' is not supported') elif self.model_type == "mha": x = torch.transpose(x, 1, 2) Qx = torch.transpose(torch.matmul(x, self.Q), 0, 1) HK = torch.transpose(torch.matmul(H, self.K), 0, 1) HV = torch.transpose(torch.matmul(H, self.V), 0, 1) # multi-head attention (mha) multihead_attn = nn.MultiheadAttention(self.emb_m, self.nheads).to(x.device) attn_output, _ = multihead_attn(Qx, HK, HV) # context c = torch.squeeze(attn_output, dim=0) # debug prints # print("shapes:", c.shape, Qx.shape) return c ### define Time-based Sequence Model (TBSM) ### class TBSM_Net(nn.Module): def __init__( self, m_spa, ln_emb, ln_bot, ln_top, arch_interaction_op, arch_interaction_itself, ln_mlp, ln_tsl, tsl_interaction_op, tsl_mechanism, ts_length, ndevices=-1, model_type="", tsl_seq=False, tsl_proj=True, tsl_inner="def", tsl_num_heads=1, mha_num_heads=8, rnn_num_layers=5, debug_mode=False, ): super(TBSM_Net, self).__init__() # save arguments self.ndevices = ndevices self.debug_mode = debug_mode self.ln_bot = ln_bot self.ln_top = ln_top self.ln_tsl = ln_tsl self.ts_length = ts_length self.tsl_interaction_op = tsl_interaction_op self.tsl_mechanism = tsl_mechanism self.model_type = model_type self.tsl_seq = tsl_seq self.tsl_proj = tsl_proj self.tsl_inner = tsl_inner self.tsl_num_heads = tsl_num_heads self.mha_num_heads = mha_num_heads self.rnn_num_layers = rnn_num_layers self.ams = nn.ModuleList() self.mlps = nn.ModuleList() if self.model_type == "tsl": self.num_mlps = int(self.tsl_num_heads) # number of tsl components else: self.num_mlps = 1 #debug prints if self.debug_mode: print(self.model_type) print(ln_bot) print(ln_top) print(ln_emb) # embedding layer (implemented through dlrm tower, without last layer sigmoid) if "qr" in model_type: self.dlrm = dlrm.DLRM_Net( m_spa, ln_emb, ln_bot, ln_top, arch_interaction_op, arch_interaction_itself, qr_flag=True, qr_operation="add", qr_collisions=4, qr_threshold=100000 ) print("Using QR embedding method.") else: self.dlrm = dlrm.DLRM_Net( m_spa, ln_emb, ln_bot, ln_top, arch_interaction_op, arch_interaction_itself, ) # prepare data needed for tsl layer construction if self.model_type == "tsl": if not self.tsl_seq: self.ts_array = [self.ts_length] * self.num_mlps else: self.ts_array = [] m = self.ts_length / self.tsl_num_heads for j in range(self.tsl_num_heads, 0, -1): t = min(self.ts_length, round(m * j)) self.ts_array.append(t) elif self.model_type == "mha": self.ts_array = [self.ts_length] else: self.ts_array = [] # construction of one or more tsl components for ts in self.ts_array: ln_tsl = np.concatenate((np.array([ts]), self.ln_tsl)) ln_tsl = np.append(ln_tsl, ts) # create tsl mechanism am = TSL_Net( arch_interaction_op=self.tsl_interaction_op, arch_attention_mechanism=self.tsl_mechanism, ln=ln_tsl, model_type=self.model_type, tsl_inner=self.tsl_inner, mha_num_heads=self.mha_num_heads, ln_top=self.ln_top, ) self.ams.append(am) # tsl MLPs (with sigmoid on last layer) for _ in range(self.num_mlps): mlp_tsl = dlrm.DLRM_Net().create_mlp(ln_mlp, ln_mlp.size - 2) self.mlps.append(mlp_tsl) # top mlp if needed if self.num_mlps > 1: f_mlp = np.array([self.num_mlps, self.num_mlps + 4, 1]) self.final_mlp = dlrm.DLRM_Net().create_mlp(f_mlp, f_mlp.size - 2) # Offsets need to be stored beforehand if args.run_fast. if args.run_fast: # Constant offsets tensor - resize if needed. self.max_offset = 10000000 self.offsets = torch.tensor(list(range(self.max_offset))) self.offsets_moved = False def forward(self, x, lS_o, lS_i): # Move offsets to device if needed and not already done. if args.run_fast and not self.offsets_moved: self.offsets = self.offsets.to(x[0].device) self.offsets_moved = True # data point is history H and last entry w n = x[0].shape[0] # batch_size ts = len(x) H = torch.zeros(n, self.ts_length, self.ln_top[-1]).to(x[0].device) # Compute H using either fast or original approach depending on args.run_fast. if args.run_fast: # j determines access indices of input; first, determine j bounds and get all inputs. j_lower = (ts - self.ts_length - 1) j_upper = (ts - 1) # Concatenate x[j]s using j bounds. concatenated_x = torch.cat(x[j_lower : j_upper]) # Set offsets and increase size if needed. curr_max_offset = (j_upper - j_lower) * n if curr_max_offset > self.max_offset + 1: # Resize offsets to 2x required size. self.offsets = torch.tensor(list(range(curr_max_offset * 2))).to(self.offsets.device) self.max_offset = curr_max_offset * 2 concatenated_lS_o = [self.offsets[: curr_max_offset] for j in range(len(lS_o[0]))] # Concatenate lS_i[0, 1, 2]s. concatenated_lS_i = [torch.cat([lS_i[i][j] for i in range(j_lower, j_upper)]) for j in range(len(lS_i[0]))] # oj determines access indices of output; determine oj bounds to assign output values in H. oj is just j indices adjusted to start at 0. oj_lower = 0 - (ts - self.ts_length - 1) oj_upper = (ts - 1) - (ts - self.ts_length - 1) # After fetching all inputs, run through DLRM. concatenated_dlrm_output = self.dlrm(concatenated_x, concatenated_lS_o, concatenated_lS_i) # Reshape output with new TS dimension and transpose to get H output. transposed_concatenated_dlrm_output = torch.transpose(concatenated_dlrm_output.reshape((j_upper - j_lower), n, self.ln_top[-1]), 0, 1) if self.model_type == "tsl" and self.tsl_proj: dlrm_output = Functional.normalize(transposed_concatenated_dlrm_output, p=2, dim=2) else: dlrm_output = transposed_concatenated_dlrm_output # Assign the output to H with correct output bounds. H[:, oj_lower : oj_upper, :] = dlrm_output else: # split point into first part (history) # and last item for j in range(ts - self.ts_length - 1, ts - 1): oj = j - (ts - self.ts_length - 1) v = self.dlrm(x[j], lS_o[j], lS_i[j]) if self.model_type == "tsl" and self.tsl_proj: v = Functional.normalize(v, p=2, dim=1) H[:, oj, :] = v w = self.dlrm(x[-1], lS_o[-1], lS_i[-1]) # project onto sphere if self.model_type == "tsl" and self.tsl_proj: w = Functional.normalize(w, p=2, dim=1) # print("data: ", x[-1], lS_o[-1], lS_i[-1]) (mini_batch_size, _) = w.shape # for cases when model is tsl or mha if self.model_type != "rnn": # create MLP for each TSL component # each ams[] element is one component for j in range(self.num_mlps): ts = self.ts_length - self.ts_array[j] c = self.ams[j](w, H[:, ts:, :]) c = torch.reshape(c, (mini_batch_size, -1)) # concat context and w z = torch.cat([c, w], dim=1) # obtain probability of a click as a result of MLP p = dlrm.DLRM_Net().apply_mlp(z, self.mlps[j]) if j == 0: ps = p else: ps = torch.cat((ps, p), dim=1) if ps.shape[1] > 1: p_out = dlrm.DLRM_Net().apply_mlp(ps, self.final_mlp) else: p_out = ps # RNN based on LSTM cells case, context is final hidden state else: hidden_dim = w.shape[1] # equal to dim(w) = dim(c) level = self.rnn_num_layers # num stacks of rnns Ht = H.permute(1, 0, 2) rnn = nn.LSTM(int(self.ln_top[-1]), int(hidden_dim), int(level)).to(x[0].device) h0 = torch.randn(level, n, hidden_dim).to(x[0].device) c0 = torch.randn(level, n, hidden_dim).to(x[0].device) output, (hn, cn) = rnn(Ht, (h0, c0)) hn, cn = torch.squeeze(hn[level - 1, :, :]), \ torch.squeeze(cn[level - 1, :, :]) if self.debug_mode: print(w.shape, output.shape, hn.shape) # concat context and w z = torch.cat([hn, w], dim=1) p_out = dlrm.DLRM_Net().apply_mlp(z, self.mlps[0]) return p_out # construct tbsm model or read it from the file specified # by args.save_model def get_tbsm(args, use_gpu): # train, test, or train-test modes = args.mode.split("-") model_file = args.save_model if args.debug_mode: print("model_file: ", model_file) print("model_type: ", args.model_type) if use_gpu: ngpus = torch.cuda.device_count() # 1 devicenum = "cuda:" + str(args.device_num % ngpus) print("device:", devicenum) device = torch.device(devicenum) print("Using {} GPU(s)...".format(ngpus)) else: device = torch.device("cpu") print("Using CPU...") # prepare dlrm arch m_spa = args.arch_sparse_feature_size # this is an array of sizes of cat features ln_emb = np.fromstring(args.arch_embedding_size, dtype=int, sep="-") num_fea = ln_emb.size + 1 # user: num sparse + bot_mlp(all dense) ln_bot = np.fromstring(args.arch_mlp_bot, dtype=int, sep="-") # m_den = ln_bot[0] ln_bot[ln_bot.size - 1] = m_spa # enforcing m_den_out = ln_bot[ln_bot.size - 1] # must be == m_spa (embed dim) if args.arch_interaction_op == "dot": # approach 1: all # num_int = num_fea * num_fea + m_den_out # approach 2: unique if args.arch_interaction_itself: num_int = (num_fea * (num_fea + 1)) // 2 + m_den_out else: num_int = (num_fea * (num_fea - 1)) // 2 + m_den_out elif args.arch_interaction_op == "cat": num_int = num_fea * m_den_out else: sys.exit( "ERROR: --arch-interaction-op=" + args.arch_interaction_op + " is not supported" ) arch_mlp_top_adjusted = str(num_int) + "-" + args.arch_mlp_top ln_top = np.fromstring(arch_mlp_top_adjusted, dtype=int, sep="-") # sigmoid_top = len(ln_top) - 2 # used only if length_ts == 1 # attention mlp (will be automatically adjusted so that first and last # layer correspond to number of vectors (ts_length) used in attention) ln_atn = np.fromstring(args.tsl_mlp, dtype=int, sep="-") # context MLP (with automatically adjusted first layer) if args.model_type == "mha": num_cat = (int(args.mha_num_heads) + 1) * ln_top[-1] # mha with k heads + w else: # tsl or rnn num_cat = 2 * ln_top[-1] # [c,w] arch_mlp_adjusted = str(num_cat) + "-" + args.arch_mlp ln_mlp = np.fromstring(arch_mlp_adjusted, dtype=int, sep="-") # construct TBSM tbsm = TBSM_Net( m_spa, ln_emb, ln_bot, ln_top, args.arch_interaction_op, args.arch_interaction_itself, ln_mlp, ln_atn, args.tsl_interaction_op, args.tsl_mechanism, args.ts_length, -1, args.model_type, args.tsl_seq, args.tsl_proj, args.tsl_inner, args.tsl_num_heads, args.mha_num_heads, args.rnn_num_layers, args.debug_mode, ) # move model to gpu if use_gpu: tbsm = tbsm.to(device) # .cuda() # load existing pre-trained model if needed if path.exists(model_file): if modes[0] == "test" or (len(modes) > 1 and modes[1] == "test"): if use_gpu: ld_model = torch.load( model_file, map_location=torch.device('cuda') ) else: # when targeting inference on CPU ld_model = torch.load(model_file, map_location=torch.device('cpu')) tbsm.load_state_dict(ld_model['model_state_dict']) return tbsm, device def data_wrap(X, lS_o, lS_i, use_gpu, device): if use_gpu: # .cuda() return ([xj.to(device) for xj in X], [[S_o.to(device) for S_o in row] for row in lS_o], [[S_i.to(device) for S_i in row] for row in lS_i]) else: return X, lS_o, lS_i def time_wrap(use_gpu, device): if use_gpu: torch.cuda.synchronize(device) return time.time() def loss_fn_wrap(Z, T, use_gpu, device): if use_gpu: return loss_fn(Z, T.to(device)) else: return loss_fn(Z, T) loss_fn = torch.nn.BCELoss(reduction="mean") # iterate through validation data, which can be used to determine the best seed and # during main training for deciding to save the current model def iterate_val_data(val_ld, tbsm, use_gpu, device): # NOTE: call to tbsm.eval() not needed here, see # https://discuss.pytorch.org/t/model-eval-vs-with-torch-no-grad/19615 total_loss_val = 0 total_accu_test = 0 total_samp_test = 0 for _, (X, lS_o, lS_i, T_test) in enumerate(val_ld): batchSize = X[0].shape[0] Z_test = tbsm(*data_wrap(X, lS_o, lS_i, use_gpu, device )) # # compute loss and accuracy z = Z_test.detach().cpu().numpy() # numpy array t = T_test.detach().cpu().numpy() # numpy array A_test = np.sum((np.round(z, 0) == t).astype(np.uint8)) total_accu_test += A_test total_samp_test += batchSize E_test = loss_fn_wrap(Z_test, T_test, use_gpu, device) L_test = E_test.detach().cpu().numpy() # numpy array total_loss_val += (L_test * batchSize) return total_accu_test, total_samp_test, total_loss_val # iterate through training data, which is called once every epoch. It updates weights, # computes loss, accuracy, saves model if needed and calls iterate_val_data() function. # isMainTraining is True for main training and False for fast seed selection def iterate_train_data(args, train_ld, val_ld, tbsm, k, use_gpu, device, writer, losses, accuracies, isMainTraining): # select number of batches if isMainTraining: nbatches = len(train_ld) if args.num_batches == 0 else args.num_batches else: nbatches = len(train_ld) # specify the optimizer algorithm optimizer = torch.optim.Adagrad(tbsm.parameters(), lr=args.learning_rate) total_time = 0 total_loss = 0 total_accu = 0 total_iter = 0 total_samp = 0 max_gA_test = 0 for j, (X, lS_o, lS_i, T) in enumerate(train_ld): if j >= nbatches: break t1 = time_wrap(use_gpu, device) batchSize = X[0].shape[0] # forward pass Z = tbsm(*data_wrap(X, lS_o, lS_i, use_gpu, device )) # loss E = loss_fn_wrap(Z, T, use_gpu, device) # compute loss and accuracy L = E.detach().cpu().numpy() # numpy array z = Z.detach().cpu().numpy() # numpy array t = T.detach().cpu().numpy() # numpy array # rounding t A = np.sum((np.round(z, 0) == np.round(t, 0)).astype(np.uint8)) optimizer.zero_grad() # backward pass E.backward(retain_graph=True) # weights update optimizer.step() t2 = time_wrap(use_gpu, device) total_time += t2 - t1 total_loss += (L * batchSize) total_accu += A total_iter += 1 total_samp += batchSize print_tl = ((j + 1) % args.print_freq == 0) or (j + 1 == nbatches) # print time, loss and accuracy if print_tl and isMainTraining: gT = 1000.0 * total_time / total_iter if args.print_time else -1 total_time = 0 gL = total_loss / total_samp total_loss = 0 gA = total_accu / total_samp total_accu = 0 str_run_type = "inference" if args.inference_only else "training" print( "Finished {} it {}/{} of epoch {}, ".format( str_run_type, j + 1, nbatches, k ) + "{:.2f} ms/it, loss {:.8f}, accuracy {:3.3f} %".format( gT, gL, gA * 100 ) ) total_iter = 0 total_samp = 0 if isMainTraining: should_test = ( (args.test_freq > 0 and (j + 1) % args.test_freq == 0) or j + 1 == nbatches ) else: should_test = (j == min(int(0.05 * len(train_ld)), len(train_ld) - 1)) # validation run if should_test: total_accu_test, total_samp_test, total_loss_val = iterate_val_data(val_ld, tbsm, use_gpu, device) gA_test = total_accu_test / total_samp_test if not isMainTraining: break gL_test = total_loss_val / total_samp_test print("At epoch {:d} validation accuracy is {:3.3f} %". format(k, gA_test * 100)) if args.enable_summary and isMainTraining: writer.add_scalars('train and val loss', {'train_loss': gL, 'val_loss': gL_test}, k * len(train_ld) + j) writer.add_scalars('train and val accuracy', {'train_acc': gA * 100, 'val_acc': gA_test * 100}, k * len(train_ld) + j) losses = np.append(losses, np.array([[j, gL, gL_test]]), axis=0) accuracies = np.append(accuracies, np.array([[j, gA * 100, gA_test * 100]]), axis=0) # save model if best so far if gA_test > max_gA_test and isMainTraining: print("Saving current model...") max_gA_test = gA_test model_ = tbsm torch.save( { "model_state_dict": model_.state_dict(), # "opt_state_dict": optimizer.state_dict(), }, args.save_model, ) if not isMainTraining: return gA_test # selects best seed, and does main model training def train_tbsm(args, use_gpu): # prepare the data train_ld, _ = tp.make_tbsm_data_and_loader(args, "train") val_ld, _ = tp.make_tbsm_data_and_loader(args, "val") # setup initial values isMainTraining = False writer = SummaryWriter() losses = np.empty((0,3), np.float32) accuracies = np.empty((0,3), np.float32) # selects best seed out of 5. Sometimes Adagrad gets stuck early, this # seems to occur randomly depending on initial weight values and # is independent of chosen model: N-inner, dot etc. # this procedure is used to reduce the probability of this happening. def select(args): seeds = np.random.randint(2, 10000, size=5) if args.debug_mode: print(seeds) best_index = 0 max_val_accuracy = 0.0 testpoint = min(int(0.05 * len(train_ld)), len(train_ld) - 1) print("testpoint, total batches: ", testpoint, len(train_ld)) for i, seed in enumerate(seeds): set_seed(seed, use_gpu) tbsm, device = get_tbsm(args, use_gpu) gA_test = iterate_train_data(args, train_ld, val_ld, tbsm, 0, use_gpu, device, writer, losses, accuracies, isMainTraining) if args.debug_mode: print("select: ", i, seed, gA_test, max_val_accuracy) if gA_test > max_val_accuracy: best_index = i max_val_accuracy = gA_test return seeds[best_index] # select best seed if needed if args.no_select_seed or path.exists(args.save_model): seed = args.numpy_rand_seed else: print("Choosing best seed...") seed = select(args) set_seed(seed, use_gpu) print("selected seed:", seed) # create or load TBSM tbsm, device = get_tbsm(args, use_gpu) if args.debug_mode: print("initial parameters (weights and bias):") for name, param in tbsm.named_parameters(): print(name) print(param.detach().cpu().numpy()) # main training loop isMainTraining = True print("time/loss/accuracy (if enabled):") with torch.autograd.profiler.profile(args.enable_profiling, use_gpu) as prof: for k in range(args.nepochs): iterate_train_data(args, train_ld, val_ld, tbsm, k, use_gpu, device, writer, losses, accuracies, isMainTraining) # collect metrics and other statistics about the run if args.enable_summary: with open('summary.npy', 'wb') as acc_loss: np.save(acc_loss, losses) np.save(acc_loss, accuracies) writer.close() # debug prints if args.debug_mode: print("final parameters (weights and bias):") for name, param in tbsm.named_parameters(): print(name) print(param.detach().cpu().numpy()) # profiling if args.enable_profiling: with open("tbsm_pytorch.prof", "w") as prof_f: prof_f.write( prof.key_averages(group_by_input_shape=True).table( sort_by="self_cpu_time_total" ) ) prof.export_chrome_trace("./tbsm_pytorch.json") return # evaluates model on test data and computes AUC metric def test_tbsm(args, use_gpu): # prepare data test_ld, N_test = tp.make_tbsm_data_and_loader(args, "test") # setup initial values z_test = np.zeros((N_test, ), dtype=np.float) t_test = np.zeros((N_test, ), dtype=np.float) # check saved model exists if not path.exists(args.save_model): sys.exit("Can't find saved model. Exiting...") # create or load TBSM tbsm, device = get_tbsm(args, use_gpu) print(args.save_model) # main eval loop # NOTE: call to tbsm.eval() not needed here, see # https://discuss.pytorch.org/t/model-eval-vs-with-torch-no-grad/19615 offset = 0 for _, (X, lS_o, lS_i, T) in enumerate(test_ld): batchSize = X[0].shape[0] Z = tbsm(*data_wrap(X, lS_o, lS_i, use_gpu, device )) z_test[offset: offset + batchSize] = np.squeeze(Z.detach().cpu().numpy(), axis=1) t_test[offset: offset + batchSize] = np.squeeze(T.detach().cpu().numpy(), axis=1) offset += batchSize if args.quality_metric == "auc": # compute AUC metric auc_score = 100.0 * roc_auc_score(t_test.astype(int), z_test) print("auc score: ", auc_score) else: sys.exit("Metric not supported.") if __name__ == "__main__": ### import packages ### import sys import argparse ### parse arguments ### parser = argparse.ArgumentParser(description="Time Based Sequence Model (TBSM)") # path to dlrm parser.add_argument("--dlrm-path", type=str, default="") # data type: taobao or synthetic (generic) parser.add_argument("--datatype", type=str, default="synthetic") # mode: train or inference or both parser.add_argument("--mode", type=str, default="train") # train, test, train-test # data locations parser.add_argument("--raw-train-file", type=str, default="./input/train.txt") parser.add_argument("--pro-train-file", type=str, default="./output/train.npz") parser.add_argument("--raw-test-file", type=str, default="./input/test.txt") parser.add_argument("--pro-test-file", type=str, default="./output/test.npz") parser.add_argument("--pro-val-file", type=str, default="./output/val.npz") parser.add_argument("--num-train-pts", type=int, default=100) parser.add_argument("--num-val-pts", type=int, default=20) # time series length for train/val and test parser.add_argument("--ts-length", type=int, default=20) # model_type = "tsl", "mha", "rnn" parser.add_argument("--model-type", type=str, default="tsl") # tsl, mha, rnn parser.add_argument("--tsl-seq", action="store_true", default=False) # k-seq method parser.add_argument("--tsl-proj", action="store_true", default=True) # sphere proj parser.add_argument("--tsl-inner", type=str, default="def") # ind, def, dot parser.add_argument("--tsl-num-heads", type=int, default=1) # num tsl components parser.add_argument("--mha-num-heads", type=int, default=8) # num mha heads parser.add_argument("--rnn-num-layers", type=int, default=5) # num rnn layers # num positive (and negative) points per user parser.add_argument("--points-per-user", type=int, default=10) # model arch related parameters # embedding dim for all sparse features (same for all features) parser.add_argument("--arch-sparse-feature-size", type=int, default=4) # emb_dim # number of distinct values for each sparse feature parser.add_argument("--arch-embedding-size", type=str, default="4-3-2") # vectors # for taobao use "987994-4162024-9439") # MLP 1: num dense fea --> embedding dim for sparse fea (out_dim enforced) parser.add_argument("--arch-mlp-bot", type=str, default="1-4") # MLP 2: num_interactions + bot[-1] --> top[-1] # (in_dim adjusted, out_dim can be any) parser.add_argument("--arch-mlp-top", type=str, default="2-2") # MLP 3: attention: ts_length --> ts_length (both adjusted) parser.add_argument("--tsl-mlp", type=str, default="2-2") # MLP 4: final prob. of click: 2 * top[-1] --> [0,1] (in_dim adjusted) parser.add_argument("--arch-mlp", type=str, default="4-1") # interactions parser.add_argument("--arch-interaction-op", type=str, default="dot") parser.add_argument("--arch-interaction-itself", action="store_true", default=False) parser.add_argument("--tsl-interaction-op", type=str, default="dot") parser.add_argument("--tsl-mechanism", type=str, default="mlp") # mul or MLP # data parser.add_argument("--num-batches", type=int, default=0) # training parser.add_argument("--mini-batch-size", type=int, default=1) parser.add_argument("--nepochs", type=int, default=1) parser.add_argument("--learning-rate", type=float, default=0.05) parser.add_argument("--print-precision", type=int, default=5) parser.add_argument("--numpy-rand-seed", type=int, default=123) parser.add_argument("--no-select-seed", action="store_true", default=False) # inference parser.add_argument("--quality-metric", type=str, default="auc") parser.add_argument("--test-freq", type=int, default=0) parser.add_argument("--inference-only", type=bool, default=False) # saving model parser.add_argument("--save-model", type=str, default="./output/model.pt") # gpu parser.add_argument("--use-gpu", action="store_true", default=False) parser.add_argument("--device-num", type=int, default=0) # debugging and profiling parser.add_argument("--debug-mode", action="store_true", default=False) parser.add_argument("--print-freq", type=int, default=1) parser.add_argument("--print-time", action="store_true", default=False) parser.add_argument("--enable-summary", action="store_true", default=False) parser.add_argument("--enable-profiling", action="store_true", default=False) parser.add_argument("--run-fast", action="store_true", default=False) args = parser.parse_args() # the code requires access to dlrm model if not path.exists(str(args.dlrm_path)): sys.exit("Please provide path to DLRM as --dlrm-path") sys.path.insert(1, args.dlrm_path) import dlrm_s_pytorch as dlrm if args.datatype == "taobao" and args.arch_embedding_size != "987994-4162024-9439": sys.exit( "ERROR: arch-embedding-size for taobao " + " needs to be 987994-4162024-9439" ) if args.tsl_inner not in ["def", "ind"] and int(args.tsl_num_heads) > 1: sys.exit( "ERROR: dot product " + " assumes one tsl component (due to redundancy)" ) # model_type = "tsl", "mha", "rnn" print("dlrm path: ", args.dlrm_path) print("model_type: ", args.model_type) print("time series length: ", args.ts_length) print("seed: ", args.numpy_rand_seed) print("model_file:", args.save_model) ### some basic setup ### use_gpu = args.use_gpu and torch.cuda.is_available() set_seed(args.numpy_rand_seed, use_gpu) np.set_printoptions(precision=args.print_precision) torch.set_printoptions(precision=args.print_precision) print("use-gpu:", use_gpu) # possible modes: # "train-test" for both training and metric computation on test data, # "train" for training model # "test" for metric computation on test data using saved trained model modes = args.mode.split("-") if modes[0] == "train": train_tbsm(args, use_gpu) if modes[0] == "test" or (len(modes) > 1 and modes[1] == "test"): test_tbsm(args, use_gpu)
35,338
36.917382
148
py
tbsm
tbsm-main/tbsm_synthetic.py
# Copyright (c) Facebook, Inc. and its affiliates. # # This source code is licensed under the MIT license found in the # LICENSE file in the root directory of this source tree. # miscellaneous from os import path import sys # numpy and scikit-learn import numpy as np from sklearn.metrics import roc_auc_score # pytorch import torch import torch.nn as nn from torch.nn.parameter import Parameter # In synthetic experiment we generate the output vectors z of the embedding # layer directly, therefore we create a custom TBSM, rather than instantiate # an existing general model. # Synthetic experiment code # It generates time series data in D dimensions # with the property that binary label has some dependency # on coupling between time series components in pairs of dimensions. def synthetic_experiment(): N, Nt, D, T = 50000, 5000, 5, 10 auc_results = np.empty((0, 5), np.float32) def generate_data(N, high): H = np.random.uniform(low=-1.0, high=1.0, size=N * D * T).reshape(N, T, D) w = np.random.uniform(low=-1.0, high=1.0, size=N * D).reshape(N, 1, D) return H, w for K in range(0, 31, 10): print("num q terms: ", K) # ----- train set ------ H, w = generate_data(N, 1.0) wt = np.transpose(w, (0, 2, 1)) p = np.zeros(D * K, dtype=np.int).reshape(K, D) for j in range(K): p[j, :] = np.random.permutation(D) wt2 = wt[:, p[j], :] wt = wt + wt2 Q = np.matmul(H[:, :, :], wt[:, :, :]) # similarity coefs Q = np.squeeze(Q, axis=2) R = np.mean(Q, axis=1) R = np.sign(R) # s1 = np.count_nonzero(R > 0) # print(Q.shape) # print("num pos, total: ", s1, N) R = R + 1 t_train = R.reshape(N, 1) z_train = np.concatenate((H, w), axis=1) # ----- test set ------ H, w = generate_data(Nt, 1.0) wt = np.transpose(w, (0, 2, 1)) for j in range(K): wt2 = wt[:, p[j], :] wt = wt + wt2 Q = np.matmul(H[:, :, :], wt[:, :, :]) # dot product Q = np.squeeze(Q, axis=2) R = np.mean(Q, axis=1) R = np.sign(R) + 1 t_test = R.reshape(Nt, 1) z_test = np.concatenate((H, w), axis=1) # debug prints # print(z_train.shape, t_train.shape) class SyntheticDataset: def __init__(self, F, y): self.F = F self.y = y def __getitem__(self, index): if isinstance(index, slice): return [ self[idx] for idx in range( index.start or 0, index.stop or len(self), index.step or 1 ) ] return self.F[index], self.y[index] def __len__(self): return len(self.y) ztraind = SyntheticDataset(z_train, t_train) ztestd = SyntheticDataset(z_test, t_test) def collate_zfn(list_of_tuples): data = list(zip(*list_of_tuples)) F = torch.tensor(data[0], dtype=torch.float) y = torch.tensor(data[1], dtype=torch.float) # y = torch.unsqueeze(y, 1) return F, y ztrain_ld = torch.utils.data.DataLoader( ztraind, batch_size=128, num_workers=0, collate_fn=collate_zfn, shuffle=True ) ztest_ld = torch.utils.data.DataLoader( ztestd, batch_size=Nt, num_workers=0, collate_fn=collate_zfn, ) ### define TBSM in PyTorch ### class TBSM_SubNet(nn.Module): def __init__( self, mode, num_inner, D, T, ): super(TBSM_SubNet, self).__init__() self.mode = mode self.num_inner = num_inner if self.mode in ["def", "ind", "dot"]: if self.mode in ["def", "ind"]: self.A = [] mean = 0.0 std_dev = np.sqrt(2 / (D + D)) for _ in range(self.num_inner): E = np.eye(D, dtype=np.float32) W = np.random.normal(mean, std_dev, size=(1, D, D)) \ .astype(np.float32) self.A.append(Parameter(torch.tensor(E + W), requires_grad=True)) d = self.num_inner * T # d = self.num_inner * D + D ln_mlp = np.array([d, 2 * d, 1]) self.mlp = dlrm.DLRM_Net().create_mlp(ln_mlp, ln_mlp.size - 2) elif self.mode == "mha": m = D # dim self.nheads = 8 self.emb_m = self.nheads * m # mha emb dim mean = 0.0 std_dev = np.sqrt(2 / (m + m)) # np.sqrt(1 / m) # np.sqrt(1 / n) qm = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.Q = Parameter(torch.tensor(qm), requires_grad=True) km = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.K = Parameter(torch.tensor(km), requires_grad=True) vm = np.random.normal(mean, std_dev, size=(1, m, self.emb_m)) \ .astype(np.float32) self.V = Parameter(torch.tensor(vm), requires_grad=True) d = self.nheads * m ln_mlp = np.array([d, 2 * d, 1]) self.mlp = dlrm.DLRM_Net().create_mlp(ln_mlp, ln_mlp.size - 2) else: d = D * (T + 1) ln_mlp = np.array([d, 2 * d, 1]) self.mlp = dlrm.DLRM_Net().create_mlp(ln_mlp, ln_mlp.size - 2) def forward(self, x): # H * w H = x[:, :-1, :] w = torch.unsqueeze(x[:, -1, :], dim=1) w = torch.transpose(w, 1, 2) # inner products if self.mode in ["def", "ind"]: for j in range(self.num_inner): aw = torch.matmul(self.A[j], w) if self.mode == "def": aw = torch.matmul(self.A[j].permute(0, 2, 1), aw) a1 = torch.bmm(H, aw) if j == 0: z = a1 else: z = torch.cat([z, a1], dim=1) z = torch.squeeze(z, dim=2) elif self.mode == "dot": z = torch.bmm(H, w) z = torch.squeeze(z, dim=2) elif self.mode == "mha": w = torch.transpose(w, 1, 2) # print("mha shapes: ", w.shape, self.Q.shape) Qx = torch.transpose(torch.matmul(w, self.Q), 0, 1) HK = torch.transpose(torch.matmul(H, self.K), 0, 1) HV = torch.transpose(torch.matmul(H, self.V), 0, 1) multihead_attn = nn.MultiheadAttention(self.emb_m, self.nheads) attn_output, _ = multihead_attn(Qx, HK, HV) # print("attn shape: ", attn_output.shape) z = torch.squeeze(attn_output, dim=0) else: # concat H = torch.flatten(H, start_dim=1, end_dim=2) w = torch.flatten(w, start_dim=1, end_dim=2) z = torch.cat([H, w], dim=1) # obtain probability of a click as a result of MLP p = dlrm.DLRM_Net().apply_mlp(z, self.mlp) return p def train_inner(znet): loss_fn = torch.nn.BCELoss(reduction="mean") # loss_fn = torch.nn.L1Loss(reduction="mean") optimizer = torch.optim.Adagrad(znet.parameters(), lr=0.05) # optimizer = torch.optim.SGD(znet.parameters(), lr=0.05) znet.train() nepochs = 1 for _ in range(nepochs): TA = 0 TS = 0 for _, (X, y) in enumerate(ztrain_ld): batchSize = X.shape[0] # forward pass Z = znet(X) # loss # print("Z, y: ", Z.shape, y.shape) E = loss_fn(Z, y) # compute loss and accuracy # L = E.detach().cpu().numpy() # numpy array z = Z.detach().cpu().numpy() # numpy array t = y.detach().cpu().numpy() # numpy array # rounding t: smooth labels case A = np.sum((np.round(z, 0) == np.round(t, 0)).astype(np.uint16)) TA += A TS += batchSize optimizer.zero_grad() # backward pass E.backward(retain_graph=True) # optimizer optimizer.step() # if j % 500 == 0: # acc = 100.0 * TA / TS # print("j, acc: ", j, acc) # TA = 0 # TS = 0 z_final = np.zeros(Nt, dtype=np.float) offset = 0 znet.eval() for _, (X, _) in enumerate(ztest_ld): batchSize = X.shape[0] Z = znet(X) z_final[offset: offset + batchSize] = \ np.squeeze(Z.detach().cpu().numpy(), axis=1) offset += batchSize # E = loss_fn(Z, y) # L = E.detach().cpu().numpy() # numpy array # loss_net = L # print(znet.num_inner, znet.mode, ": ", loss_net) auc_net = 100.0 * roc_auc_score(t_test.astype(int), z_final) print(znet.num_inner, znet.mode, ": ", auc_net) return auc_net dim = T znet = TBSM_SubNet("dot", 1, D, dim) # c or c,w res1 = train_inner(znet) znet = TBSM_SubNet("def", 1, D, dim) # c or c,w res2 = train_inner(znet) znet = TBSM_SubNet("def", 4, D, dim) # c or c,w res3 = train_inner(znet) znet = TBSM_SubNet("def", 8, D, dim) # c or c,w res4 = train_inner(znet) znet = TBSM_SubNet("mha", 1, D, dim) # c or c,w res5 = train_inner(znet) auc_results = np.append(auc_results, np.array([[res1, res2, res3, res4, res5]]), axis=0) print(auc_results) # np.savez_compressed( # 'auc_synthetic.npz', # auc_results=auc_results, # ) if __name__ == "__main__": import argparse parser = argparse.ArgumentParser(description="Synthetic data experiments (TBSM)") # path to dlrm parser.add_argument("--dlrm-path", type=str, default="") args = parser.parse_args() if not path.exists(str(args.dlrm_path)): sys.exit("Please provide path to DLRM as --dlrm-path") sys.path.insert(1, args.dlrm_path) import dlrm_s_pytorch as dlrm synthetic_experiment()
11,465
36.106796
88
py
tbsm
tbsm-main/tbsm_data_pytorch.py
# Copyright (c) Facebook, Inc. and its affiliates. # # This source code is licensed under the MIT license found in the # LICENSE file in the root directory of this source tree. ### import packages ### from __future__ import absolute_import, division, print_function, unicode_literals # miscellaneous from os import path import sys # numpy and scikit-learn import numpy as np # pytorch import torch # dataset (either synthetic or Taobao) class TBSMDataset(): def __init__( self, datatype, mode, ts_length=20, points_per_user=4, numpy_rand_seed=7, raw_path="", pro_data="", spa_fea_sizes="", num_pts=1, # pts to train or test ): # save arguments if mode == "train": self.numpy_rand_seed = numpy_rand_seed else: self.numpy_rand_seed = numpy_rand_seed + 31 self.mode = mode # save dataset parameters self.total = num_pts # number of lines in txt to process self.ts_length = ts_length self.points_per_user = points_per_user # pos and neg points per user self.spa_fea_sizes = spa_fea_sizes self.M = 200 # max history length # split the datafile into path and filename lstr = raw_path.split("/") self.d_path = "/".join(lstr[0:-1]) + "/" self.d_file = lstr[-1] # preprocess data if needed if path.exists(str(pro_data)): print("Reading pre-processed data=%s" % (str(pro_data))) file = str(pro_data) else: file = str(pro_data) levels = np.fromstring(self.spa_fea_sizes, dtype=int, sep="-") if datatype == "taobao": self.Unum = levels[0] # 987994 num of users self.Inum = levels[1] # 4162024 num of items self.Cnum = levels[2] # 9439 num of categories print("Reading raw data=%s" % (str(raw_path))) if self.mode == "test": self.build_taobao_test( raw_path, file, ) else: self.build_taobao_train_or_val( raw_path, file, ) elif datatype == "synthetic": self.build_synthetic_train_or_val( file, ) # load data with np.load(file) as data: self.X_cat = data["X_cat"] self.X_int = data["X_int"] self.y = data["y"] # common part between train/val and test generation # truncates (if needed) and shuffles data points def truncate_and_save(self, out_file, do_shuffle, t, users, items, cats, times, y): # truncate. If for some users we didn't generate had too short history # we truncate the unused portion of the pre-allocated matrix. if t < self.total_out: users = users[:t, :] items = items[:t, :] cats = cats[:t, :] times = times[:t, :] y = y[:t] # shuffle if do_shuffle: indices = np.arange(len(y)) indices = np.random.permutation(indices) users = users[indices] items = items[indices] cats = cats[indices] times = times[indices] y = y[indices] N = len(y) X_cat = np.zeros((3, N, self.ts_length + 1), dtype="i4") # 4 byte int X_int = np.zeros((1, N, self.ts_length + 1), dtype=np.float) X_cat[0, :, :] = users X_cat[1, :, :] = items X_cat[2, :, :] = cats X_int[0, :, :] = times # saving to compressed numpy file if not path.exists(out_file): np.savez_compressed( out_file, X_cat=X_cat, X_int=X_int, y=y, ) return # processes raw train or validation into npz format required by training # for train data out of each line in raw datafile produces several randomly chosen # datapoints, max number of datapoints per user is specified by points_per_user # argument, for validation data produces one datapoint per user. def build_taobao_train_or_val(self, raw_path, out_file): with open(str(raw_path)) as f: for i, _ in enumerate(f): if i % 50000 == 0: print("pre-processing line: ", i) self.total = min(self.total, i + 1) print("total lines: ", self.total) self.total_out = self.total * self.points_per_user * 2 # pos + neg points print("Total number of points in raw datafile: ", self.total) print("Total number of points in output will be at most: ", self.total_out) np.random.seed(self.numpy_rand_seed) r_target = np.arange(0, self.M - 1) time = np.arange(self.ts_length + 1, dtype=np.int32) / (self.ts_length + 1) # time = np.ones(self.ts_length + 1, dtype=np.int32) users = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int items = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int cats = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int times = np.zeros((self.total_out, self.ts_length + 1), dtype=np.float) y = np.zeros(self.total_out, dtype="i4") # 4 byte int # determine how many datapoints to take from each user based on the length of # user behavior sequence # ind=0, 1, 2, 3,... t < 10, 20, 30, 40, 50, 60, ... k = 20 regime = np.zeros(k, dtype=np.int) regime[1], regime[2], regime[3] = 1, 3, 6 for j in range(4, k): regime[j] = self.points_per_user if self.mode == "val": self.points_per_user = 1 for j in range(k): regime[j] = np.min([regime[j], self.points_per_user]) last = self.M - 1 # max index of last item # try to generate the desired number of points (time series) per each user. # if history is short it may not succeed to generate sufficiently different # time series for a particular user. t, t_pos, t_neg, t_short = 0, 0, 0, 0 with open(str(raw_path)) as f: for i, line in enumerate(f): if i % 1000 == 0: print("processing line: ", i, t, t_pos, t_neg, t_short) if i >= self.total: break units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") neg_item_hist_list = units[6].split(",") neg_cate_hist_list = units[7].split(",") user = np.array(np.maximum(np.int32(units[0]) - self.Inum, 0), dtype=np.int32) # y[i] = np.int32(units[3]) items_ = np.array( list(map(lambda x: np.maximum(np.int32(x), 0), item_hist_list)), dtype=np.int32 ) cats_ = np.array( list(map(lambda x: np.maximum(np.int32(x) - self.Inum - self.Unum, 0), cate_hist_list)), dtype=np.int32 ) neg_items_ = np.array( list(map(lambda x: np.maximum(np.int32(x), 0), neg_item_hist_list)), dtype=np.int32 ) neg_cats_ = np.array( list(map(lambda x: np.maximum(np.int32(x) - self.Inum - self.Unum, 0), neg_cate_hist_list)), dtype=np.int32 ) # select datapoints first = np.argmax(items_ > 0) ind = int((last - first) // 10) # index into regime array # pos for _ in range(regime[ind]): a1 = min(first + self.ts_length, last - 1) end = np.random.randint(a1, last) indices = np.arange(end - self.ts_length, end + 1) if items_[indices[0]] == 0: t_short += 1 items[t] = items_[indices] cats[t] = cats_[indices] users[t] = np.full(self.ts_length + 1, user) times[t] = time y[t] = 1 # check if np.any(users[t] < 0) or np.any(items[t] < 0) \ or np.any(cats[t] < 0): sys.exit("Categorical feature less than zero after \ processing. Aborting...") t += 1 t_pos += 1 # neg for _ in range(regime[ind]): a1 = min(first + self.ts_length - 1, last - 1) end = np.random.randint(a1, last) indices = np.arange(end - self.ts_length + 1, end + 1) if items_[indices[0]] == 0: t_short += 1 items[t, :-1] = items_[indices] cats[t, :-1] = cats_[indices] neg_indices = np.random.choice(r_target, 1, replace=False) # random final item items[t, -1] = neg_items_[neg_indices] cats[t, -1] = neg_cats_[neg_indices] users[t] = np.full(self.ts_length + 1, user) times[t] = time y[t] = 0 # check if np.any(users[t] < 0) or np.any(items[t] < 0) \ or np.any(cats[t] < 0): sys.exit("Categorical feature less than zero after \ processing. Aborting...") t += 1 t_neg += 1 print("total points, pos points, neg points: ", t, t_pos, t_neg) self.truncate_and_save(out_file, True, t, users, items, cats, times, y) return # processes raw test datafile into npz format required to be used by # inference step, produces one datapoint per user by taking last ts-length items def build_taobao_test(self, raw_path, out_file): with open(str(raw_path)) as f: for i, _ in enumerate(f): if i % 50000 == 0: print("pre-processing line: ", i) self.total = i + 1 self.total_out = self.total # pos + neg points print("ts_length: ", self.ts_length) print("Total number of points in raw datafile: ", self.total) print("Total number of points in output will be at most: ", self.total_out) time = np.arange(self.ts_length + 1, dtype=np.int32) / (self.ts_length + 1) users = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int items = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int cats = np.zeros((self.total_out, self.ts_length + 1), dtype="i4") # 4 byte int times = np.zeros((self.total_out, self.ts_length + 1), dtype=np.float) y = np.zeros(self.total_out, dtype="i4") # 4 byte int # try to generate the desired number of points (time series) per each user. # if history is short it may not succeed to generate sufficiently different # time series for a particular user. t, t_pos, t_neg = 0, 0, 0 with open(str(raw_path)) as f: for i, line in enumerate(f): if i % 1000 == 0: print("processing line: ", i, t, t_pos, t_neg) if i >= self.total: break units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") user = np.array(np.maximum(np.int32(units[0]) - self.Inum, 0), dtype=np.int32) y[t] = np.int32(units[3]) items_ = np.array( list(map(lambda x: np.maximum(np.int32(x), 0), item_hist_list)), dtype=np.int32 ) cats_ = np.array( list(map(lambda x: np.maximum(np.int32(x) - self.Inum - self.Unum, 0), cate_hist_list)), dtype=np.int32 ) # get pts items[t] = items_[-(self.ts_length + 1):] cats[t] = cats_[-(self.ts_length + 1):] users[t] = np.full(self.ts_length + 1, user) times[t] = time # check if np.any(users[t] < 0) or np.any(items[t] < 0) \ or np.any(cats[t] < 0): sys.exit("Categorical feature less than zero after \ processing. Aborting...") if y[t] == 1: t_pos += 1 else: t_neg += 1 t += 1 print("total points, pos points, neg points: ", t, t_pos, t_neg) self.truncate_and_save(out_file, False, t, users, items, cats, times, y) return # builds small synthetic data mimicking the structure of taobao data def build_synthetic_train_or_val(self, out_file): np.random.seed(123) fea_sizes = np.fromstring(self.spa_fea_sizes, dtype=int, sep="-") maxval = np.min(fea_sizes) num_s = len(fea_sizes) X_cat = np.random.randint(maxval, size=(num_s, self.total, self.ts_length + 1), dtype="i4") # 4 byte int X_int = np.random.uniform(0, 1, size=(1, self.total, self.ts_length + 1)) y = np.random.randint(0, 2, self.total, dtype="i4") # 4 byte int # saving to compressed numpy file if not path.exists(out_file): np.savez_compressed( out_file, X_cat=X_cat, X_int=X_int, y=y, ) return def __getitem__(self, index): if isinstance(index, slice): return [ self[idx] for idx in range( index.start or 0, index.stop or len(self), index.step or 1 ) ] return self.X_cat[:, index, :], self.X_int[:, index, :], self.y[index] def __len__(self): return len(self.y) # defines transform to be performed during each call to batch, # used by loader def collate_wrapper_tbsm(list_of_tuples): # turns tuple into X, S_o, S_i, take last ts_length items data = list(zip(*list_of_tuples)) all_cat = torch.tensor(data[0], dtype=torch.long) all_int = torch.tensor(data[1], dtype=torch.float) # print("shapes:", all_cat.shape, all_int.shape) num_den_fea = all_int.shape[1] num_cat_fea = all_cat.shape[1] batchSize = all_cat.shape[0] ts_len = all_cat.shape[2] all_int = torch.reshape(all_int, (batchSize, num_den_fea * ts_len)) X = [] lS_i = [] lS_o = [] # transform data into the form used in dlrm nn for j in range(ts_len): lS_i_h = [] for i in range(num_cat_fea): lS_i_h.append(all_cat[:, i, j]) lS_o_h = [torch.tensor(range(batchSize)) for _ in range(len(lS_i_h))] lS_i.append(lS_i_h) lS_o.append(lS_o_h) X.append(all_int[:, j].view(-1, 1)) T = torch.tensor(data[2], dtype=torch.float32).view(-1, 1) return X, lS_o, lS_i, T # creates a loader (train, val or test data) to be used in the main training loop # or during inference step def make_tbsm_data_and_loader(args, mode): if mode == "train": raw = args.raw_train_file proc = args.pro_train_file numpts = args.num_train_pts batchsize = args.mini_batch_size doshuffle = True elif mode == "val": raw = args.raw_train_file proc = args.pro_val_file numpts = args.num_val_pts batchsize = 25000 doshuffle = True else: raw = args.raw_test_file proc = args.pro_test_file numpts = 1 batchsize = 25000 doshuffle = False data = TBSMDataset( args.datatype, mode, args.ts_length, args.points_per_user, args.numpy_rand_seed, raw, proc, args.arch_embedding_size, numpts, ) loader = torch.utils.data.DataLoader( data, batch_size=batchsize, num_workers=0, collate_fn=collate_wrapper_tbsm, shuffle=doshuffle, ) return loader, len(data)
16,772
36.52349
88
py
tbsm
tbsm-main/tools/taobao_prepare.py
# Copyright (c) 2019 UIC-Paper # # This source code is licensed under the MIT license found in the # LICENSE file in the root directory of this source tree. # # Source: https://github.com/UIC-Paper/MIMN import cPickle as pkl import pandas as pd import random import numpy as np RAW_DATA_FILE = './data/taobao_data/UserBehavior.csv' DATASET_PKL = './data/taobao_data/dataset.pkl' Test_File = "./data/taobao_data/taobao_test.txt" Train_File = "./data/taobao_data/taobao_train.txt" Train_handle = open(Train_File, 'w') Test_handle = open(Test_File, 'w') Feature_handle = open("./data/taobao_data/taobao_feature.pkl",'w') MAX_LEN_ITEM = 200 def to_df(file_name): df = pd.read_csv(RAW_DATA_FILE, header=None, names=['uid', 'iid', 'cid', 'btag', 'time']) return df def remap(df): item_key = sorted(df['iid'].unique().tolist()) item_len = len(item_key) item_map = dict(zip(item_key, range(item_len))) df['iid'] = df['iid'].map(lambda x: item_map[x]) user_key = sorted(df['uid'].unique().tolist()) user_len = len(user_key) user_map = dict(zip(user_key, range(item_len, item_len + user_len))) df['uid'] = df['uid'].map(lambda x: user_map[x]) cate_key = sorted(df['cid'].unique().tolist()) cate_len = len(cate_key) cate_map = dict(zip(cate_key, range(user_len + item_len, user_len + item_len + cate_len))) df['cid'] = df['cid'].map(lambda x: cate_map[x]) btag_key = sorted(df['btag'].unique().tolist()) btag_len = len(btag_key) btag_map = dict(zip(btag_key, range(user_len + item_len + cate_len, user_len + item_len + cate_len + btag_len))) df['btag'] = df['btag'].map(lambda x: btag_map[x]) print(item_len, user_len, cate_len, btag_len) return df, item_len, user_len + item_len + cate_len + btag_len + 1 #+1 is for unknown target btag def gen_user_item_group(df, item_cnt, feature_size): user_df = df.sort_values(['uid', 'time']).groupby('uid') item_df = df.sort_values(['iid', 'time']).groupby('iid') print("group completed") return user_df, item_df def gen_dataset(user_df, item_df, item_cnt, feature_size, dataset_pkl): train_sample_list = [] test_sample_list = [] # get each user's last touch point time print(len(user_df)) user_last_touch_time = [] for uid, hist in user_df: user_last_touch_time.append(hist['time'].tolist()[-1]) print("get user last touch time completed") user_last_touch_time_sorted = sorted(user_last_touch_time) split_time = user_last_touch_time_sorted[int(len(user_last_touch_time_sorted) * 0.7)] cnt = 0 for uid, hist in user_df: cnt += 1 print(cnt) item_hist = hist['iid'].tolist() cate_hist = hist['cid'].tolist() btag_hist = hist['btag'].tolist() target_item_time = hist['time'].tolist()[-1] target_item = item_hist[-1] target_item_cate = cate_hist[-1] target_item_btag = feature_size label = 1 test = (target_item_time > split_time) # neg sampling neg = random.randint(0, 1) if neg == 1: label = 0 while target_item == item_hist[-1]: target_item = random.randint(0, item_cnt - 1) target_item_cate = item_df.get_group(target_item)['cid'].tolist()[0] target_item_btag = feature_size # the item history part of the sample item_part = [] for i in range(len(item_hist) - 1): item_part.append([uid, item_hist[i], cate_hist[i], btag_hist[i]]) item_part.append([uid, target_item, target_item_cate, target_item_btag]) # item_part_len = min(len(item_part), MAX_LEN_ITEM) # choose the item side information: which user has clicked the target item # padding history with 0 if len(item_part) <= MAX_LEN_ITEM: item_part_pad = [[0] * 4] * (MAX_LEN_ITEM - len(item_part)) + item_part else: item_part_pad = item_part[len(item_part) - MAX_LEN_ITEM:len(item_part)] # gen sample # sample = (label, item_part_pad, item_part_len, user_part_pad, user_part_len) if test: # test_set.append(sample) cat_list = [] item_list = [] # btag_list = [] for i in range(len(item_part_pad)): item_list.append(item_part_pad[i][1]) cat_list.append(item_part_pad[i][2]) # cat_list.append(item_part_pad[i][0]) test_sample_list.append(str(uid) + "\t" + str(target_item) + "\t" + str(target_item_cate) + "\t" + str(label) + "\t" + ",".join(map(str, item_list)) + "\t" +",".join(map(str, cat_list))+"\n") else: cat_list = [] item_list = [] # btag_list = [] for i in range(len(item_part_pad)): item_list.append(item_part_pad[i][1]) cat_list.append(item_part_pad[i][2]) train_sample_list.append(str(uid) + "\t" + str(target_item) + "\t" + str(target_item_cate) + "\t" + str(label) + "\t" + ",".join(map(str, item_list)) + "\t" +",".join(map(str, cat_list))+"\n") train_sample_length_quant = len(train_sample_list)/256*256 test_sample_length_quant = len(test_sample_list)/256*256 print("length",len(train_sample_list)) train_sample_list = train_sample_list[:train_sample_length_quant] test_sample_list = test_sample_list[:test_sample_length_quant] random.shuffle(train_sample_list) print("length",len(train_sample_list)) return train_sample_list, test_sample_list def produce_neg_item_hist_with_cate(train_file, test_file): item_dict = {} sample_count = 0 hist_seq = 0 for line in train_file: units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") hist_list = zip(item_hist_list, cate_hist_list) hist_seq = len(hist_list) sample_count += 1 for item in hist_list: item_dict.setdefault(str(item),0) for line in test_file: units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") hist_list = zip(item_hist_list, cate_hist_list) hist_seq = len(hist_list) sample_count += 1 for item in hist_list: item_dict.setdefault(str(item),0) del(item_dict["('0', '0')"]) neg_array = np.random.choice(np.array(item_dict.keys()), (sample_count, hist_seq+20)) neg_list = neg_array.tolist() sample_count = 0 for line in train_file: units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") hist_list = zip(item_hist_list, cate_hist_list) hist_seq = len(hist_list) neg_hist_list = [] for item in neg_list[sample_count]: item = eval(item) if item not in hist_list: neg_hist_list.append(item) if len(neg_hist_list) == hist_seq: break sample_count += 1 neg_item_list, neg_cate_list = zip(*neg_hist_list) Train_handle.write(line.strip() + "\t" + ",".join(neg_item_list) + "\t" + ",".join(neg_cate_list) + "\n" ) for line in test_file: units = line.strip().split("\t") item_hist_list = units[4].split(",") cate_hist_list = units[5].split(",") hist_list = zip(item_hist_list, cate_hist_list) hist_seq = len(hist_list) neg_hist_list = [] for item in neg_list[sample_count]: item = eval(item) if item not in hist_list: neg_hist_list.append(item) if len(neg_hist_list) == hist_seq: break sample_count += 1 neg_item_list, neg_cate_list = zip(*neg_hist_list) Test_handle.write(line.strip() + "\t" + ",".join(neg_item_list) + "\t" + ",".join(neg_cate_list) + "\n" ) def main(): df = to_df(RAW_DATA_FILE) df, item_cnt, feature_size = remap(df) print("feature_size", item_cnt, feature_size) feature_total_num = feature_size + 1 pkl.dump(feature_total_num, Feature_handle) user_df, item_df = gen_user_item_group(df, item_cnt, feature_size) train_sample_list, test_sample_list = gen_dataset(user_df, item_df, item_cnt, feature_size, DATASET_PKL) produce_neg_item_hist_with_cate(train_sample_list, test_sample_list) if __name__ == '__main__': main()
8,508
36.484581
204
py
rllab
rllab-master/setup.py
# setup.py from setuptools import setup,find_packages setup( name='rllab', packages=[package for package in find_packages() if package.startswith('rllab')], version='0.1.0', )
205
19.6
52
py
rllab
rllab-master/examples/trpo_cartpole_recurrent.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.policies.gaussian_gru_policy import GaussianGRUPolicy from rllab.optimizers.conjugate_gradient_optimizer import ConjugateGradientOptimizer, FiniteDifferenceHvp from rllab.misc.instrument import run_experiment_lite def run_task(*_): env = normalize(CartpoleEnv()) policy = GaussianGRUPolicy( env_spec=env.spec, ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=10, discount=0.99, step_size=0.01, optimizer=ConjugateGradientOptimizer(hvp_approach=FiniteDifferenceHvp(base_eps=1e-5)) ) algo.train() run_experiment_lite( run_task, n_parallel=1, seed=1, )
1,003
25.421053
105
py
rllab
rllab-master/examples/cluster_gym_mujoco_demo.py
from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.normalized_env import normalize from sandbox.rocky.tf.envs.base import TfEnv from sandbox.rocky.tf.policies.gaussian_mlp_policy import GaussianMLPPolicy from sandbox.rocky.tf.algos.trpo import TRPO from rllab.misc.instrument import run_experiment_lite from rllab.envs.gym_env import GymEnv import sys from rllab.misc.instrument import VariantGenerator, variant class VG(VariantGenerator): @variant def step_size(self): return [0.01, 0.05, 0.1] @variant def seed(self): return [1, 11, 21, 31, 41] def run_task(vv): env = TfEnv(normalize(GymEnv('HalfCheetah-v1', record_video=False, record_log=False))) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32), name="policy" ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=40, discount=0.99, step_size=vv["step_size"], # Uncomment both lines (this and the plot parameter below) to enable plotting # plot=True, ) algo.train() variants = VG().variants() for v in variants: run_experiment_lite( run_task, exp_prefix="first_exp", # Number of parallel workers for sampling n_parallel=1, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=v["seed"], # mode="local", mode="ec2", variant=v, # plot=True, # terminate_machine=False, ) sys.exit()
1,919
25.30137
93
py
rllab
rllab-master/examples/trpo_cartpole.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy env = normalize(CartpoleEnv()) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=40, discount=0.99, step_size=0.01, ) algo.train()
713
24.5
89
py
rllab
rllab-master/examples/trpo_cartpole_pickled.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import run_experiment_lite from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy def run_task(*_): env = normalize(CartpoleEnv()) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=1000, discount=0.99, step_size=0.01, # Uncomment both lines (this and the plot parameter below) to enable plotting #plot=True ) algo.train() run_experiment_lite( run_task, # Number of parallel workers for sampling n_parallel=2, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=1, #plot=True )
1,287
27
93
py
rllab
rllab-master/examples/vpg_2.py
from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy from rllab.envs.normalized_env import normalize import numpy as np import theano import theano.tensor as TT from lasagne.updates import adam # normalize() makes sure that the actions for the environment lies # within the range [-1, 1] (only works for environments with continuous actions) env = normalize(CartpoleEnv()) # Initialize a neural network policy with a single hidden layer of 8 hidden units policy = GaussianMLPPolicy(env.spec, hidden_sizes=(8,)) # Initialize a linear baseline estimator using default hand-crafted features baseline = LinearFeatureBaseline(env.spec) # We will collect 100 trajectories per iteration N = 100 # Each trajectory will have at most 100 time steps T = 100 # Number of iterations n_itr = 100 # Set the discount factor for the problem discount = 0.99 # Learning rate for the gradient update learning_rate = 0.1 # Construct the computation graph # Create a Theano variable for storing the observations # We could have simply written `observations_var = TT.matrix('observations')` instead for this example. However, # doing it in a slightly more abstract way allows us to delegate to the environment for handling the correct data # type for the variable. For instance, for an environment with discrete observations, we might want to use integer # types if the observations are represented as one-hot vectors. observations_var = env.observation_space.new_tensor_variable( 'observations', # It should have 1 extra dimension since we want to represent a list of observations extra_dims=1 ) actions_var = env.action_space.new_tensor_variable( 'actions', extra_dims=1 ) advantages_var = TT.vector('advantages') # policy.dist_info_sym returns a dictionary, whose values are symbolic expressions for quantities related to the # distribution of the actions. For a Gaussian policy, it contains the mean and (log) standard deviation. dist_info_vars = policy.dist_info_sym(observations_var) # policy.distribution returns a distribution object under rllab.distributions. It contains many utilities for computing # distribution-related quantities, given the computed dist_info_vars. Below we use dist.log_likelihood_sym to compute # the symbolic log-likelihood. For this example, the corresponding distribution is an instance of the class # rllab.distributions.DiagonalGaussian dist = policy.distribution # Note that we negate the objective, since most optimizers assume a # minimization problem surr = - TT.mean(dist.log_likelihood_sym(actions_var, dist_info_vars) * advantages_var) # Get the list of trainable parameters. params = policy.get_params(trainable=True) grads = theano.grad(surr, params) f_train = theano.function( inputs=[observations_var, actions_var, advantages_var], outputs=None, updates=adam(grads, params, learning_rate=learning_rate), allow_input_downcast=True ) for _ in range(n_itr): paths = [] for _ in range(N): observations = [] actions = [] rewards = [] observation = env.reset() for _ in range(T): # policy.get_action() returns a pair of values. The second one returns a dictionary, whose values contains # sufficient statistics for the action distribution. It should at least contain entries that would be # returned by calling policy.dist_info(), which is the non-symbolic analog of policy.dist_info_sym(). # Storing these statistics is useful, e.g., when forming importance sampling ratios. In our case it is # not needed. action, _ = policy.get_action(observation) # Recall that the last entry of the tuple stores diagnostic information about the environment. In our # case it is not needed. next_observation, reward, terminal, _ = env.step(action) observations.append(observation) actions.append(action) rewards.append(reward) observation = next_observation if terminal: # Finish rollout if terminal state reached break # We need to compute the empirical return for each time step along the # trajectory path = dict( observations=np.array(observations), actions=np.array(actions), rewards=np.array(rewards), ) path_baseline = baseline.predict(path) advantages = [] returns = [] return_so_far = 0 for t in range(len(rewards) - 1, -1, -1): return_so_far = rewards[t] + discount * return_so_far returns.append(return_so_far) advantage = return_so_far - path_baseline[t] advantages.append(advantage) # The advantages are stored backwards in time, so we need to revert it advantages = np.array(advantages[::-1]) # And we need to do the same thing for the list of returns returns = np.array(returns[::-1]) advantages = (advantages - np.mean(advantages)) / (np.std(advantages) + 1e-8) path["advantages"] = advantages path["returns"] = returns paths.append(path) baseline.fit(paths) observations = np.concatenate([p["observations"] for p in paths]) actions = np.concatenate([p["actions"] for p in paths]) advantages = np.concatenate([p["advantages"] for p in paths]) f_train(observations, actions, advantages) print('Average Return:', np.mean([sum(p["rewards"]) for p in paths]))
5,669
40.086957
119
py
rllab
rllab-master/examples/cluster_demo.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import stub, run_experiment_lite from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy import sys def run_task(v): env = normalize(CartpoleEnv()) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=40, discount=0.99, step_size=v["step_size"], # Uncomment both lines (this and the plot parameter below) to enable plotting # plot=True, ) algo.train() for step_size in [0.01, 0.05, 0.1]: for seed in [1, 11, 21, 31, 41]: run_experiment_lite( run_task, exp_prefix="first_exp", # Number of parallel workers for sampling n_parallel=1, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=seed, # mode="local", mode="ec2", variant=dict(step_size=step_size, seed=seed) # plot=True, # terminate_machine=False, ) sys.exit()
1,682
29.6
93
py
rllab
rllab-master/examples/trpo_gym_cartpole.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.gym_env import GymEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import run_experiment_lite from rllab.policies.categorical_mlp_policy import CategoricalMLPPolicy def run_task(*_): # Please note that different environments with different action spaces may # require different policies. For example with a Discrete action space, a # CategoricalMLPPolicy works, but for a Box action space may need to use # a GaussianMLPPolicy (see the trpo_gym_pendulum.py example) env = normalize(GymEnv("CartPole-v0")) policy = CategoricalMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=env.horizon, n_itr=50, discount=0.99, step_size=0.01, # Uncomment both lines (this and the plot parameter below) to enable plotting # plot=True, ) algo.train() run_experiment_lite( run_task, # Number of parallel workers for sampling n_parallel=1, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=1, # plot=True, )
1,597
30.96
93
py
rllab
rllab-master/examples/trpo_gym_pendulum.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.gym_env import GymEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import run_experiment_lite from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy def run_task(*_): # Please note that different environments with different action spaces may require different # policies. For example with a Box action space, a GaussianMLPPolicy works, but for a Discrete # action space may need to use a CategoricalMLPPolicy (see the trpo_gym_cartpole.py example) env = normalize(GymEnv("Pendulum-v0")) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=env.horizon, n_itr=50, discount=0.99, step_size=0.01, # Uncomment both lines (this and the plot parameter below) to enable plotting # plot=True, ) algo.train() run_experiment_lite( run_task, # Number of parallel workers for sampling n_parallel=1, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=1, # plot=True, )
1,582
31.306122
98
py
rllab
rllab-master/examples/trpo_point.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from examples.point_env import PointEnv from rllab.envs.normalized_env import normalize from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy env = normalize(PointEnv()) policy = GaussianMLPPolicy( env_spec=env.spec, ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, ) algo.train()
478
25.611111
73
py
rllab
rllab-master/examples/ddpg_cartpole.py
from rllab.algos.ddpg import DDPG from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import run_experiment_lite from rllab.exploration_strategies.ou_strategy import OUStrategy from rllab.policies.deterministic_mlp_policy import DeterministicMLPPolicy from rllab.q_functions.continuous_mlp_q_function import ContinuousMLPQFunction def run_task(*_): env = normalize(CartpoleEnv()) policy = DeterministicMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) es = OUStrategy(env_spec=env.spec) qf = ContinuousMLPQFunction(env_spec=env.spec) algo = DDPG( env=env, policy=policy, es=es, qf=qf, batch_size=32, max_path_length=100, epoch_length=1000, min_pool_size=10000, n_epochs=1000, discount=0.99, scale_reward=0.01, qf_learning_rate=1e-3, policy_learning_rate=1e-4, # Uncomment both lines (this and the plot parameter below) to enable plotting # plot=True, ) algo.train() run_experiment_lite( run_task, # Number of parallel workers for sampling n_parallel=1, # Only keep the snapshot parameters for the last iteration snapshot_mode="last", # Specifies the seed for the experiment. If this is not provided, a random seed # will be used seed=1, # plot=True, )
1,538
28.037736
93
py
rllab
rllab-master/examples/trpo_swimmer.py
from rllab.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.mujoco.swimmer_env import SwimmerEnv from rllab.envs.normalized_env import normalize from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy env = normalize(SwimmerEnv()) policy = GaussianMLPPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=500, n_itr=40, discount=0.99, step_size=0.01, ) algo.train()
711
24.428571
89
py
rllab
rllab-master/examples/point_env.py
from rllab.envs.base import Env from rllab.spaces import Box from rllab.envs.base import Step import numpy as np class PointEnv(Env): @property def observation_space(self): return Box(low=-np.inf, high=np.inf, shape=(2,)) @property def action_space(self): return Box(low=-0.1, high=0.1, shape=(2,)) def reset(self): self._state = np.random.uniform(-1, 1, size=(2,)) observation = np.copy(self._state) return observation def step(self, action): self._state = self._state + action x, y = self._state reward = - (x ** 2 + y ** 2) ** 0.5 done = abs(x) < 0.01 and abs(y) < 0.01 next_observation = np.copy(self._state) return Step(observation=next_observation, reward=reward, done=done) def render(self): print('current state:', self._state)
866
26.967742
75
py
rllab
rllab-master/examples/__init__.py
0
0
0
py
rllab
rllab-master/examples/nop_cartpole.py
from rllab.algos.nop import NOP from rllab.baselines.zero_baseline import ZeroBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from rllab.policies.uniform_control_policy import UniformControlPolicy env = normalize(CartpoleEnv()) policy = UniformControlPolicy( env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. ) baseline = ZeroBaseline(env_spec=env.spec) algo = NOP( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=40, discount=0.99, step_size=0.01, ) algo.train()
665
23.666667
89
py
rllab
rllab-master/examples/vpg_1.py
from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy from rllab.envs.normalized_env import normalize import numpy as np import theano import theano.tensor as TT from lasagne.updates import adam # normalize() makes sure that the actions for the environment lies # within the range [-1, 1] (only works for environments with continuous actions) env = normalize(CartpoleEnv()) # Initialize a neural network policy with a single hidden layer of 8 hidden units policy = GaussianMLPPolicy(env.spec, hidden_sizes=(8,)) # We will collect 100 trajectories per iteration N = 100 # Each trajectory will have at most 100 time steps T = 100 # Number of iterations n_itr = 100 # Set the discount factor for the problem discount = 0.99 # Learning rate for the gradient update learning_rate = 0.01 # Construct the computation graph # Create a Theano variable for storing the observations # We could have simply written `observations_var = TT.matrix('observations')` instead for this example. However, # doing it in a slightly more abstract way allows us to delegate to the environment for handling the correct data # type for the variable. For instance, for an environment with discrete observations, we might want to use integer # types if the observations are represented as one-hot vectors. observations_var = env.observation_space.new_tensor_variable( 'observations', # It should have 1 extra dimension since we want to represent a list of observations extra_dims=1 ) actions_var = env.action_space.new_tensor_variable( 'actions', extra_dims=1 ) returns_var = TT.vector('returns') # policy.dist_info_sym returns a dictionary, whose values are symbolic expressions for quantities related to the # distribution of the actions. For a Gaussian policy, it contains the mean and the logarithm of the standard deviation. dist_info_vars = policy.dist_info_sym(observations_var) # policy.distribution returns a distribution object under rllab.distributions. It contains many utilities for computing # distribution-related quantities, given the computed dist_info_vars. Below we use dist.log_likelihood_sym to compute # the symbolic log-likelihood. For this example, the corresponding distribution is an instance of the class # rllab.distributions.DiagonalGaussian dist = policy.distribution # Note that we negate the objective, since most optimizers assume a minimization problem surr = - TT.mean(dist.log_likelihood_sym(actions_var, dist_info_vars) * returns_var) # Get the list of trainable parameters. params = policy.get_params(trainable=True) grads = theano.grad(surr, params) f_train = theano.function( inputs=[observations_var, actions_var, returns_var], outputs=None, updates=adam(grads, params, learning_rate=learning_rate), allow_input_downcast=True ) for _ in range(n_itr): paths = [] for _ in range(N): observations = [] actions = [] rewards = [] observation = env.reset() for _ in range(T): # policy.get_action() returns a pair of values. The second one returns a dictionary, whose values contains # sufficient statistics for the action distribution. It should at least contain entries that would be # returned by calling policy.dist_info(), which is the non-symbolic analog of policy.dist_info_sym(). # Storing these statistics is useful, e.g., when forming importance sampling ratios. In our case it is # not needed. action, _ = policy.get_action(observation) # Recall that the last entry of the tuple stores diagnostic information about the environment. In our # case it is not needed. next_observation, reward, terminal, _ = env.step(action) observations.append(observation) actions.append(action) rewards.append(reward) observation = next_observation if terminal: # Finish rollout if terminal state reached break # We need to compute the empirical return for each time step along the # trajectory returns = [] return_so_far = 0 for t in range(len(rewards) - 1, -1, -1): return_so_far = rewards[t] + discount * return_so_far returns.append(return_so_far) # The returns are stored backwards in time, so we need to revert it returns = returns[::-1] paths.append(dict( observations=np.array(observations), actions=np.array(actions), rewards=np.array(rewards), returns=np.array(returns) )) observations = np.concatenate([p["observations"] for p in paths]) actions = np.concatenate([p["actions"] for p in paths]) returns = np.concatenate([p["returns"] for p in paths]) f_train(observations, actions, returns) print('Average Return:', np.mean([sum(p["rewards"]) for p in paths]))
5,002
40.347107
119
py
rllab
rllab-master/examples/trpo_gym_tf_cartpole.py
from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.gym_env import GymEnv from rllab.envs.normalized_env import normalize from rllab.misc.instrument import stub, run_experiment_lite from sandbox.rocky.tf.envs.base import TfEnv from sandbox.rocky.tf.policies.categorical_mlp_policy import CategoricalMLPPolicy from sandbox.rocky.tf.algos.trpo import TRPO stub(globals()) # Need to wrap in a tf environment and force_reset to true # see https://github.com/openai/rllab/issues/87#issuecomment-282519288 env = TfEnv(normalize(GymEnv("CartPole-v0", force_reset=True))) policy = CategoricalMLPPolicy( name="policy", env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=200, n_itr=120, discount=0.99, step_size=0.01, # optimizer=ConjugateGradientOptimizer(hvp_approach=FiniteDifferenceHvp(base_eps=1e-5)) ) run_experiment_lite( algo.train(), n_parallel=1, snapshot_mode="last", seed=1 )
1,194
26.790698
91
py
rllab
rllab-master/sandbox/__init__.py
0
0
0
py
rllab
rllab-master/sandbox/rocky/__init__.py
0
0
0
py
rllab
rllab-master/sandbox/rocky/tf/__init__.py
0
0
0
py
rllab
rllab-master/sandbox/rocky/tf/launchers/vpg_cartpole.py
from sandbox.rocky.tf.algos.vpg import VPG from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from sandbox.rocky.tf.policies.gaussian_mlp_policy import GaussianMLPPolicy from sandbox.rocky.tf.envs.base import TfEnv from rllab.misc.instrument import stub, run_experiment_lite env = TfEnv(normalize(CartpoleEnv())) policy = GaussianMLPPolicy( name="policy", env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = VPG( env=env, policy=policy, baseline=baseline, batch_size=10000, max_path_length=100, n_itr=40, discount=0.99, optimizer_args=dict( tf_optimizer_args=dict( learning_rate=0.01, ) ) ) algo.train()
949
26.142857
89
py
rllab
rllab-master/sandbox/rocky/tf/launchers/trpo_cartpole_recurrent.py
from sandbox.rocky.tf.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from sandbox.rocky.tf.policies.gaussian_gru_policy import GaussianGRUPolicy from sandbox.rocky.tf.policies.gaussian_lstm_policy import GaussianLSTMPolicy from sandbox.rocky.tf.envs.base import TfEnv import sandbox.rocky.tf.core.layers as L from sandbox.rocky.tf.optimizers.conjugate_gradient_optimizer import ConjugateGradientOptimizer, FiniteDifferenceHvp from rllab.misc.instrument import stub, run_experiment_lite env = TfEnv(normalize(CartpoleEnv())) policy = GaussianLSTMPolicy( name="policy", env_spec=env.spec, lstm_layer_cls=L.TfBasicLSTMLayer, # gru_layer_cls=L.GRULayer, ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=10, discount=0.99, step_size=0.01, optimizer=ConjugateGradientOptimizer(hvp_approach=FiniteDifferenceHvp(base_eps=1e-5)) ) algo.train()
1,148
31.828571
116
py
rllab
rllab-master/sandbox/rocky/tf/launchers/trpo_cartpole.py
from sandbox.rocky.tf.algos.trpo import TRPO from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline from rllab.envs.box2d.cartpole_env import CartpoleEnv from rllab.envs.normalized_env import normalize from sandbox.rocky.tf.optimizers.conjugate_gradient_optimizer import ConjugateGradientOptimizer from sandbox.rocky.tf.optimizers.conjugate_gradient_optimizer import FiniteDifferenceHvp from sandbox.rocky.tf.policies.gaussian_mlp_policy import GaussianMLPPolicy from sandbox.rocky.tf.envs.base import TfEnv from rllab.misc.instrument import stub, run_experiment_lite env = TfEnv(normalize(CartpoleEnv())) policy = GaussianMLPPolicy( name="policy", env_spec=env.spec, # The neural network policy should have two hidden layers, each with 32 hidden units. hidden_sizes=(32, 32) ) baseline = LinearFeatureBaseline(env_spec=env.spec) algo = TRPO( env=env, policy=policy, baseline=baseline, batch_size=4000, max_path_length=100, n_itr=40, discount=0.99, step_size=0.01, # optimizer=ConjugateGradientOptimizer(hvp_approach=FiniteDifferenceHvp(base_eps=1e-5)) ) algo.train()
1,144
31.714286
95
py
rllab
rllab-master/sandbox/rocky/tf/launchers/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/core/network.py
import sandbox.rocky.tf.core.layers as L import tensorflow as tf import numpy as np import itertools from rllab.core.serializable import Serializable from sandbox.rocky.tf.core.parameterized import Parameterized from sandbox.rocky.tf.core.layers_powered import LayersPowered class MLP(LayersPowered, Serializable): def __init__(self, name, output_dim, hidden_sizes, hidden_nonlinearity, output_nonlinearity, hidden_W_init=L.XavierUniformInitializer(), hidden_b_init=tf.zeros_initializer(), output_W_init=L.XavierUniformInitializer(), output_b_init=tf.zeros_initializer(), input_var=None, input_layer=None, input_shape=None, batch_normalization=False, weight_normalization=False, ): Serializable.quick_init(self, locals()) with tf.variable_scope(name): if input_layer is None: l_in = L.InputLayer(shape=(None,) + input_shape, input_var=input_var, name="input") else: l_in = input_layer self._layers = [l_in] l_hid = l_in if batch_normalization: l_hid = L.batch_norm(l_hid) for idx, hidden_size in enumerate(hidden_sizes): l_hid = L.DenseLayer( l_hid, num_units=hidden_size, nonlinearity=hidden_nonlinearity, name="hidden_%d" % idx, W=hidden_W_init, b=hidden_b_init, weight_normalization=weight_normalization ) if batch_normalization: l_hid = L.batch_norm(l_hid) self._layers.append(l_hid) l_out = L.DenseLayer( l_hid, num_units=output_dim, nonlinearity=output_nonlinearity, name="output", W=output_W_init, b=output_b_init, weight_normalization=weight_normalization ) if batch_normalization: l_out = L.batch_norm(l_out) self._layers.append(l_out) self._l_in = l_in self._l_out = l_out # self._input_var = l_in.input_var self._output = L.get_output(l_out) LayersPowered.__init__(self, l_out) @property def input_layer(self): return self._l_in @property def output_layer(self): return self._l_out @property def input_var(self): return self._l_in.input_var @property def layers(self): return self._layers @property def output(self): return self._output class ConvNetwork(LayersPowered, Serializable): def __init__(self, name, input_shape, output_dim, conv_filters, conv_filter_sizes, conv_strides, conv_pads, hidden_sizes, hidden_nonlinearity, output_nonlinearity, hidden_W_init=L.XavierUniformInitializer(), hidden_b_init=tf.zeros_initializer(), output_W_init=L.XavierUniformInitializer(), output_b_init=tf.zeros_initializer(), input_var=None, input_layer=None, batch_normalization=False, weight_normalization=False): Serializable.quick_init(self, locals()) """ A network composed of several convolution layers followed by some fc layers. input_shape: (width,height,channel) HOWEVER, network inputs are assumed flattened. This network will first unflatten the inputs and then apply the standard convolutions and so on. conv_filters: a list of numbers of convolution kernel conv_filter_sizes: a list of sizes (int) of the convolution kernels conv_strides: a list of strides (int) of the conv kernels conv_pads: a list of pad formats (either 'SAME' or 'VALID') hidden_nonlinearity: a nonlinearity from tf.nn, shared by all conv and fc layers hidden_sizes: a list of numbers of hidden units for all fc layers """ with tf.variable_scope(name): if input_layer is not None: l_in = input_layer l_hid = l_in elif len(input_shape) == 3: l_in = L.InputLayer(shape=(None, np.prod(input_shape)), input_var=input_var, name="input") l_hid = L.reshape(l_in, ([0],) + input_shape, name="reshape_input") elif len(input_shape) == 2: l_in = L.InputLayer(shape=(None, np.prod(input_shape)), input_var=input_var, name="input") input_shape = (1,) + input_shape l_hid = L.reshape(l_in, ([0],) + input_shape, name="reshape_input") else: l_in = L.InputLayer(shape=(None,) + input_shape, input_var=input_var, name="input") l_hid = l_in if batch_normalization: l_hid = L.batch_norm(l_hid) for idx, conv_filter, filter_size, stride, pad in zip( range(len(conv_filters)), conv_filters, conv_filter_sizes, conv_strides, conv_pads, ): l_hid = L.Conv2DLayer( l_hid, num_filters=conv_filter, filter_size=filter_size, stride=(stride, stride), pad=pad, nonlinearity=hidden_nonlinearity, name="conv_hidden_%d" % idx, weight_normalization=weight_normalization, ) if batch_normalization: l_hid = L.batch_norm(l_hid) if output_nonlinearity == L.spatial_expected_softmax: assert len(hidden_sizes) == 0 assert output_dim == conv_filters[-1] * 2 l_hid.nonlinearity = tf.identity l_out = L.SpatialExpectedSoftmaxLayer(l_hid) else: l_hid = L.flatten(l_hid, name="conv_flatten") for idx, hidden_size in enumerate(hidden_sizes): l_hid = L.DenseLayer( l_hid, num_units=hidden_size, nonlinearity=hidden_nonlinearity, name="hidden_%d" % idx, W=hidden_W_init, b=hidden_b_init, weight_normalization=weight_normalization, ) if batch_normalization: l_hid = L.batch_norm(l_hid) l_out = L.DenseLayer( l_hid, num_units=output_dim, nonlinearity=output_nonlinearity, name="output", W=output_W_init, b=output_b_init, weight_normalization=weight_normalization, ) if batch_normalization: l_out = L.batch_norm(l_out) self._l_in = l_in self._l_out = l_out # self._input_var = l_in.input_var LayersPowered.__init__(self, l_out) @property def input_layer(self): return self._l_in @property def output_layer(self): return self._l_out @property def input_var(self): return self._l_in.input_var class GRUNetwork(object): def __init__(self, name, input_shape, output_dim, hidden_dim, hidden_nonlinearity=tf.nn.relu, gru_layer_cls=L.GRULayer, output_nonlinearity=None, input_var=None, input_layer=None, layer_args=None): with tf.variable_scope(name): if input_layer is None: l_in = L.InputLayer(shape=(None, None) + input_shape, input_var=input_var, name="input") else: l_in = input_layer l_step_input = L.InputLayer(shape=(None,) + input_shape, name="step_input") l_step_prev_state = L.InputLayer(shape=(None, hidden_dim), name="step_prev_state") if layer_args is None: layer_args = dict() l_gru = gru_layer_cls(l_in, num_units=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, hidden_init_trainable=False, name="gru", **layer_args) l_gru_flat = L.ReshapeLayer( l_gru, shape=(-1, hidden_dim), name="gru_flat" ) l_output_flat = L.DenseLayer( l_gru_flat, num_units=output_dim, nonlinearity=output_nonlinearity, name="output_flat" ) l_output = L.OpLayer( l_output_flat, op=lambda flat_output, l_input: tf.reshape(flat_output, tf.stack((tf.shape(l_input)[0], tf.shape(l_input)[1], -1))), shape_op=lambda flat_output_shape, l_input_shape: (l_input_shape[0], l_input_shape[1], flat_output_shape[-1]), extras=[l_in], name="output" ) l_step_state = l_gru.get_step_layer(l_step_input, l_step_prev_state, name="step_state") l_step_hidden = l_step_state l_step_output = L.DenseLayer( l_step_hidden, num_units=output_dim, nonlinearity=output_nonlinearity, W=l_output_flat.W, b=l_output_flat.b, name="step_output" ) self._l_in = l_in self._hid_init_param = l_gru.h0 self._l_gru = l_gru self._l_out = l_output self._l_step_input = l_step_input self._l_step_prev_state = l_step_prev_state self._l_step_hidden = l_step_hidden self._l_step_state = l_step_state self._l_step_output = l_step_output self._hidden_dim = hidden_dim @property def state_dim(self): return self._hidden_dim @property def hidden_dim(self): return self._hidden_dim @property def input_layer(self): return self._l_in @property def input_var(self): return self._l_in.input_var @property def output_layer(self): return self._l_out @property def recurrent_layer(self): return self._l_gru @property def step_input_layer(self): return self._l_step_input @property def step_prev_state_layer(self): return self._l_step_prev_state @property def step_hidden_layer(self): return self._l_step_hidden @property def step_state_layer(self): return self._l_step_state @property def step_output_layer(self): return self._l_step_output @property def hid_init_param(self): return self._hid_init_param @property def state_init_param(self): return self._hid_init_param class LSTMNetwork(object): def __init__(self, name, input_shape, output_dim, hidden_dim, hidden_nonlinearity=tf.nn.relu, lstm_layer_cls=L.LSTMLayer, output_nonlinearity=None, input_var=None, input_layer=None, forget_bias=1.0, use_peepholes=False, layer_args=None): with tf.variable_scope(name): if input_layer is None: l_in = L.InputLayer(shape=(None, None) + input_shape, input_var=input_var, name="input") else: l_in = input_layer l_step_input = L.InputLayer(shape=(None,) + input_shape, name="step_input") # contains previous hidden and cell state l_step_prev_state = L.InputLayer(shape=(None, hidden_dim * 2), name="step_prev_state") if layer_args is None: layer_args = dict() l_lstm = lstm_layer_cls(l_in, num_units=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, hidden_init_trainable=False, name="lstm", forget_bias=forget_bias, cell_init_trainable=False, use_peepholes=use_peepholes, **layer_args) l_lstm_flat = L.ReshapeLayer( l_lstm, shape=(-1, hidden_dim), name="lstm_flat" ) l_output_flat = L.DenseLayer( l_lstm_flat, num_units=output_dim, nonlinearity=output_nonlinearity, name="output_flat" ) l_output = L.OpLayer( l_output_flat, op=lambda flat_output, l_input: tf.reshape(flat_output, tf.stack((tf.shape(l_input)[0], tf.shape(l_input)[1], -1))), shape_op=lambda flat_output_shape, l_input_shape: (l_input_shape[0], l_input_shape[1], flat_output_shape[-1]), extras=[l_in], name="output" ) l_step_state = l_lstm.get_step_layer(l_step_input, l_step_prev_state, name="step_state") l_step_hidden = L.SliceLayer(l_step_state, indices=slice(hidden_dim), name="step_hidden") l_step_cell = L.SliceLayer(l_step_state, indices=slice(hidden_dim, None), name="step_cell") l_step_output = L.DenseLayer( l_step_hidden, num_units=output_dim, nonlinearity=output_nonlinearity, W=l_output_flat.W, b=l_output_flat.b, name="step_output" ) self._l_in = l_in self._hid_init_param = l_lstm.h0 self._cell_init_param = l_lstm.c0 self._l_lstm = l_lstm self._l_out = l_output self._l_step_input = l_step_input self._l_step_prev_state = l_step_prev_state self._l_step_hidden = l_step_hidden self._l_step_cell = l_step_cell self._l_step_state = l_step_state self._l_step_output = l_step_output self._hidden_dim = hidden_dim @property def state_dim(self): return self._hidden_dim * 2 @property def input_layer(self): return self._l_in @property def input_var(self): return self._l_in.input_var @property def output_layer(self): return self._l_out @property def recurrent_layer(self): return self._l_lstm @property def step_input_layer(self): return self._l_step_input @property def step_prev_state_layer(self): return self._l_step_prev_state @property def step_hidden_layer(self): return self._l_step_hidden @property def step_state_layer(self): return self._l_step_state @property def step_cell_layer(self): return self._l_step_cell @property def step_output_layer(self): return self._l_step_output @property def hid_init_param(self): return self._hid_init_param @property def cell_init_param(self): return self._cell_init_param @property def state_init_param(self): return tf.concat(axis=0, values=[self._hid_init_param, self._cell_init_param]) class ConvMergeNetwork(LayersPowered, Serializable): """ This network allows the input to consist of a convolution-friendly component, plus a non-convolution-friendly component. These two components will be concatenated in the fully connected layers. There can also be a list of optional layers for the non-convolution-friendly component alone. The input to the network should be a matrix where each row is a single input entry, with both the aforementioned components flattened out and then concatenated together """ def __init__(self, name, input_shape, extra_input_shape, output_dim, hidden_sizes, conv_filters, conv_filter_sizes, conv_strides, conv_pads, extra_hidden_sizes=None, hidden_W_init=L.XavierUniformInitializer(), hidden_b_init=tf.zeros_initializer(), output_W_init=L.XavierUniformInitializer(), output_b_init=tf.zeros_initializer(), hidden_nonlinearity=tf.nn.relu, output_nonlinearity=None, input_var=None, input_layer=None): Serializable.quick_init(self, locals()) if extra_hidden_sizes is None: extra_hidden_sizes = [] with tf.variable_scope(name): input_flat_dim = np.prod(input_shape) extra_input_flat_dim = np.prod(extra_input_shape) total_input_flat_dim = input_flat_dim + extra_input_flat_dim if input_layer is None: l_in = L.InputLayer(shape=(None, total_input_flat_dim), input_var=input_var, name="input") else: l_in = input_layer l_conv_in = L.reshape( L.SliceLayer( l_in, indices=slice(input_flat_dim), name="conv_slice" ), ([0],) + input_shape, name="conv_reshaped" ) l_extra_in = L.reshape( L.SliceLayer( l_in, indices=slice(input_flat_dim, None), name="extra_slice" ), ([0],) + extra_input_shape, name="extra_reshaped" ) l_conv_hid = l_conv_in for idx, conv_filter, filter_size, stride, pad in zip( range(len(conv_filters)), conv_filters, conv_filter_sizes, conv_strides, conv_pads, ): l_conv_hid = L.Conv2DLayer( l_conv_hid, num_filters=conv_filter, filter_size=filter_size, stride=(stride, stride), pad=pad, nonlinearity=hidden_nonlinearity, name="conv_hidden_%d" % idx, ) l_extra_hid = l_extra_in for idx, hidden_size in enumerate(extra_hidden_sizes): l_extra_hid = L.DenseLayer( l_extra_hid, num_units=hidden_size, nonlinearity=hidden_nonlinearity, name="extra_hidden_%d" % idx, W=hidden_W_init, b=hidden_b_init, ) l_joint_hid = L.concat( [L.flatten(l_conv_hid, name="conv_hidden_flat"), l_extra_hid], name="joint_hidden" ) for idx, hidden_size in enumerate(hidden_sizes): l_joint_hid = L.DenseLayer( l_joint_hid, num_units=hidden_size, nonlinearity=hidden_nonlinearity, name="joint_hidden_%d" % idx, W=hidden_W_init, b=hidden_b_init, ) l_out = L.DenseLayer( l_joint_hid, num_units=output_dim, nonlinearity=output_nonlinearity, name="output", W=output_W_init, b=output_b_init, ) self._l_in = l_in self._l_out = l_out LayersPowered.__init__(self, [l_out], input_layers=[l_in]) @property def input_layer(self): return self._l_in @property def output_layer(self): return self._l_out @property def input_var(self): return self._l_in.input_var
19,707
35.837383
155
py
rllab
rllab-master/sandbox/rocky/tf/core/layers.py
# -*- coding: utf-8 -*- import functools import numpy as np import math import tensorflow as tf from tensorflow.python.training import moving_averages from collections import OrderedDict from collections import deque from itertools import chain from inspect import getargspec from difflib import get_close_matches from warnings import warn class G(object): pass G._n_layers = 0 def create_param(spec, shape, name, trainable=True, regularizable=True): if not hasattr(spec, '__call__'): assert isinstance(spec, (tf.Tensor, tf.Variable)) return spec assert hasattr(spec, '__call__') if regularizable: # use the default regularizer regularizer = None else: # do not regularize this variable regularizer = lambda _: tf.constant(0.) return tf.get_variable( name=name, shape=shape, initializer=spec, trainable=trainable, regularizer=regularizer, dtype=tf.float32 ) def as_tuple(x, N, t=None): try: X = tuple(x) except TypeError: X = (x,) * N if (t is not None) and not all(isinstance(v, t) for v in X): raise TypeError("expected a single value or an iterable " "of {0}, got {1} instead".format(t.__name__, x)) if len(X) != N: raise ValueError("expected a single value or an iterable " "with length {0}, got {1} instead".format(N, x)) return X def conv_output_length(input_length, filter_size, stride, pad=0): """Helper function to compute the output size of a convolution operation This function computes the length along a single axis, which corresponds to a 1D convolution. It can also be used for convolutions with higher dimensionalities by using it individually for each axis. Parameters ---------- input_length : int or None The size of the input. filter_size : int The size of the filter. stride : int The stride of the convolution operation. pad : int, 'full' or 'same' (default: 0) By default, the convolution is only computed where the input and the filter fully overlap (a valid convolution). When ``stride=1``, this yields an output that is smaller than the input by ``filter_size - 1``. The `pad` argument allows you to implicitly pad the input with zeros, extending the output size. A single integer results in symmetric zero-padding of the given size on both borders. ``'full'`` pads with one less than the filter size on both sides. This is equivalent to computing the convolution wherever the input and the filter overlap by at least one position. ``'same'`` pads with half the filter size on both sides (one less on the second side for an even filter size). When ``stride=1``, this results in an output size equal to the input size. Returns ------- int or None The output size corresponding to the given convolution parameters, or ``None`` if `input_size` is ``None``. Raises ------ ValueError When an invalid padding is specified, a `ValueError` is raised. """ if input_length is None: return None if pad == 'valid': output_length = input_length - filter_size + 1 elif pad == 'full': output_length = input_length + filter_size - 1 elif pad == 'same': output_length = input_length elif isinstance(pad, int): output_length = input_length + 2 * pad - filter_size + 1 else: raise ValueError('Invalid pad: {0}'.format(pad)) # This is the integer arithmetic equivalent to # np.ceil(output_length / stride) output_length = (output_length + stride - 1) // stride return output_length class Layer(object): def __init__(self, incoming, name=None, variable_reuse=None, weight_normalization=False, **kwargs): if isinstance(incoming, tuple): self.input_shape = incoming self.input_layer = None else: self.input_shape = incoming.output_shape self.input_layer = incoming self.params = OrderedDict() self.weight_normalization = weight_normalization if name is None: name = "%s_%d" % (type(self).__name__, G._n_layers) G._n_layers += 1 self.name = name self.variable_reuse = variable_reuse self.get_output_kwargs = [] if any(d is not None and d <= 0 for d in self.input_shape): raise ValueError(( "Cannot create Layer with a non-positive input_shape " "dimension. input_shape=%r, self.name=%r") % ( self.input_shape, self.name)) @property def output_shape(self): shape = self.get_output_shape_for(self.input_shape) if any(isinstance(s, (tf.Variable, tf.Tensor)) for s in shape): raise ValueError("%s returned a symbolic output shape from its " "get_output_shape_for() method: %r. This is not " "allowed; shapes must be tuples of integers for " "fixed-size dimensions and Nones for variable " "dimensions." % (self.__class__.__name__, shape)) return shape def get_output_shape_for(self, input_shape): raise NotImplementedError def get_output_for(self, input, **kwargs): raise NotImplementedError def add_param_plain(self, spec, shape, name, **tags): with tf.variable_scope(self.name, reuse=self.variable_reuse): tags['trainable'] = tags.get('trainable', True) tags['regularizable'] = tags.get('regularizable', True) param = create_param(spec, shape, name, **tags) self.params[param] = set(tag for tag, value in list(tags.items()) if value) return param def add_param(self, spec, shape, name, **kwargs): param = self.add_param_plain(spec, shape, name, **kwargs) if name is not None and name.startswith("W") and self.weight_normalization: # Hacky: check if the parameter is a weight matrix. If so, apply weight normalization if len(param.get_shape()) == 2: v = param g = self.add_param_plain(tf.ones_initializer(), (shape[1],), name=name + "_wn/g") param = v * (tf.reshape(g, (1, -1)) / tf.sqrt(tf.reduce_sum(tf.square(v), 0, keep_dims=True))) elif len(param.get_shape()) == 4: v = param g = self.add_param_plain(tf.ones_initializer(), (shape[3],), name=name + "_wn/g") param = v * (tf.reshape(g, (1, 1, 1, -1)) / tf.sqrt(tf.reduce_sum(tf.square(v), [0, 1, 2], keep_dims=True))) else: raise NotImplementedError return param def get_params(self, **tags): result = list(self.params.keys()) only = set(tag for tag, value in list(tags.items()) if value) if only: # retain all parameters that have all of the tags in `only` result = [param for param in result if not (only - self.params[param])] exclude = set(tag for tag, value in list(tags.items()) if not value) if exclude: # retain all parameters that have none of the tags in `exclude` result = [param for param in result if not (self.params[param] & exclude)] return result class InputLayer(Layer): def __init__(self, shape, input_var=None, **kwargs): super(InputLayer, self).__init__(shape, **kwargs) self.shape = shape if input_var is None: if self.name is not None: with tf.variable_scope(self.name): input_var = tf.placeholder(tf.float32, shape=shape, name="input") else: input_var = tf.placeholder(tf.float32, shape=shape, name="input") self.input_var = input_var @Layer.output_shape.getter def output_shape(self): return self.shape class MergeLayer(Layer): def __init__(self, incomings, name=None, **kwargs): self.input_shapes = [incoming if isinstance(incoming, tuple) else incoming.output_shape for incoming in incomings] self.input_layers = [None if isinstance(incoming, tuple) else incoming for incoming in incomings] self.name = name self.params = OrderedDict() self.get_output_kwargs = [] @Layer.output_shape.getter def output_shape(self): shape = self.get_output_shape_for(self.input_shapes) if any(isinstance(s, (tf.Variable, tf.Tensor)) for s in shape): raise ValueError("%s returned a symbolic output shape from its " "get_output_shape_for() method: %r. This is not " "allowed; shapes must be tuples of integers for " "fixed-size dimensions and Nones for variable " "dimensions." % (self.__class__.__name__, shape)) return shape def get_output_shape_for(self, input_shapes): raise NotImplementedError def get_output_for(self, inputs, **kwargs): raise NotImplementedError class ConcatLayer(MergeLayer): """ Concatenates multiple inputs along the specified axis. Inputs should have the same shape except for the dimension specified in axis, which can have different sizes. Parameters ----------- incomings : a list of :class:`Layer` instances or tuples The layers feeding into this layer, or expected input shapes axis : int Axis which inputs are joined over """ def __init__(self, incomings, axis=1, **kwargs): super(ConcatLayer, self).__init__(incomings, **kwargs) self.axis = axis def get_output_shape_for(self, input_shapes): # Infer the output shape by grabbing, for each axis, the first # input size that is not `None` (if there is any) output_shape = [next((s for s in sizes if s is not None), None) for sizes in zip(*input_shapes)] def match(shape1, shape2): return (len(shape1) == len(shape2) and all(i == self.axis or s1 is None or s2 is None or s1 == s2 for i, (s1, s2) in enumerate(zip(shape1, shape2)))) # Check for compatibility with inferred output shape if not all(match(shape, output_shape) for shape in input_shapes): raise ValueError("Mismatch: input shapes must be the same except " "in the concatenation axis") # Infer output shape on concatenation axis and return sizes = [input_shape[self.axis] for input_shape in input_shapes] concat_size = None if any(s is None for s in sizes) else sum(sizes) output_shape[self.axis] = concat_size return tuple(output_shape) def get_output_for(self, inputs, **kwargs): dtypes = [x.dtype.as_numpy_dtype for x in inputs] if len(set(dtypes)) > 1: # need to convert to common data type common_dtype = np.core.numerictypes.find_common_type([], dtypes) inputs = [tf.cast(x, common_dtype) for x in inputs] return tf.concat(axis=self.axis, values=inputs) concat = ConcatLayer # shortcut class XavierUniformInitializer(object): def __call__(self, shape, dtype=tf.float32, *args, **kwargs): if len(shape) == 2: n_inputs, n_outputs = shape else: receptive_field_size = np.prod(shape[:2]) n_inputs = shape[-2] * receptive_field_size n_outputs = shape[-1] * receptive_field_size init_range = math.sqrt(6.0 / (n_inputs + n_outputs)) return tf.random_uniform_initializer(-init_range, init_range, dtype=dtype)(shape) class HeUniformInitializer(object): def __call__(self, shape, dtype=tf.float32, *args, **kwargs): if len(shape) == 2: n_inputs, _ = shape else: receptive_field_size = np.prod(shape[:2]) n_inputs = shape[-2] * receptive_field_size init_range = math.sqrt(1.0 / n_inputs) return tf.random_uniform_initializer(-init_range, init_range, dtype=dtype)(shape) def py_ortho_init(scale): def _init(shape): u, s, v = np.linalg.svd(np.random.uniform(size=shape)) return np.cast['float32'](u * scale) return _init class OrthogonalInitializer(object): def __init__(self, scale=1.1): self.scale = scale def __call__(self, shape, dtype=tf.float32, *args, **kwargs): result, = tf.py_func(py_ortho_init(self.scale), [shape], [tf.float32]) result.set_shape(shape) return result class ParamLayer(Layer): def __init__(self, incoming, num_units, param=tf.zeros_initializer(), trainable=True, **kwargs): super(ParamLayer, self).__init__(incoming, **kwargs) self.num_units = num_units self.param = self.add_param( param, (num_units,), name="param", trainable=trainable ) def get_output_shape_for(self, input_shape): return input_shape[:-1] + (self.num_units,) def get_output_for(self, input, **kwargs): ndim = input.get_shape().ndims reshaped_param = tf.reshape(self.param, (1,) * (ndim - 1) + (self.num_units,)) tile_arg = tf.concat(axis=0, values=[tf.shape(input)[:ndim - 1], [1]]) tiled = tf.tile(reshaped_param, tile_arg) return tiled class OpLayer(MergeLayer): def __init__(self, incoming, op, shape_op=lambda x: x, extras=None, **kwargs): if extras is None: extras = [] incomings = [incoming] + extras super(OpLayer, self).__init__(incomings, **kwargs) self.op = op self.shape_op = shape_op self.incomings = incomings def get_output_shape_for(self, input_shapes): return self.shape_op(*input_shapes) def get_output_for(self, inputs, **kwargs): return self.op(*inputs) class DenseLayer(Layer): def __init__(self, incoming, num_units, nonlinearity=None, W=XavierUniformInitializer(), b=tf.zeros_initializer(), **kwargs): super(DenseLayer, self).__init__(incoming, **kwargs) self.nonlinearity = tf.identity if nonlinearity is None else nonlinearity self.num_units = num_units num_inputs = int(np.prod(self.input_shape[1:])) self.W = self.add_param(W, (num_inputs, num_units), name="W") if b is None: self.b = None else: self.b = self.add_param(b, (num_units,), name="b", regularizable=False) def get_output_shape_for(self, input_shape): return (input_shape[0], self.num_units) def get_output_for(self, input, **kwargs): if input.get_shape().ndims > 2: # if the input has more than two dimensions, flatten it into a # batch of feature vectors. input = tf.reshape(input, tf.stack([tf.shape(input)[0], -1])) activation = tf.matmul(input, self.W) if self.b is not None: activation = activation + tf.expand_dims(self.b, 0) return self.nonlinearity(activation) class BaseConvLayer(Layer): def __init__(self, incoming, num_filters, filter_size, stride=1, pad="VALID", untie_biases=False, W=XavierUniformInitializer(), b=tf.zeros_initializer(), nonlinearity=tf.nn.relu, n=None, **kwargs): """ Input is assumed to be of shape batch*height*width*channels """ super(BaseConvLayer, self).__init__(incoming, **kwargs) if nonlinearity is None: self.nonlinearity = tf.identity else: self.nonlinearity = nonlinearity if n is None: n = len(self.input_shape) - 2 elif n != len(self.input_shape) - 2: raise ValueError("Tried to create a %dD convolution layer with " "input shape %r. Expected %d input dimensions " "(batchsize, channels, %d spatial dimensions)." % (n, self.input_shape, n + 2, n)) self.n = n self.num_filters = num_filters self.filter_size = as_tuple(filter_size, n, int) self.stride = as_tuple(stride, n, int) self.untie_biases = untie_biases self.pad = pad if pad == 'SAME': if any(s % 2 == 0 for s in self.filter_size): raise NotImplementedError( '`same` padding requires odd filter size.') self.W = self.add_param(W, self.get_W_shape(), name="W") if b is None: self.b = None else: if self.untie_biases: biases_shape = self.output_shape[1:3] + (num_filters,) # + self.output_shape[2:] else: biases_shape = (num_filters,) self.b = self.add_param(b, biases_shape, name="b", regularizable=False) def get_W_shape(self): """Get the shape of the weight matrix `W`. Returns ------- tuple of int The shape of the weight matrix. """ num_input_channels = self.input_shape[-1] return self.filter_size + (num_input_channels, self.num_filters) def get_output_shape_for(self, input_shape): if self.pad == 'SAME': pad = ('same',) * self.n elif self.pad == 'VALID': pad = (0,) * self.n else: import ipdb; ipdb.set_trace() raise NotImplementedError # pad = self.pad if isinstance(self.pad, tuple) else (self.pad,) * self.n batchsize = input_shape[0] return ((batchsize,) + tuple(conv_output_length(input, filter, stride, p) for input, filter, stride, p in zip(input_shape[1:3], self.filter_size, self.stride, pad))) + (self.num_filters,) def get_output_for(self, input, **kwargs): conved = self.convolve(input, **kwargs) if self.b is None: activation = conved elif self.untie_biases: # raise NotImplementedError activation = conved + tf.expand_dims(self.b, 0) else: activation = conved + tf.reshape(self.b, (1, 1, 1, self.num_filters)) return self.nonlinearity(activation) def convolve(self, input, **kwargs): """ Symbolically convolves `input` with ``self.W``, producing an output of shape ``self.output_shape``. To be implemented by subclasses. Parameters ---------- input : Theano tensor The input minibatch to convolve **kwargs Any additional keyword arguments from :meth:`get_output_for` Returns ------- Theano tensor `input` convolved according to the configuration of this layer, without any bias or nonlinearity applied. """ raise NotImplementedError("BaseConvLayer does not implement the " "convolve() method. You will want to " "use a subclass such as Conv2DLayer.") class Conv2DLayer(BaseConvLayer): def __init__(self, incoming, num_filters, filter_size, stride=(1, 1), pad="VALID", untie_biases=False, W=XavierUniformInitializer(), b=tf.zeros_initializer(), nonlinearity=tf.nn.relu, convolution=tf.nn.conv2d, **kwargs): super(Conv2DLayer, self).__init__(incoming=incoming, num_filters=num_filters, filter_size=filter_size, stride=stride, pad=pad, untie_biases=untie_biases, W=W, b=b, nonlinearity=nonlinearity, n=2, **kwargs) self.convolution = convolution def convolve(self, input, **kwargs): conved = self.convolution(input, self.W, strides=(1,) + self.stride + (1,), padding=self.pad) return conved def pool_output_length(input_length, pool_size, stride, pad): if input_length is None or pool_size is None: return None if pad == "SAME": return int(np.ceil(float(input_length) / float(stride))) return int(np.ceil(float(input_length - pool_size + 1) / float(stride))) class Pool2DLayer(Layer): def __init__(self, incoming, pool_size, stride=None, pad="VALID", mode='max', **kwargs): super(Pool2DLayer, self).__init__(incoming, **kwargs) self.pool_size = as_tuple(pool_size, 2) if len(self.input_shape) != 4: raise ValueError("Tried to create a 2D pooling layer with " "input shape %r. Expected 4 input dimensions " "(batchsize, 2 spatial dimensions, channels)." % (self.input_shape,)) if stride is None: self.stride = self.pool_size else: self.stride = as_tuple(stride, 2) self.pad = pad self.mode = mode def get_output_shape_for(self, input_shape): output_shape = list(input_shape) # copy / convert to mutable list output_shape[1] = pool_output_length(input_shape[1], pool_size=self.pool_size[0], stride=self.stride[0], pad=self.pad, ) output_shape[2] = pool_output_length(input_shape[2], pool_size=self.pool_size[1], stride=self.stride[1], pad=self.pad, ) return tuple(output_shape) def get_output_for(self, input, **kwargs): assert self.mode == "max" pooled = tf.nn.max_pool( input, ksize=(1,) + self.pool_size + (1,), strides=(1,) + self.stride + (1,), padding=self.pad, ) return pooled def spatial_expected_softmax(x, temp=1): assert len(x.get_shape()) == 4 vals = [] for dim in [0, 1]: dim_val = x.get_shape()[dim + 1].value lin = tf.linspace(-1.0, 1.0, dim_val) lin = tf.expand_dims(lin, 1 - dim) lin = tf.expand_dims(lin, 0) lin = tf.expand_dims(lin, 3) m = tf.reduce_max(x, [1, 2], keep_dims=True) e = tf.exp((x - m) / temp) + 1e-5 val = tf.reduce_sum(e * lin, [1, 2]) / (tf.reduce_sum(e, [1, 2])) vals.append(tf.expand_dims(val, 2)) return tf.reshape(tf.concat(axis=2, values=vals), [-1, x.get_shape()[-1].value * 2]) class SpatialExpectedSoftmaxLayer(Layer): """ Computes the softmax across a spatial region, separately for each channel, followed by an expectation operation. """ def __init__(self, incoming, **kwargs): super().__init__(incoming, **kwargs) # self.temp = self.add_param(tf.ones_initializer, shape=(), name="temperature") def get_output_shape_for(self, input_shape): return (input_shape[0], input_shape[-1] * 2) def get_output_for(self, input, **kwargs): return spatial_expected_softmax(input)#, self.temp) # max_ = tf.reduce_max(input, reduction_indices=[1, 2], keep_dims=True) # exp = tf.exp(input - max_) + 1e-5 # vals = [] # # for dim in [0, 1]: # dim_val = input.get_shape()[dim + 1].value # lin = tf.linspace(-1.0, 1.0, dim_val) # lin = tf.expand_dims(lin, 1 - dim) # lin = tf.expand_dims(lin, 0) # lin = tf.expand_dims(lin, 3) # m = tf.reduce_max(input, [1, 2], keep_dims=True) # e = tf.exp(input - m) + 1e-5 # val = tf.reduce_sum(e * lin, [1, 2]) / (tf.reduce_sum(e, [1, 2])) # vals.append(tf.expand_dims(val, 2)) # # return tf.reshape(tf.concat(2, vals), [-1, input.get_shape()[-1].value * 2]) # import ipdb; ipdb.set_trace() # input.get_shape() # exp / tf.reduce_sum(exp, reduction_indices=[1, 2], keep_dims=True) # import ipdb; # ipdb.set_trace() # spatial softmax? # for dim in range(2): # val = obs.get_shape()[dim + 1].value # lin = tf.linspace(-1.0, 1.0, val) # lin = tf.expand_dims(lin, 1 - dim) # lin = tf.expand_dims(lin, 0) # lin = tf.expand_dims(lin, 3) # m = tf.reduce_max(e, [1, 2], keep_dims=True) # e = tf.exp(e - m) + 1e-3 # val = tf.reduce_sum(e * lin, [1, 2]) / (tf.reduce_sum(e, [1, 2])) class DropoutLayer(Layer): def __init__(self, incoming, p=0.5, rescale=True, **kwargs): super(DropoutLayer, self).__init__(incoming, **kwargs) self.p = p self.rescale = rescale def get_output_for(self, input, deterministic=False, **kwargs): """ Parameters ---------- input : tensor output from the previous layer deterministic : bool If true dropout and scaling is disabled, see notes """ if deterministic or self.p == 0: return input else: # Using theano constant to prevent upcasting # one = T.constant(1) retain_prob = 1. - self.p if self.rescale: input /= retain_prob # use nonsymbolic shape for dropout mask if possible return tf.nn.dropout(input, keep_prob=retain_prob) def get_output_shape_for(self, input_shape): return input_shape # TODO: add Conv3DLayer class FlattenLayer(Layer): """ A layer that flattens its input. The leading ``outdim-1`` dimensions of the output will have the same shape as the input. The remaining dimensions are collapsed into the last dimension. Parameters ---------- incoming : a :class:`Layer` instance or a tuple The layer feeding into this layer, or the expected input shape. outdim : int The number of dimensions in the output. See Also -------- flatten : Shortcut """ def __init__(self, incoming, outdim=2, **kwargs): super(FlattenLayer, self).__init__(incoming, **kwargs) self.outdim = outdim if outdim < 1: raise ValueError('Dim must be >0, was %i', outdim) def get_output_shape_for(self, input_shape): to_flatten = input_shape[self.outdim - 1:] if any(s is None for s in to_flatten): flattened = None else: flattened = int(np.prod(to_flatten)) return input_shape[:self.outdim - 1] + (flattened,) def get_output_for(self, input, **kwargs): # total_entries = tf.reduce_prod(tf.shape(input)) pre_shape = tf.shape(input)[:self.outdim - 1] to_flatten = tf.reduce_prod(tf.shape(input)[self.outdim - 1:]) return tf.reshape(input, tf.concat(axis=0, values=[pre_shape, tf.stack([to_flatten])])) flatten = FlattenLayer # shortcut class ReshapeLayer(Layer): def __init__(self, incoming, shape, **kwargs): super(ReshapeLayer, self).__init__(incoming, **kwargs) shape = tuple(shape) for s in shape: if isinstance(s, int): if s == 0 or s < - 1: raise ValueError("`shape` integers must be positive or -1") elif isinstance(s, list): if len(s) != 1 or not isinstance(s[0], int) or s[0] < 0: raise ValueError("`shape` input references must be " "single-element lists of int >= 0") elif isinstance(s, (tf.Tensor, tf.Variable)): # T.TensorVariable): raise NotImplementedError # if s.ndim != 0: # raise ValueError( # "A symbolic variable in a shape specification must be " # "a scalar, but had %i dimensions" % s.ndim) else: raise ValueError("`shape` must be a tuple of int and/or [int]") if sum(s == -1 for s in shape) > 1: raise ValueError("`shape` cannot contain multiple -1") self.shape = shape # try computing the output shape once as a sanity check self.get_output_shape_for(self.input_shape) def get_output_shape_for(self, input_shape, **kwargs): # Initialize output shape from shape specification output_shape = list(self.shape) # First, replace all `[i]` with the corresponding input dimension, and # mask parts of the shapes thus becoming irrelevant for -1 inference masked_input_shape = list(input_shape) masked_output_shape = list(output_shape) for dim, o in enumerate(output_shape): if isinstance(o, list): if o[0] >= len(input_shape): raise ValueError("specification contains [%d], but input " "shape has %d dimensions only" % (o[0], len(input_shape))) output_shape[dim] = input_shape[o[0]] masked_output_shape[dim] = input_shape[o[0]] if (input_shape[o[0]] is None) \ and (masked_input_shape[o[0]] is None): # first time we copied this unknown input size: mask # it, we have a 1:1 correspondence between out[dim] and # in[o[0]] and can ignore it for -1 inference even if # it is unknown. masked_input_shape[o[0]] = 1 masked_output_shape[dim] = 1 # Secondly, replace all symbolic shapes with `None`, as we cannot # infer their size here. for dim, o in enumerate(output_shape): if isinstance(o, (tf.Tensor, tf.Variable)): # T.TensorVariable): raise NotImplementedError # output_shape[dim] = None # masked_output_shape[dim] = None # From the shapes, compute the sizes of the input and output tensor input_size = (None if any(x is None for x in masked_input_shape) else np.prod(masked_input_shape)) output_size = (None if any(x is None for x in masked_output_shape) else np.prod(masked_output_shape)) del masked_input_shape, masked_output_shape # Finally, infer value for -1 if needed if -1 in output_shape: dim = output_shape.index(-1) if (input_size is None) or (output_size is None): output_shape[dim] = None output_size = None else: output_size *= -1 output_shape[dim] = input_size // output_size output_size *= output_shape[dim] # Sanity check if (input_size is not None) and (output_size is not None) \ and (input_size != output_size): raise ValueError("%s cannot be reshaped to specification %s. " "The total size mismatches." % (input_shape, self.shape)) return tuple(output_shape) def get_output_for(self, input, **kwargs): # Replace all `[i]` with the corresponding input dimension output_shape = list(self.shape) for dim, o in enumerate(output_shape): if isinstance(o, list): output_shape[dim] = tf.shape(input)[o[0]] # Everything else is handled by Theano return tf.reshape(input, tf.stack(output_shape)) reshape = ReshapeLayer # shortcut class SliceLayer(Layer): def __init__(self, incoming, indices, axis=-1, **kwargs): super(SliceLayer, self).__init__(incoming, **kwargs) self.slice = indices self.axis = axis def get_output_shape_for(self, input_shape): output_shape = list(input_shape) if isinstance(self.slice, int): del output_shape[self.axis] elif input_shape[self.axis] is not None: output_shape[self.axis] = len( list(range(*self.slice.indices(input_shape[self.axis])))) else: output_shape[self.axis] = None return tuple(output_shape) def get_output_for(self, input, **kwargs): axis = self.axis ndims = input.get_shape().ndims if axis < 0: axis += ndims if isinstance(self.slice, int) and self.slice < 0: return tf.reverse(input, [self.axis + 1])[ (slice(None),) * axis + (-1 - self.slice,) + (slice(None),) * (ndims - axis - 1) ] # import ipdb; ipdb.set_trace() return input[(slice(None),) * axis + (self.slice,) + (slice(None),) * (ndims - axis - 1)] class DimshuffleLayer(Layer): def __init__(self, incoming, pattern, **kwargs): super(DimshuffleLayer, self).__init__(incoming, **kwargs) # Sanity check the pattern used_dims = set() for p in pattern: if isinstance(p, int): # Dimension p if p in used_dims: raise ValueError("pattern contains dimension {0} more " "than once".format(p)) used_dims.add(p) elif p == 'x': # Broadcast pass else: raise ValueError("pattern should only contain dimension" "indices or 'x', not {0}".format(p)) self.pattern = pattern # try computing the output shape once as a sanity check self.get_output_shape_for(self.input_shape) def get_output_shape_for(self, input_shape): # Build output shape while keeping track of the dimensions that we are # attempting to collapse, so we can ensure that they are broadcastable output_shape = [] dims_used = [False] * len(input_shape) for p in self.pattern: if isinstance(p, int): if p < 0 or p >= len(input_shape): raise ValueError("pattern contains {0}, but input shape " "has {1} dimensions " "only".format(p, len(input_shape))) # Dimension p o = input_shape[p] dims_used[p] = True elif p == 'x': # Broadcast; will be of size 1 o = 1 output_shape.append(o) for i, (dim_size, used) in enumerate(zip(input_shape, dims_used)): if not used and dim_size != 1 and dim_size is not None: raise ValueError( "pattern attempted to collapse dimension " "{0} of size {1}; dimensions with size != 1/None are not" "broadcastable and cannot be " "collapsed".format(i, dim_size)) return tuple(output_shape) def get_output_for(self, input, **kwargs): return tf.transpose(input, self.pattern) dimshuffle = DimshuffleLayer # shortcut def apply_ln(layer): def _normalize(x, prefix): EPS = 1e-5 dim = x.get_shape()[-1].value bias_name = prefix + "_ln/bias" scale_name = prefix + "_ln/scale" if bias_name not in layer.norm_params: layer.norm_params[bias_name] = layer.add_param( tf.zeros_initializer(), (dim,), name=bias_name, regularizable=False) if scale_name not in layer.norm_params: layer.norm_params[scale_name] = layer.add_param( tf.ones_initializer(), (dim,), name=scale_name) bias = layer.norm_params[bias_name] scale = layer.norm_params[scale_name] mean, var = tf.nn.moments(x, axes=[1], keep_dims=True) x_normed = (x - mean) / tf.sqrt(var + EPS) return x_normed * scale + bias return _normalize class GRULayer(Layer): """ A gated recurrent unit implements the following update mechanism: Reset gate: r(t) = f_r(x(t) @ W_xr + h(t-1) @ W_hr + b_r) Update gate: u(t) = f_u(x(t) @ W_xu + h(t-1) @ W_hu + b_u) Cell gate: c(t) = f_c(x(t) @ W_xc + r(t) * (h(t-1) @ W_hc) + b_c) New hidden state: h(t) = (1 - u(t)) * h(t-1) + u_t * c(t) Note that the reset, update, and cell vectors must have the same dimension as the hidden state """ def __init__(self, incoming, num_units, hidden_nonlinearity, gate_nonlinearity=tf.nn.sigmoid, W_x_init=XavierUniformInitializer(), W_h_init=OrthogonalInitializer(), b_init=tf.zeros_initializer(), hidden_init=tf.zeros_initializer(), hidden_init_trainable=False, layer_normalization=False, **kwargs): if hidden_nonlinearity is None: hidden_nonlinearity = tf.identity if gate_nonlinearity is None: gate_nonlinearity = tf.identity super(GRULayer, self).__init__(incoming, **kwargs) input_shape = self.input_shape[2:] input_dim = np.prod(input_shape) self.layer_normalization = layer_normalization # Weights for the initial hidden state self.h0 = self.add_param(hidden_init, (num_units,), name="h0", trainable=hidden_init_trainable, regularizable=False) # Weights for the reset gate self.W_xr = self.add_param(W_x_init, (input_dim, num_units), name="W_xr") self.W_hr = self.add_param(W_h_init, (num_units, num_units), name="W_hr") self.b_r = self.add_param(b_init, (num_units,), name="b_r", regularizable=False) # Weights for the update gate self.W_xu = self.add_param(W_x_init, (input_dim, num_units), name="W_xu") self.W_hu = self.add_param(W_h_init, (num_units, num_units), name="W_hu") self.b_u = self.add_param(b_init, (num_units,), name="b_u", regularizable=False) # Weights for the cell gate self.W_xc = self.add_param(W_x_init, (input_dim, num_units), name="W_xc") self.W_hc = self.add_param(W_h_init, (num_units, num_units), name="W_hc") self.b_c = self.add_param(b_init, (num_units,), name="b_c", regularizable=False) self.W_x_ruc = tf.concat(axis=1, values=[self.W_xr, self.W_xu, self.W_xc]) self.W_h_ruc = tf.concat(axis=1, values=[self.W_hr, self.W_hu, self.W_hc]) self.W_x_ru = tf.concat(axis=1, values=[self.W_xr, self.W_xu]) self.W_h_ru = tf.concat(axis=1, values=[self.W_hr, self.W_hu]) self.b_ruc = tf.concat(axis=0, values=[self.b_r, self.b_u, self.b_c]) self.gate_nonlinearity = gate_nonlinearity self.num_units = num_units self.nonlinearity = hidden_nonlinearity self.norm_params = dict() # pre-run the step method to initialize the normalization parameters h_dummy = tf.placeholder(dtype=tf.float32, shape=(None, num_units), name="h_dummy") x_dummy = tf.placeholder(dtype=tf.float32, shape=(None, input_dim), name="x_dummy") self.step(h_dummy, x_dummy) def step(self, hprev, x): if self.layer_normalization: ln = apply_ln(self) x_ru = ln(tf.matmul(x, self.W_x_ru), "x_ru") h_ru = ln(tf.matmul(hprev, self.W_h_ru), "h_ru") x_r, x_u = tf.split(axis=1, num_or_size_splits=2, value=x_ru) h_r, h_u = tf.split(axis=1, num_or_size_splits=2, value=h_ru) x_c = ln(tf.matmul(x, self.W_xc), "x_c") h_c = ln(tf.matmul(hprev, self.W_hc), "h_c") r = self.gate_nonlinearity(x_r + h_r) u = self.gate_nonlinearity(x_u + h_u) c = self.nonlinearity(x_c + r * h_c) h = (1 - u) * hprev + u * c return h else: xb_ruc = tf.matmul(x, self.W_x_ruc) + tf.reshape(self.b_ruc, (1, -1)) h_ruc = tf.matmul(hprev, self.W_h_ruc) xb_r, xb_u, xb_c = tf.split(axis=1, num_or_size_splits=3, value=xb_ruc) h_r, h_u, h_c = tf.split(axis=1, num_or_size_splits=3, value=h_ruc) r = self.gate_nonlinearity(xb_r + h_r) u = self.gate_nonlinearity(xb_u + h_u) c = self.nonlinearity(xb_c + r * h_c) h = (1 - u) * hprev + u * c return h def get_step_layer(self, l_in, l_prev_hidden, name=None): return GRUStepLayer(incomings=[l_in, l_prev_hidden], recurrent_layer=self, name=name) def get_output_shape_for(self, input_shape): n_batch, n_steps = input_shape[:2] return n_batch, n_steps, self.num_units def get_output_for(self, input, **kwargs): input_shape = tf.shape(input) n_batches = input_shape[0] n_steps = input_shape[1] input = tf.reshape(input, tf.stack([n_batches, n_steps, -1])) if 'recurrent_state' in kwargs and self in kwargs['recurrent_state']: h0s = kwargs['recurrent_state'][self] else: h0s = tf.tile( tf.reshape(self.h0, (1, self.num_units)), (n_batches, 1) ) # flatten extra dimensions shuffled_input = tf.transpose(input, (1, 0, 2)) hs = tf.scan( self.step, elems=shuffled_input, initializer=h0s ) shuffled_hs = tf.transpose(hs, (1, 0, 2)) if 'recurrent_state_output' in kwargs: kwargs['recurrent_state_output'][self] = shuffled_hs return shuffled_hs class GRUStepLayer(MergeLayer): def __init__(self, incomings, recurrent_layer, **kwargs): super(GRUStepLayer, self).__init__(incomings, **kwargs) self._gru_layer = recurrent_layer def get_params(self, **tags): return self._gru_layer.get_params(**tags) def get_output_shape_for(self, input_shapes): n_batch = input_shapes[0][0] return n_batch, self._gru_layer.num_units def get_output_for(self, inputs, **kwargs): x, hprev = inputs n_batch = tf.shape(x)[0] x = tf.reshape(x, tf.stack([n_batch, -1])) x.set_shape((None, self.input_shapes[0][1])) return self._gru_layer.step(hprev, x) class TfGRULayer(Layer): """ Use TensorFlow's built-in GRU implementation """ def __init__(self, incoming, num_units, hidden_nonlinearity, horizon=None, hidden_init_trainable=False, **kwargs): assert len(incoming.output_shape) == 3 input_dim = incoming.shape[2] gru = tf.nn.rnn_cell.GRUCell(num_units=num_units, activation=hidden_nonlinearity) self.num_units = num_units self.horizon = horizon self.gru = gru self.hidden_nonlinearity = hidden_nonlinearity Layer.__init__(self, incoming=incoming, **kwargs) # dummy input variable input_dummy = tf.placeholder(tf.float32, (None, input_dim), "input_dummy") hidden_dummy = tf.placeholder(tf.float32, (None, num_units), "hidden_dummy") with tf.variable_scope(self.name) as vs: gru(input_dummy, hidden_dummy, scope=vs) vs.reuse_variables() self.scope = vs all_vars = [v for v in tf.global_variables() if v.name.startswith(vs.name)] trainable_vars = [v for v in tf.trainable_variables() if v.name.startswith(vs.name)] for var in trainable_vars: self.add_param(spec=var, shape=None, name=None, trainable=True) for var in set(all_vars) - set(trainable_vars): self.add_param(spec=var, shape=None, name=None, trainable=False) self.h0 = self.add_param(tf.zeros_initializer(), (num_units,), name="h0", trainable=hidden_init_trainable, regularizable=False) def step(self, hprev, x): return self.gru(x, hprev, scope=self.scope)[1] def get_output_for(self, input, **kwargs): input_shape = tf.shape(input) n_batches = input_shape[0] state = tf.tile( tf.reshape(self.h0, (1, self.num_units)), (n_batches, 1) ) state.set_shape((None, self.num_units)) if self.horizon is not None: outputs = [] for idx in range(self.horizon): output, state = self.gru(input[:, idx, :], state, scope=self.scope) # self.name) outputs.append(tf.expand_dims(output, 1)) outputs = tf.concat(axis=1, values=outputs) return outputs else: n_steps = input_shape[1] input = tf.reshape(input, tf.stack([n_batches, n_steps, -1])) # flatten extra dimensions shuffled_input = tf.transpose(input, (1, 0, 2)) shuffled_input.set_shape((None, None, self.input_shape[-1])) hs = tf.scan( self.step, elems=shuffled_input, initializer=state ) shuffled_hs = tf.transpose(hs, (1, 0, 2)) return shuffled_hs def get_output_shape_for(self, input_shape): n_batch, n_steps = input_shape[:2] return n_batch, n_steps, self.num_units def get_step_layer(self, l_in, l_prev_hidden, name=None): return GRUStepLayer(incomings=[l_in, l_prev_hidden], recurrent_layer=self, name=name) class PseudoLSTMLayer(Layer): """ A Pseudo LSTM unit implements the following update mechanism: Incoming gate: i(t) = σ(W_hi @ h(t-1)) + W_xi @ x(t) + b_i) Forget gate: f(t) = σ(W_hf @ h(t-1)) + W_xf @ x(t) + b_f) Out gate: o(t) = σ(W_ho @ h(t-1)) + W_xo @ x(t) + b_o) New cell gate: c_new(t) = ϕ(W_hc @ (o(t) * h(t-1)) + W_xc @ x(t) + b_c) Cell gate: c(t) = f(t) * c(t-1) + i(t) * c_new(t) Hidden state: h(t) = ϕ(c(t)) Output: out = h(t) If gate_squash_inputs is set to True, we have the following updates instead: Out gate: o(t) = σ(W_ho @ h(t-1)) + W_xo @ x(t) + b_o) Incoming gate: i(t) = σ(W_hi @ (o(t) * h(t-1)) + W_xi @ x(t) + b_i) Forget gate: f(t) = σ(W_hf @ (o(t) * h(t-1)) + W_xf @ x(t) + b_f) New cell gate: c_new(t) = ϕ(W_hc @ (o(t) * h(t-1)) + W_xc @ x(t) + b_c) Cell state: c(t) = f(t) * c(t-1) + i(t) * c_new(t) Hidden state: h(t) = ϕ(c(t)) Output: out = h(t) Note that the incoming, forget, cell, and out vectors must have the same dimension as the hidden state The notation is slightly different from http://r2rt.com/written-memories-understanding-deriving-and-extending-the-lstm.html: here we introduce the cell gate and swap its role with the hidden state, so that the output is the same as the hidden state (and we can use this as a drop-in replacement for LSTMLayer). """ def __init__(self, incoming, num_units, hidden_nonlinearity=tf.tanh, gate_nonlinearity=tf.nn.sigmoid, W_x_init=XavierUniformInitializer(), W_h_init=OrthogonalInitializer(), forget_bias=1.0, b_init=tf.zeros_initializer(), hidden_init=tf.zeros_initializer(), hidden_init_trainable=False, cell_init=tf.zeros_initializer(), cell_init_trainable=False, gate_squash_inputs=False, layer_normalization=False, **kwargs): if hidden_nonlinearity is None: hidden_nonlinearity = tf.identity if gate_nonlinearity is None: gate_nonlinearity = tf.identity super(PseudoLSTMLayer, self).__init__(incoming, **kwargs) self.layer_normalization = layer_normalization input_shape = self.input_shape[2:] input_dim = np.prod(input_shape) # Weights for the initial hidden state (this is actually not used, since the initial hidden state is # determined by the initial cell state via h0 = self.nonlinearity(c0)). It is here merely for # interface convenience self.h0 = self.add_param(hidden_init, (num_units,), name="h0", trainable=hidden_init_trainable, regularizable=False) # Weights for the initial cell state self.c0 = self.add_param(cell_init, (num_units,), name="c0", trainable=cell_init_trainable, regularizable=False) # Weights for the incoming gate self.W_xi = self.add_param(W_x_init, (input_dim, num_units), name="W_xi") self.W_hi = self.add_param(W_h_init, (num_units, num_units), name="W_hi") self.b_i = self.add_param(b_init, (num_units,), name="b_i", regularizable=False) # Weights for the forget gate self.W_xf = self.add_param(W_x_init, (input_dim, num_units), name="W_xf") self.W_hf = self.add_param(W_h_init, (num_units, num_units), name="W_hf") self.b_f = self.add_param(b_init, (num_units,), name="b_f", regularizable=False) # Weights for the out gate self.W_xo = self.add_param(W_x_init, (input_dim, num_units), name="W_xo") self.W_ho = self.add_param(W_h_init, (num_units, num_units), name="W_ho") self.b_o = self.add_param(b_init, (num_units,), name="b_o", regularizable=False) # Weights for the cell gate self.W_xc = self.add_param(W_x_init, (input_dim, num_units), name="W_xc") self.W_hc = self.add_param(W_h_init, (num_units, num_units), name="W_hc") self.b_c = self.add_param(b_init, (num_units,), name="b_c", regularizable=False) self.gate_nonlinearity = gate_nonlinearity self.num_units = num_units self.nonlinearity = hidden_nonlinearity self.forget_bias = forget_bias self.gate_squash_inputs = gate_squash_inputs self.W_x_ifo = tf.concat(axis=1, values=[self.W_xi, self.W_xf, self.W_xo]) self.W_h_ifo = tf.concat(axis=1, values=[self.W_hi, self.W_hf, self.W_ho]) self.W_x_if = tf.concat(axis=1, values=[self.W_xi, self.W_xf]) self.W_h_if = tf.concat(axis=1, values=[self.W_hi, self.W_hf]) self.norm_params = dict() def step(self, hcprev, x): hprev = hcprev[:, :self.num_units] cprev = hcprev[:, self.num_units:] if self.layer_normalization: ln = apply_ln(self) else: ln = lambda x, *args: x if self.gate_squash_inputs: """ Out gate: o(t) = σ(W_ho @ h(t-1)) + W_xo @ x(t) + b_o) Incoming gate: i(t) = σ(W_hi @ (o(t) * h(t-1)) + W_xi @ x(t) + b_i) Forget gate: f(t) = σ(W_hf @ (o(t) * h(t-1)) + W_xf @ x(t) + b_f) New cell gate: c_new(t) = ϕ(W_hc @ (o(t) * h(t-1)) + W_xc @ x(t) + b_c) Cell state: c(t) = f(t) * c(t-1) + i(t) * c_new(t) Hidden state: h(t) = ϕ(c(t)) Output: out = h(t) """ o = self.nonlinearity( ln(tf.matmul(hprev, self.W_ho), "h_o") + ln(tf.matmul(x, self.W_xo), "x_o") + self.b_o ) x_if = ln(tf.matmul(x, self.W_x_if), "x_if") h_if = ln(tf.matmul(o * hprev, self.W_h_if), "h_if") x_i, x_f = tf.split(axis=1, num_or_size_splits=2, value=x_if) h_i, h_f = tf.split(axis=1, num_or_size_splits=2, value=h_if) i = self.gate_nonlinearity(x_i + h_i + self.b_i) f = self.gate_nonlinearity(x_f + h_f + self.b_f + self.forget_bias) c_new = self.nonlinearity( ln(tf.matmul(o * hprev, self.W_hc), "h_c") + ln(tf.matmul(x, self.W_xc), "x_c") + self.b_c ) c = f * cprev + i * c_new h = self.nonlinearity(ln(c, "c")) return tf.concat(axis=1, values=[h, c]) else: """ Incoming gate: i(t) = σ(W_hi @ h(t-1)) + W_xi @ x(t) + b_i) Forget gate: f(t) = σ(W_hf @ h(t-1)) + W_xf @ x(t) + b_f) Out gate: o(t) = σ(W_ho @ h(t-1)) + W_xo @ x(t) + b_o) New cell gate: c_new(t) = ϕ(W_hc @ (o(t) * h(t-1)) + W_xc @ x(t) + b_c) Cell gate: c(t) = f(t) * c(t-1) + i(t) * c_new(t) Hidden state: h(t) = ϕ(c(t)) Output: out = h(t) """ x_ifo = ln(tf.matmul(x, self.W_x_ifo), "x_ifo") h_ifo = ln(tf.matmul(hprev, self.W_h_ifo), "h_ifo") x_i, x_f, x_o = tf.split(axis=1, num_or_size_splits=3, value=x_ifo) h_i, h_f, h_o = tf.split(axis=1, num_or_size_splits=3, value=h_ifo) i = self.gate_nonlinearity(x_i + h_i + self.b_i) f = self.gate_nonlinearity(x_f + h_f + self.b_f + self.forget_bias) o = self.gate_nonlinearity(x_o + h_o + self.b_o) c_new = self.nonlinearity( ln(tf.matmul(o * hprev, self.W_hc), "h_c") + ln(tf.matmul(x, self.W_xc), "x_c") + self.b_c ) c = f * cprev + i * c_new h = self.nonlinearity(ln(c, "c")) return tf.concat(axis=1, values=[h, c]) def get_step_layer(self, l_in, l_prev_state, name=None): return LSTMStepLayer(incomings=[l_in, l_prev_state], recurrent_layer=self, name=name) def get_output_shape_for(self, input_shape): n_batch, n_steps = input_shape[:2] return n_batch, n_steps, self.num_units def get_output_for(self, input, **kwargs): input_shape = tf.shape(input) n_batches = input_shape[0] n_steps = input_shape[1] input = tf.reshape(input, tf.stack([n_batches, n_steps, -1])) c0s = tf.tile( tf.reshape(self.c0, (1, self.num_units)), (n_batches, 1) ) h0s = self.nonlinearity(c0s) # flatten extra dimensions shuffled_input = tf.transpose(input, (1, 0, 2)) hcs = tf.scan( self.step, elems=shuffled_input, initializer=tf.concat(axis=1, values=[h0s, c0s]) ) shuffled_hcs = tf.transpose(hcs, (1, 0, 2)) shuffled_hs = shuffled_hcs[:, :, :self.num_units] shuffled_cs = shuffled_hcs[:, :, self.num_units:] return shuffled_hs class LSTMLayer(Layer): """ A LSTM unit implements the following update mechanism: Incoming gate: i(t) = f_i(x(t) @ W_xi + h(t-1) @ W_hi + w_ci * c(t-1) + b_i) Forget gate: f(t) = f_f(x(t) @ W_xf + h(t-1) @ W_hf + w_cf * c(t-1) + b_f) Cell gate: c(t) = f(t) * c(t - 1) + i(t) * f_c(x(t) @ W_xc + h(t-1) @ W_hc + b_c) Out gate: o(t) = f_o(x(t) @ W_xo + h(t-1) W_ho + w_co * c(t) + b_o) New hidden state: h(t) = o(t) * f_h(c(t)) Note that the incoming, forget, cell, and out vectors must have the same dimension as the hidden state """ def __init__(self, incoming, num_units, hidden_nonlinearity=tf.tanh, gate_nonlinearity=tf.nn.sigmoid, W_x_init=XavierUniformInitializer(), W_h_init=OrthogonalInitializer(), forget_bias=1.0, use_peepholes=False, w_init=tf.random_normal_initializer(stddev=0.1), b_init=tf.zeros_initializer(), hidden_init=tf.zeros_initializer(), hidden_init_trainable=False, cell_init=tf.zeros_initializer(), cell_init_trainable=False, layer_normalization=False, **kwargs): if hidden_nonlinearity is None: hidden_nonlinearity = tf.identity if gate_nonlinearity is None: gate_nonlinearity = tf.identity super(LSTMLayer, self).__init__(incoming, **kwargs) self.layer_normalization = layer_normalization input_shape = self.input_shape[2:] input_dim = np.prod(input_shape) # Weights for the initial hidden state self.h0 = self.add_param(hidden_init, (num_units,), name="h0", trainable=hidden_init_trainable, regularizable=False) # Weights for the initial cell state self.c0 = self.add_param(cell_init, (num_units,), name="c0", trainable=cell_init_trainable, regularizable=False) # Weights for the incoming gate self.W_xi = self.add_param(W_x_init, (input_dim, num_units), name="W_xi") self.W_hi = self.add_param(W_h_init, (num_units, num_units), name="W_hi") if use_peepholes: self.w_ci = self.add_param(w_init, (num_units,), name="w_ci") self.b_i = self.add_param(b_init, (num_units,), name="b_i", regularizable=False) # Weights for the forget gate self.W_xf = self.add_param(W_x_init, (input_dim, num_units), name="W_xf") self.W_hf = self.add_param(W_h_init, (num_units, num_units), name="W_hf") if use_peepholes: self.w_cf = self.add_param(w_init, (num_units,), name="w_cf") self.b_f = self.add_param(b_init, (num_units,), name="b_f", regularizable=False) # Weights for the cell gate self.W_xc = self.add_param(W_x_init, (input_dim, num_units), name="W_xc") self.W_hc = self.add_param(W_h_init, (num_units, num_units), name="W_hc") self.b_c = self.add_param(b_init, (num_units,), name="b_c", regularizable=False) # Weights for the reset gate self.W_xr = self.add_param(W_x_init, (input_dim, num_units), name="W_xr") self.W_hr = self.add_param(W_h_init, (num_units, num_units), name="W_hr") self.b_r = self.add_param(b_init, (num_units,), name="b_r", regularizable=False) # Weights for the out gate self.W_xo = self.add_param(W_x_init, (input_dim, num_units), name="W_xo") self.W_ho = self.add_param(W_h_init, (num_units, num_units), name="W_ho") if use_peepholes: self.w_co = self.add_param(w_init, (num_units,), name="w_co") self.b_o = self.add_param(b_init, (num_units,), name="b_o", regularizable=False) self.gate_nonlinearity = gate_nonlinearity self.num_units = num_units self.nonlinearity = hidden_nonlinearity self.forget_bias = forget_bias self.use_peepholes = use_peepholes self.W_x_ifco = tf.concat(axis=1, values=[self.W_xi, self.W_xf, self.W_xc, self.W_xo]) self.W_h_ifco = tf.concat(axis=1, values=[self.W_hi, self.W_hf, self.W_hc, self.W_ho]) if use_peepholes: self.w_c_ifo = tf.concat(axis=0, values=[self.w_ci, self.w_cf, self.w_co]) self.norm_params = dict() def step(self, hcprev, x): """ Incoming gate: i(t) = f_i(x(t) @ W_xi + h(t-1) @ W_hi + w_ci * c(t-1) + b_i) Forget gate: f(t) = f_f(x(t) @ W_xf + h(t-1) @ W_hf + w_cf * c(t-1) + b_f) Cell gate: c(t) = f(t) * c(t - 1) + i(t) * f_c(x(t) @ W_xc + h(t-1) @ W_hc + b_c) Out gate: o(t) = f_o(x(t) @ W_xo + h(t-1) W_ho + w_co * c(t) + b_o) New hidden state: h(t) = o(t) * f_h(c(t)) """ hprev = hcprev[:, :self.num_units] cprev = hcprev[:, self.num_units:] if self.layer_normalization: ln = apply_ln(self) else: ln = lambda x, *args: x x_ifco = ln(tf.matmul(x, self.W_x_ifco), "x_ifco") h_ifco = ln(tf.matmul(hprev, self.W_h_ifco), "h_ifco") x_i, x_f, x_c, x_o = tf.split(axis=1, num_or_size_splits=4, value=x_ifco) h_i, h_f, h_c, h_o = tf.split(axis=1, num_or_size_splits=4, value=h_ifco) if self.use_peepholes: i = self.gate_nonlinearity(x_i + h_i + self.w_ci * cprev + self.b_i) f = self.gate_nonlinearity(x_f + h_f + self.w_cf * cprev + self.b_f + self.forget_bias) o = self.gate_nonlinearity(x_o + h_o + self.w_co * cprev + self.b_o) else: i = self.gate_nonlinearity(x_i + h_i + self.b_i) f = self.gate_nonlinearity(x_f + h_f + self.b_f + self.forget_bias) o = self.gate_nonlinearity(x_o + h_o + self.b_o) c = f * cprev + i * self.nonlinearity(x_c + h_c + self.b_c) h = o * self.nonlinearity(ln(c, "c")) return tf.concat(axis=1, values=[h, c]) def get_step_layer(self, l_in, l_prev_state, name=None): return LSTMStepLayer(incomings=[l_in, l_prev_state], recurrent_layer=self, name=name) def get_output_shape_for(self, input_shape): n_batch, n_steps = input_shape[:2] return n_batch, n_steps, self.num_units def get_output_for(self, input, **kwargs): input_shape = tf.shape(input) n_batches = input_shape[0] n_steps = input_shape[1] input = tf.reshape(input, tf.stack([n_batches, n_steps, -1])) h0s = tf.tile( tf.reshape(self.h0, (1, self.num_units)), (n_batches, 1) ) c0s = tf.tile( tf.reshape(self.c0, (1, self.num_units)), (n_batches, 1) ) # flatten extra dimensions shuffled_input = tf.transpose(input, (1, 0, 2)) hcs = tf.scan( self.step, elems=shuffled_input, initializer=tf.concat(axis=1, values=[h0s, c0s]) ) shuffled_hcs = tf.transpose(hcs, (1, 0, 2)) shuffled_hs = shuffled_hcs[:, :, :self.num_units] shuffled_cs = shuffled_hcs[:, :, self.num_units:] if 'recurrent_state_output' in kwargs: kwargs['recurrent_state_output'][self] = shuffled_hcs return shuffled_hs class LSTMStepLayer(MergeLayer): def __init__(self, incomings, recurrent_layer, **kwargs): super(LSTMStepLayer, self).__init__(incomings, **kwargs) self._recurrent_layer = recurrent_layer def get_params(self, **tags): return self._recurrent_layer.get_params(**tags) def get_output_shape_for(self, input_shapes): n_batch = input_shapes[0][0] return n_batch, 2 * self._recurrent_layer.num_units def get_output_for(self, inputs, **kwargs): x, hcprev = inputs n_batch = tf.shape(x)[0] x = tf.reshape(x, tf.stack([n_batch, -1])) hc = self._recurrent_layer.step(hcprev, x) return hc class TfBasicLSTMLayer(Layer): """ Use TensorFlow's built-in (basic) LSTM implementation """ def __init__(self, incoming, num_units, hidden_nonlinearity, horizon=None, hidden_init_trainable=False, forget_bias=1.0, use_peepholes=False, **kwargs): assert not use_peepholes, "Basic LSTM does not support peepholes!" assert len(incoming.output_shape) == 3 input_dim = incoming.shape[2] lstm = tf.contrib.rnn.BasicLSTMCell( num_units=num_units, activation=hidden_nonlinearity, state_is_tuple=True, forget_bias=forget_bias ) self.num_units = num_units self.horizon = horizon self.lstm = lstm self.hidden_nonlinearity = hidden_nonlinearity Layer.__init__(self, incoming=incoming, **kwargs) # dummy input variable input_dummy = tf.placeholder(tf.float32, (None, input_dim), "input_dummy") hidden_dummy = tf.placeholder(tf.float32, (None, num_units), "hidden_dummy") cell_dummy = tf.placeholder(tf.float32, (None, num_units), "cell_dummy") with tf.variable_scope(self.name) as vs: lstm(input_dummy, (cell_dummy, hidden_dummy), scope=vs) vs.reuse_variables() self.scope = vs all_vars = [v for v in tf.global_variables() if v.name.startswith(vs.name)] trainable_vars = [v for v in tf.trainable_variables() if v.name.startswith(vs.name)] for var in trainable_vars: self.add_param(spec=var, shape=None, name=None, trainable=True) for var in set(all_vars) - set(trainable_vars): self.add_param(spec=var, shape=None, name=None, trainable=False) self.h0 = self.add_param(tf.zeros_initializer(), (num_units,), name="h0", trainable=hidden_init_trainable, regularizable=False) self.c0 = self.add_param(tf.zeros_initializer(), (num_units,), name="c0", trainable=hidden_init_trainable, regularizable=False) def step(self, hcprev, x): hprev = hcprev[:, :self.num_units] cprev = hcprev[:, self.num_units:] x.set_shape((None, self.input_shape[-1])) c, h = self.lstm(x, (cprev, hprev), scope=self.scope)[1] return tf.concat(axis=1, values=[h, c]) def get_output_for(self, input, **kwargs): input_shape = tf.shape(input) n_batches = input_shape[0] h0s = tf.tile( tf.reshape(self.h0, (1, self.num_units)), (n_batches, 1) ) h0s.set_shape((None, self.num_units)) c0s = tf.tile( tf.reshape(self.c0, (1, self.num_units)), (n_batches, 1) ) c0s.set_shape((None, self.num_units)) state = (c0s, h0s) if self.horizon is not None: outputs = [] for idx in range(self.horizon): output, state = self.lstm(input[:, idx, :], state, scope=self.scope) # self.name) outputs.append(tf.expand_dims(output, 1)) outputs = tf.concat(axis=1, values=outputs) return outputs else: n_steps = input_shape[1] input = tf.reshape(input, tf.stack([n_batches, n_steps, -1])) # flatten extra dimensions shuffled_input = tf.transpose(input, (1, 0, 2)) shuffled_input.set_shape((None, None, self.input_shape[-1])) hcs = tf.scan( self.step, elems=shuffled_input, initializer=tf.concat(axis=1, values=[h0s, c0s]), ) shuffled_hcs = tf.transpose(hcs, (1, 0, 2)) shuffled_hs = shuffled_hcs[:, :, :self.num_units] shuffled_cs = shuffled_hcs[:, :, self.num_units:] return shuffled_hs def get_output_shape_for(self, input_shape): n_batch, n_steps = input_shape[:2] return n_batch, n_steps, self.num_units def get_step_layer(self, l_in, l_prev_state, name=None): return LSTMStepLayer(incomings=[l_in, l_prev_state], recurrent_layer=self, name=name) def get_all_layers(layer, treat_as_input=None): """ :type layer: Layer | list[Layer] :rtype: list[Layer] """ # We perform a depth-first search. We add a layer to the result list only # after adding all its incoming layers (if any) or when detecting a cycle. # We use a LIFO stack to avoid ever running into recursion depth limits. try: queue = deque(layer) except TypeError: queue = deque([layer]) seen = set() done = set() result = [] # If treat_as_input is given, we pretend we've already collected all their # incoming layers. if treat_as_input is not None: seen.update(treat_as_input) while queue: # Peek at the leftmost node in the queue. layer = queue[0] if layer is None: # Some node had an input_layer set to `None`. Just ignore it. queue.popleft() elif layer not in seen: # We haven't seen this node yet: Mark it and queue all incomings # to be processed first. If there are no incomings, the node will # be appended to the result list in the next iteration. seen.add(layer) if hasattr(layer, 'input_layers'): queue.extendleft(reversed(layer.input_layers)) elif hasattr(layer, 'input_layer'): queue.appendleft(layer.input_layer) else: # We've been here before: Either we've finished all its incomings, # or we've detected a cycle. In both cases, we remove the layer # from the queue and append it to the result list. queue.popleft() if layer not in done: result.append(layer) done.add(layer) return result class NonlinearityLayer(Layer): def __init__(self, incoming, nonlinearity=tf.nn.relu, **kwargs): super(NonlinearityLayer, self).__init__(incoming, **kwargs) self.nonlinearity = (tf.identity if nonlinearity is None else nonlinearity) def get_output_for(self, input, **kwargs): return self.nonlinearity(input) def get_output_shape_for(self, input_shape): return input_shape class BatchNormLayer(Layer): def __init__(self, incoming, center=True, scale=False, epsilon=0.001, decay=0.9, beta=tf.zeros_initializer(), gamma=tf.ones_initializer(), moving_mean=tf.zeros_initializer(), moving_variance=tf.ones_initializer(), **kwargs): super(BatchNormLayer, self).__init__(incoming, **kwargs) self.center = center self.scale = scale self.epsilon = epsilon self.decay = decay input_shape = incoming.output_shape axis = list(range(len(input_shape) - 1)) params_shape = input_shape[-1:] if center: self.beta = self.add_param(beta, shape=params_shape, name='beta', trainable=True, regularizable=False) else: self.beta = None if scale: self.gamma = self.add_param(gamma, shape=params_shape, name='gamma', trainable=True, regularizable=True) else: self.gamma = None self.moving_mean = self.add_param(moving_mean, shape=params_shape, name='moving_mean', trainable=False, regularizable=False) self.moving_variance = self.add_param(moving_variance, shape=params_shape, name='moving_variance', trainable=False, regularizable=False) self.axis = axis def get_output_for(self, input, phase='train', **kwargs): if phase == 'train': # Calculate the moments based on the individual batch. mean, variance = tf.nn.moments(input, self.axis, shift=self.moving_mean) # Update the moving_mean and moving_variance moments. update_moving_mean = moving_averages.assign_moving_average( self.moving_mean, mean, self.decay) update_moving_variance = moving_averages.assign_moving_average( self.moving_variance, variance, self.decay) # Make sure the updates are computed here. with tf.control_dependencies([update_moving_mean, update_moving_variance]): output = tf.nn.batch_normalization( input, mean, variance, self.beta, self.gamma, self.epsilon) else: output = tf.nn.batch_normalization( input, self.moving_mean, self.moving_variance, self.beta, self.gamma, self.epsilon) output.set_shape(self.input_shape) return output def get_output_shape_for(self, input_shape): return input_shape def batch_norm(layer, **kwargs): nonlinearity = getattr(layer, 'nonlinearity', None) scale = True if nonlinearity is not None: layer.nonlinearity = tf.identity if nonlinearity is tf.nn.relu: scale = False if hasattr(layer, 'b') and layer.b is not None: del layer.params[layer.b] layer.b = None bn_name = (kwargs.pop('name', None) or (getattr(layer, 'name', None) and layer.name + '_bn')) layer = BatchNormLayer(layer, name=bn_name, scale=scale, **kwargs) if nonlinearity is not None: nonlin_name = bn_name and bn_name + '_nonlin' layer = NonlinearityLayer(layer, nonlinearity=nonlinearity, name=nonlin_name) return layer class ElemwiseSumLayer(MergeLayer): def __init__(self, incomings, **kwargs): super(ElemwiseSumLayer, self).__init__(incomings, **kwargs) def get_output_for(self, inputs, **kwargs): return functools.reduce(tf.add, inputs) def get_output_shape_for(self, input_shapes): assert len(set(input_shapes)) == 1 return input_shapes[0] def get_output(layer_or_layers, inputs=None, **kwargs): # track accepted kwargs used by get_output_for accepted_kwargs = {'deterministic'} # obtain topological ordering of all layers the output layer(s) depend on treat_as_input = list(inputs.keys()) if isinstance(inputs, dict) else [] all_layers = get_all_layers(layer_or_layers, treat_as_input) # initialize layer-to-expression mapping from all input layers all_outputs = dict((layer, layer.input_var) for layer in all_layers if isinstance(layer, InputLayer) and layer not in treat_as_input) # update layer-to-expression mapping from given input(s), if any if isinstance(inputs, dict): all_outputs.update((layer, tf.convert_to_tensor(expr)) for layer, expr in list(inputs.items())) elif inputs is not None: if len(all_outputs) > 1: raise ValueError("get_output() was called with a single input " "expression on a network with multiple input " "layers. Please call it with a dictionary of " "input expressions instead.") for input_layer in all_outputs: all_outputs[input_layer] = tf.convert_to_tensor(inputs) # update layer-to-expression mapping by propagating the inputs for layer in all_layers: if layer not in all_outputs: try: if isinstance(layer, MergeLayer): layer_inputs = [all_outputs[input_layer] for input_layer in layer.input_layers] else: layer_inputs = all_outputs[layer.input_layer] except KeyError: # one of the input_layer attributes must have been `None` raise ValueError("get_output() was called without giving an " "input expression for the free-floating " "layer %r. Please call it with a dictionary " "mapping this layer to an input expression." % layer) all_outputs[layer] = layer.get_output_for(layer_inputs, **kwargs) try: names, _, _, defaults = getargspec(layer.get_output_for) except TypeError: # If introspection is not possible, skip it pass else: if defaults is not None: accepted_kwargs |= set(names[-len(defaults):]) accepted_kwargs |= set(layer.get_output_kwargs) unused_kwargs = set(kwargs.keys()) - accepted_kwargs if unused_kwargs: suggestions = [] for kwarg in unused_kwargs: suggestion = get_close_matches(kwarg, accepted_kwargs) if suggestion: suggestions.append('%s (perhaps you meant %s)' % (kwarg, suggestion[0])) else: suggestions.append(kwarg) warn("get_output() was called with unused kwargs:\n\t%s" % "\n\t".join(suggestions)) # return the output(s) of the requested layer(s) only try: return [all_outputs[layer] for layer in layer_or_layers] except TypeError: return all_outputs[layer_or_layers] def unique(l): """Filters duplicates of iterable. Create a new list from l with duplicate entries removed, while preserving the original order. Parameters ---------- l : iterable Input iterable to filter of duplicates. Returns ------- list A list of elements of `l` without duplicates and in the same order. """ new_list = [] seen = set() for el in l: if el not in seen: new_list.append(el) seen.add(el) return new_list def get_all_params(layer, **tags): """ :type layer: Layer|list[Layer] """ layers = get_all_layers(layer) params = chain.from_iterable(l.get_params(**tags) for l in layers) return unique(params)
76,922
40.557536
120
py
rllab
rllab-master/sandbox/rocky/tf/core/layers_powered.py
from sandbox.rocky.tf.core.parameterized import Parameterized import sandbox.rocky.tf.core.layers as L import itertools class LayersPowered(Parameterized): def __init__(self, output_layers, input_layers=None): self._output_layers = output_layers self._input_layers = input_layers Parameterized.__init__(self) def get_params_internal(self, **tags): layers = L.get_all_layers(self._output_layers, treat_as_input=self._input_layers) params = itertools.chain.from_iterable(l.get_params(**tags) for l in layers) return L.unique(params)
592
31.944444
89
py
rllab
rllab-master/sandbox/rocky/tf/core/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/core/parameterized.py
from contextlib import contextmanager from rllab.core.serializable import Serializable from rllab.misc.tensor_utils import flatten_tensors, unflatten_tensors import tensorflow as tf load_params = True @contextmanager def suppress_params_loading(): global load_params load_params = False yield load_params = True class Parameterized(object): def __init__(self): self._cached_params = {} self._cached_param_dtypes = {} self._cached_param_shapes = {} self._cached_assign_ops = {} self._cached_assign_placeholders = {} def get_params_internal(self, **tags): """ Internal method to be implemented which does not perform caching """ raise NotImplementedError def get_params(self, **tags): """ Get the list of parameters, filtered by the provided tags. Some common tags include 'regularizable' and 'trainable' """ tag_tuple = tuple(sorted(list(tags.items()), key=lambda x: x[0])) if tag_tuple not in self._cached_params: self._cached_params[tag_tuple] = self.get_params_internal(**tags) return self._cached_params[tag_tuple] def get_param_dtypes(self, **tags): tag_tuple = tuple(sorted(list(tags.items()), key=lambda x: x[0])) if tag_tuple not in self._cached_param_dtypes: params = self.get_params(**tags) param_values = tf.get_default_session().run(params) self._cached_param_dtypes[tag_tuple] = [val.dtype for val in param_values] return self._cached_param_dtypes[tag_tuple] def get_param_shapes(self, **tags): tag_tuple = tuple(sorted(list(tags.items()), key=lambda x: x[0])) if tag_tuple not in self._cached_param_shapes: params = self.get_params(**tags) param_values = tf.get_default_session().run(params) self._cached_param_shapes[tag_tuple] = [val.shape for val in param_values] return self._cached_param_shapes[tag_tuple] def get_param_values(self, **tags): params = self.get_params(**tags) param_values = tf.get_default_session().run(params) return flatten_tensors(param_values) def set_param_values(self, flattened_params, **tags): debug = tags.pop("debug", False) param_values = unflatten_tensors( flattened_params, self.get_param_shapes(**tags)) ops = [] feed_dict = dict() for param, dtype, value in zip( self.get_params(**tags), self.get_param_dtypes(**tags), param_values): if param not in self._cached_assign_ops: assign_placeholder = tf.placeholder(dtype=param.dtype.base_dtype) assign_op = tf.assign(param, assign_placeholder) self._cached_assign_ops[param] = assign_op self._cached_assign_placeholders[param] = assign_placeholder ops.append(self._cached_assign_ops[param]) feed_dict[self._cached_assign_placeholders[param]] = value.astype(dtype) if debug: print("setting value of %s" % param.name) tf.get_default_session().run(ops, feed_dict=feed_dict) def flat_to_params(self, flattened_params, **tags): return unflatten_tensors(flattened_params, self.get_param_shapes(**tags)) def __getstate__(self): d = Serializable.__getstate__(self) global load_params if load_params: d["params"] = self.get_param_values() return d def __setstate__(self, d): Serializable.__setstate__(self, d) global load_params if load_params: tf.get_default_session().run(tf.variables_initializer(self.get_params())) self.set_param_values(d["params"]) class JointParameterized(Parameterized): def __init__(self, components): super(JointParameterized, self).__init__() self.components = components def get_params_internal(self, **tags): params = [param for comp in self.components for param in comp.get_params_internal(**tags)] # only return unique parameters return sorted(set(params), key=hash)
4,226
37.081081
98
py
rllab
rllab-master/sandbox/rocky/tf/envs/parallel_vec_env_executor.py
import numpy as np import pickle as pickle from sandbox.rocky.tf.misc import tensor_utils from rllab.misc import logger from rllab.sampler.stateful_pool import singleton_pool import uuid def worker_init_envs(G, alloc, scope, env): logger.log("initializing environment on worker %d" % G.worker_id) if not hasattr(G, 'parallel_vec_envs'): G.parallel_vec_envs = dict() G.parallel_vec_env_template = dict() G.parallel_vec_envs[scope] = [(idx, pickle.loads(pickle.dumps(env))) for idx in alloc] G.parallel_vec_env_template[scope] = env # For these two methods below, we pack the data into batch numpy arrays whenever possible, to reduce communication cost def worker_run_reset(G, flags, scope): if not hasattr(G, 'parallel_vec_envs'): logger.log("on worker %d" % G.worker_id) import traceback for line in traceback.format_stack(): logger.log(line) # log the stacktrace at least logger.log("oops") for k, v in G.__dict__.items(): logger.log(str(k) + " : " + str(v)) assert hasattr(G, 'parallel_vec_envs') assert scope in G.parallel_vec_envs N = len(G.parallel_vec_envs[scope]) env_template = G.parallel_vec_env_template[scope] obs_dim = env_template.observation_space.flat_dim ret_arr = np.zeros((N, obs_dim)) ids = [] flat_obs = [] reset_ids = [] for itr_idx, (idx, env) in enumerate(G.parallel_vec_envs[scope]): flag = flags[idx] if flag: flat_obs.append(env.reset()) reset_ids.append(itr_idx) ids.append(idx) if len(reset_ids) > 0: ret_arr[reset_ids] = env_template.observation_space.flatten_n(flat_obs) return ids, ret_arr def worker_run_step(G, action_n, scope): assert hasattr(G, 'parallel_vec_envs') assert scope in G.parallel_vec_envs env_template = G.parallel_vec_env_template[scope] ids = [] step_results = [] for (idx, env) in G.parallel_vec_envs[scope]: action = action_n[idx] ids.append(idx) step_results.append(tuple(env.step(action))) if len(step_results) == 0: return None obs, rewards, dones, env_infos = list(map(list, list(zip(*step_results)))) obs = env_template.observation_space.flatten_n(obs) rewards = np.asarray(rewards) dones = np.asarray(dones) env_infos = tensor_utils.stack_tensor_dict_list(env_infos) return ids, obs, rewards, dones, env_infos def worker_collect_env_time(G): return G.env_time class ParallelVecEnvExecutor(object): def __init__(self, env, n, max_path_length, scope=None): if scope is None: # initialize random scope scope = str(uuid.uuid4()) envs_per_worker = int(np.ceil(n * 1.0 / singleton_pool.n_parallel)) alloc_env_ids = [] rest_alloc = n start_id = 0 for _ in range(singleton_pool.n_parallel): n_allocs = min(envs_per_worker, rest_alloc) alloc_env_ids.append(list(range(start_id, start_id + n_allocs))) start_id += n_allocs rest_alloc = max(0, rest_alloc - envs_per_worker) singleton_pool.run_each(worker_init_envs, [(alloc, scope, env) for alloc in alloc_env_ids]) self._alloc_env_ids = alloc_env_ids self._action_space = env.action_space self._observation_space = env.observation_space self._num_envs = n self.scope = scope self.ts = np.zeros(n, dtype='int') self.max_path_length = max_path_length def step(self, action_n): results = singleton_pool.run_each( worker_run_step, [(action_n, self.scope) for _ in self._alloc_env_ids], ) results = [x for x in results if x is not None] ids, obs, rewards, dones, env_infos = list(zip(*results)) ids = np.concatenate(ids) obs = self.observation_space.unflatten_n(np.concatenate(obs)) rewards = np.concatenate(rewards) dones = np.concatenate(dones) env_infos = tensor_utils.split_tensor_dict_list(tensor_utils.concat_tensor_dict_list(env_infos)) if env_infos is None: env_infos = [dict() for _ in range(self.num_envs)] items = list(zip(ids, obs, rewards, dones, env_infos)) items = sorted(items, key=lambda x: x[0]) ids, obs, rewards, dones, env_infos = list(zip(*items)) obs = list(obs) rewards = np.asarray(rewards) dones = np.asarray(dones) self.ts += 1 dones[self.ts >= self.max_path_length] = True reset_obs = self._run_reset(dones) for (i, done) in enumerate(dones): if done: obs[i] = reset_obs[i] self.ts[i] = 0 return obs, rewards, dones, tensor_utils.stack_tensor_dict_list(list(env_infos)) def _run_reset(self, dones): dones = np.asarray(dones) results = singleton_pool.run_each( worker_run_reset, [(dones, self.scope) for _ in self._alloc_env_ids], ) ids, flat_obs = list(map(np.concatenate, list(zip(*results)))) zipped = list(zip(ids, flat_obs)) sorted_obs = np.asarray([x[1] for x in sorted(zipped, key=lambda x: x[0])]) done_ids, = np.where(dones) done_flat_obs = sorted_obs[done_ids] done_unflat_obs = self.observation_space.unflatten_n(done_flat_obs) all_obs = [None] * self.num_envs done_cursor = 0 for idx, done in enumerate(dones): if done: all_obs[idx] = done_unflat_obs[done_cursor] done_cursor += 1 return all_obs def reset(self): dones = np.asarray([True] * self.num_envs) return self._run_reset(dones) @property def num_envs(self): return self._num_envs @property def action_space(self): return self._action_space @property def observation_space(self): return self._observation_space def terminate(self): pass
6,057
33.225989
119
py
rllab
rllab-master/sandbox/rocky/tf/envs/base.py
from rllab.envs.proxy_env import ProxyEnv from rllab.envs.base import EnvSpec from rllab.spaces.box import Box as TheanoBox from rllab.spaces.discrete import Discrete as TheanoDiscrete from rllab.spaces.product import Product as TheanoProduct from sandbox.rocky.tf.spaces.discrete import Discrete from sandbox.rocky.tf.spaces.box import Box from sandbox.rocky.tf.spaces.product import Product from cached_property import cached_property def to_tf_space(space): if isinstance(space, TheanoBox): return Box(low=space.low, high=space.high) elif isinstance(space, TheanoDiscrete): return Discrete(space.n) elif isinstance(space, TheanoProduct): return Product(list(map(to_tf_space, space.components))) else: raise NotImplementedError class WrappedCls(object): def __init__(self, cls, env_cls, extra_kwargs): self.cls = cls self.env_cls = env_cls self.extra_kwargs = extra_kwargs def __call__(self, *args, **kwargs): return self.cls(self.env_cls(*args, **dict(self.extra_kwargs, **kwargs))) class TfEnv(ProxyEnv): @cached_property def observation_space(self): return to_tf_space(self.wrapped_env.observation_space) @cached_property def action_space(self): return to_tf_space(self.wrapped_env.action_space) @cached_property def spec(self): return EnvSpec( observation_space=self.observation_space, action_space=self.action_space, ) @property def vectorized(self): return getattr(self.wrapped_env, "vectorized", False) def vec_env_executor(self, n_envs, max_path_length): return VecTfEnv(self.wrapped_env.vec_env_executor(n_envs=n_envs, max_path_length=max_path_length)) @classmethod def wrap(cls, env_cls, **extra_kwargs): # Use a class wrapper rather than a lambda method for smoother serialization return WrappedCls(cls, env_cls, extra_kwargs) class VecTfEnv(object): def __init__(self, vec_env): self.vec_env = vec_env def reset(self): return self.vec_env.reset() @property def num_envs(self): return self.vec_env.num_envs def step(self, action_n): return self.vec_env.step(action_n) def terminate(self): self.vec_env.terminate()
2,330
28.506329
106
py
rllab
rllab-master/sandbox/rocky/tf/envs/vec_env_executor.py
import numpy as np import pickle as pickle from sandbox.rocky.tf.misc import tensor_utils class VecEnvExecutor(object): def __init__(self, envs, max_path_length): self.envs = envs self._action_space = envs[0].action_space self._observation_space = envs[0].observation_space self.ts = np.zeros(len(self.envs), dtype='int') self.max_path_length = max_path_length def step(self, action_n): all_results = [env.step(a) for (a, env) in zip(action_n, self.envs)] obs, rewards, dones, env_infos = list(map(list, list(zip(*all_results)))) dones = np.asarray(dones) rewards = np.asarray(rewards) self.ts += 1 if self.max_path_length is not None: dones[self.ts >= self.max_path_length] = True for (i, done) in enumerate(dones): if done: obs[i] = self.envs[i].reset() self.ts[i] = 0 return obs, rewards, dones, tensor_utils.stack_tensor_dict_list(env_infos) def reset(self): results = [env.reset() for env in self.envs] self.ts[:] = 0 return results @property def num_envs(self): return len(self.envs) @property def action_space(self): return self._action_space @property def observation_space(self): return self._observation_space def terminate(self): pass
1,412
27.836735
82
py
rllab
rllab-master/sandbox/rocky/tf/envs/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/distributions/recurrent_diagonal_gaussian.py
from sandbox.rocky.tf.distributions.diagonal_gaussian import DiagonalGaussian RecurrentDiagonalGaussian = DiagonalGaussian
127
17.285714
77
py
rllab
rllab-master/sandbox/rocky/tf/distributions/base.py
class Distribution(object): @property def dim(self): raise NotImplementedError def kl_sym(self, old_dist_info_vars, new_dist_info_vars): """ Compute the symbolic KL divergence of two distributions """ raise NotImplementedError def kl(self, old_dist_info, new_dist_info): """ Compute the KL divergence of two distributions """ raise NotImplementedError def likelihood_ratio_sym(self, x_var, old_dist_info_vars, new_dist_info_vars): raise NotImplementedError def entropy(self, dist_info): raise NotImplementedError def log_likelihood_sym(self, x_var, dist_info_vars): raise NotImplementedError def log_likelihood(self, xs, dist_info): raise NotImplementedError @property def dist_info_specs(self): raise NotImplementedError @property def dist_info_keys(self): return [k for k, _ in self.dist_info_specs]
982
22.97561
82
py
rllab
rllab-master/sandbox/rocky/tf/distributions/categorical.py
import numpy as np from .base import Distribution import tensorflow as tf from sandbox.rocky.tf.misc import tensor_utils TINY = 1e-8 def from_onehot(x_var): ret = np.zeros((len(x_var),), 'int32') nonzero_n, nonzero_a = np.nonzero(x_var) ret[nonzero_n] = nonzero_a return ret class Categorical(Distribution): def __init__(self, dim): self._dim = dim weights_var = tf.placeholder( dtype=tf.float32, shape=(None, dim), name="weights" ) self._f_sample = tensor_utils.compile_function( inputs=[weights_var], outputs=tf.multinomial(tf.log(weights_var + 1e-8), num_samples=1)[:, 0], ) @property def dim(self): return self._dim def kl_sym(self, old_dist_info_vars, new_dist_info_vars): """ Compute the symbolic KL divergence of two categorical distributions """ old_prob_var = old_dist_info_vars["prob"] new_prob_var = new_dist_info_vars["prob"] ndims = old_prob_var.get_shape().ndims # Assume layout is N * A return tf.reduce_sum( old_prob_var * (tf.log(old_prob_var + TINY) - tf.log(new_prob_var + TINY)), axis=ndims - 1 ) def kl(self, old_dist_info, new_dist_info): """ Compute the KL divergence of two categorical distributions """ old_prob = old_dist_info["prob"] new_prob = new_dist_info["prob"] return np.sum( old_prob * (np.log(old_prob + TINY) - np.log(new_prob + TINY)), axis=-1 ) def likelihood_ratio_sym(self, x_var, old_dist_info_vars, new_dist_info_vars): old_prob_var = old_dist_info_vars["prob"] new_prob_var = new_dist_info_vars["prob"] ndims = old_prob_var.get_shape().ndims x_var = tf.cast(x_var, tf.float32) # Assume layout is N * A return (tf.reduce_sum(new_prob_var * x_var, ndims - 1) + TINY) / \ (tf.reduce_sum(old_prob_var * x_var, ndims - 1) + TINY) def entropy_sym(self, dist_info_vars): probs = dist_info_vars["prob"] return -tf.reduce_sum(probs * tf.log(probs + TINY), axis=1) def cross_entropy_sym(self, old_dist_info_vars, new_dist_info_vars): old_prob_var = old_dist_info_vars["prob"] new_prob_var = new_dist_info_vars["prob"] ndims = old_prob_var.get_shape().ndims # Assume layout is N * A return tf.reduce_sum( old_prob_var * (- tf.log(new_prob_var + TINY)), axis=ndims - 1 ) def entropy(self, info): probs = info["prob"] return -np.sum(probs * np.log(probs + TINY), axis=1) def log_likelihood_sym(self, x_var, dist_info_vars): probs = dist_info_vars["prob"] ndims = probs.get_shape().ndims return tf.log(tf.reduce_sum(probs * tf.cast(x_var, tf.float32), ndims - 1) + TINY) def log_likelihood(self, xs, dist_info): probs = dist_info["prob"] # Assume layout is N * A return np.log(np.sum(probs * xs, axis=-1) + TINY) @property def dist_info_specs(self): return [("prob", (self.dim,))] def sample(self, dist_info): return self._f_sample(dist_info["prob"]) def sample_sym(self, dist_info): probs = dist_info["prob"] samples = tf.multinomial(tf.log(probs + 1e-8), num_samples=1)[:, 0] return tf.nn.embedding_lookup(np.eye(self.dim, dtype=np.float32), samples)
3,514
32.47619
90
py
rllab
rllab-master/sandbox/rocky/tf/distributions/recurrent_categorical.py
import tensorflow as tf import numpy as np from sandbox.rocky.tf.distributions.categorical import Categorical from sandbox.rocky.tf.distributions.base import Distribution TINY = 1e-8 class RecurrentCategorical(Distribution): def __init__(self, dim): self._cat = Categorical(dim) self._dim = dim @property def dim(self): return self._dim def kl_sym(self, old_dist_info_vars, new_dist_info_vars): """ Compute the symbolic KL divergence of two categorical distributions """ old_prob_var = old_dist_info_vars["prob"] new_prob_var = new_dist_info_vars["prob"] # Assume layout is N * T * A return tf.reduce_sum( old_prob_var * (tf.log(old_prob_var + TINY) - tf.log(new_prob_var + TINY)), axis=2 ) def kl(self, old_dist_info, new_dist_info): """ Compute the KL divergence of two categorical distributions """ old_prob = old_dist_info["prob"] new_prob = new_dist_info["prob"] return np.sum( old_prob * (np.log(old_prob + TINY) - np.log(new_prob + TINY)), axis=2 ) def likelihood_ratio_sym(self, x_var, old_dist_info_vars, new_dist_info_vars): old_prob_var = old_dist_info_vars["prob"] new_prob_var = new_dist_info_vars["prob"] # Assume layout is N * T * A a_dim = tf.shape(x_var)[2] flat_ratios = self._cat.likelihood_ratio_sym( tf.reshape(x_var, tf.stack([-1, a_dim])), dict(prob=tf.reshape(old_prob_var, tf.stack([-1, a_dim]))), dict(prob=tf.reshape(new_prob_var, tf.stack([-1, a_dim]))) ) return tf.reshape(flat_ratios, tf.shape(old_prob_var)[:2]) def entropy(self, dist_info): probs = dist_info["prob"] return -np.sum(probs * np.log(probs + TINY), axis=2) def entropy_sym(self, dist_info_vars): probs = dist_info_vars["prob"] return -tf.reduce_sum(probs * tf.log(probs + TINY), 2) def log_likelihood_sym(self, xs, dist_info_vars): probs = dist_info_vars["prob"] # Assume layout is N * T * A a_dim = tf.shape(probs)[2] # a_dim = TT.printing.Print("lala")(a_dim) flat_logli = self._cat.log_likelihood_sym( tf.reshape(xs, tf.stack([-1, a_dim])), dict(prob=tf.reshape(probs, tf.stack((-1, a_dim)))) ) return tf.reshape(flat_logli, tf.shape(probs)[:2]) def log_likelihood(self, xs, dist_info): probs = dist_info["prob"] # Assume layout is N * T * A a_dim = tf.shape(probs)[2] flat_logli = self._cat.log_likelihood_sym( xs.reshape((-1, a_dim)), dict(prob=probs.reshape((-1, a_dim))) ) return flat_logli.reshape(probs.shape[:2]) @property def dist_info_specs(self): return [("prob", (self.dim,))]
2,923
33
87
py
rllab
rllab-master/sandbox/rocky/tf/distributions/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/distributions/diagonal_gaussian.py
import tensorflow as tf import numpy as np from sandbox.rocky.tf.distributions.base import Distribution class DiagonalGaussian(Distribution): def __init__(self, dim): self._dim = dim @property def dim(self): return self._dim def kl(self, old_dist_info, new_dist_info): old_means = old_dist_info["mean"] old_log_stds = old_dist_info["log_std"] new_means = new_dist_info["mean"] new_log_stds = new_dist_info["log_std"] """ Compute the KL divergence of two multivariate Gaussian distribution with diagonal covariance matrices """ old_std = np.exp(old_log_stds) new_std = np.exp(new_log_stds) # means: (N*A) # std: (N*A) # formula: # { (\mu_1 - \mu_2)^2 + \sigma_1^2 - \sigma_2^2 } / (2\sigma_2^2) + # ln(\sigma_2/\sigma_1) numerator = np.square(old_means - new_means) + \ np.square(old_std) - np.square(new_std) denominator = 2 * np.square(new_std) + 1e-8 return np.sum( numerator / denominator + new_log_stds - old_log_stds, axis=-1) # more lossy version # return TT.sum( # numerator / denominator + TT.log(new_std) - TT.log(old_std ), axis=-1) def kl_sym(self, old_dist_info_vars, new_dist_info_vars): old_means = old_dist_info_vars["mean"] old_log_stds = old_dist_info_vars["log_std"] new_means = new_dist_info_vars["mean"] new_log_stds = new_dist_info_vars["log_std"] """ Compute the KL divergence of two multivariate Gaussian distribution with diagonal covariance matrices """ old_std = tf.exp(old_log_stds) new_std = tf.exp(new_log_stds) # means: (N*A) # std: (N*A) # formula: # { (\mu_1 - \mu_2)^2 + \sigma_1^2 - \sigma_2^2 } / (2\sigma_2^2) + # ln(\sigma_2/\sigma_1) numerator = tf.square(old_means - new_means) + \ tf.square(old_std) - tf.square(new_std) denominator = 2 * tf.square(new_std) + 1e-8 return tf.reduce_sum( numerator / denominator + new_log_stds - old_log_stds, axis=-1) def likelihood_ratio_sym(self, x_var, old_dist_info_vars, new_dist_info_vars): logli_new = self.log_likelihood_sym(x_var, new_dist_info_vars) logli_old = self.log_likelihood_sym(x_var, old_dist_info_vars) return tf.exp(logli_new - logli_old) def log_likelihood_sym(self, x_var, dist_info_vars): means = dist_info_vars["mean"] log_stds = dist_info_vars["log_std"] zs = (x_var - means) / tf.exp(log_stds) return - tf.reduce_sum(log_stds, axis=-1) - \ 0.5 * tf.reduce_sum(tf.square(zs), axis=-1) - \ 0.5 * self.dim * np.log(2 * np.pi) def sample(self, dist_info): means = dist_info["mean"] log_stds = dist_info["log_std"] rnd = np.random.normal(size=means.shape) return rnd * np.exp(log_stds) + means def log_likelihood(self, xs, dist_info): means = dist_info["mean"] log_stds = dist_info["log_std"] zs = (xs - means) / np.exp(log_stds) return - np.sum(log_stds, axis=-1) - \ 0.5 * np.sum(np.square(zs), axis=-1) - \ 0.5 * self.dim * np.log(2 * np.pi) def entropy(self, dist_info): log_stds = dist_info["log_std"] return np.sum(log_stds + np.log(np.sqrt(2 * np.pi * np.e)), axis=-1) @property def dist_info_specs(self): return [("mean", (self.dim,)), ("log_std", (self.dim,))]
3,627
36.020408
84
py
rllab
rllab-master/sandbox/rocky/tf/distributions/bernoulli.py
from .base import Distribution import tensorflow as tf import numpy as np TINY = 1e-8 class Bernoulli(Distribution): def __init__(self, dim): self._dim = dim @property def dim(self): return self._dim def kl_sym(self, old_dist_info_vars, new_dist_info_vars): old_p = old_dist_info_vars["p"] new_p = new_dist_info_vars["p"] kl = old_p * (tf.log(old_p + TINY) - tf.log(new_p + TINY)) + \ (1 - old_p) * (tf.log(1 - old_p + TINY) - tf.log(1 - new_p + TINY)) ndims = kl.get_shape().ndims return tf.reduce_sum(kl, axis=ndims - 1) def kl(self, old_dist_info, new_dist_info): old_p = old_dist_info["p"] new_p = new_dist_info["p"] kl = old_p * (np.log(old_p + TINY) - np.log(new_p + TINY)) + \ (1 - old_p) * (np.log(1 - old_p + TINY) - np.log(1 - new_p + TINY)) return np.sum(kl, axis=-1) def sample(self, dist_info): p = np.asarray(dist_info["p"]) return np.cast['int'](np.random.uniform(low=0., high=1., size=p.shape) < p) def likelihood_ratio_sym(self, x_var, old_dist_info_vars, new_dist_info_vars): old_p = old_dist_info_vars["p"] new_p = new_dist_info_vars["p"] ndims = old_p.get_shape().ndims return tf.reduce_prod(x_var * new_p / (old_p + TINY) + (1 - x_var) * (1 - new_p) / (1 - old_p + TINY), axis=ndims - 1) def log_likelihood_sym(self, x_var, dist_info_vars): p = dist_info_vars["p"] ndims = p.get_shape().ndims return tf.reduce_sum(x_var * tf.log(p + TINY) + (1 - x_var) * tf.log(1 - p + TINY), axis=ndims - 1) def log_likelihood(self, xs, dist_info): p = dist_info["p"] return np.sum(xs * np.log(p + TINY) + (1 - xs) * np.log(1 - p + TINY), axis=-1) def entropy(self, dist_info): p = dist_info["p"] return np.sum(- p * np.log(p + TINY) - (1 - p) * np.log(1 - p + TINY), axis=-1) @property def dist_info_keys(self): return ["p"]
2,050
33.183333
110
py
rllab
rllab-master/sandbox/rocky/tf/policies/base.py
from sandbox.rocky.tf.core.parameterized import Parameterized class Policy(Parameterized): def __init__(self, env_spec): Parameterized.__init__(self) self._env_spec = env_spec # Should be implemented by all policies def get_action(self, observation): raise NotImplementedError def get_actions(self, observations): raise NotImplementedError def reset(self, dones=None): pass @property def vectorized(self): """ Indicates whether the policy is vectorized. If True, it should implement get_actions(), and support resetting with multiple simultaneous states. """ return False @property def observation_space(self): return self._env_spec.observation_space @property def action_space(self): return self._env_spec.action_space @property def env_spec(self): return self._env_spec @property def recurrent(self): """ Indicates whether the policy is recurrent. :return: """ return False def log_diagnostics(self, paths): """ Log extra information per iteration based on the collected paths """ pass @property def state_info_keys(self): """ Return keys for the information related to the policy's state when taking an action. :return: """ return [k for k, _ in self.state_info_specs] @property def state_info_specs(self): """ Return keys and shapes for the information related to the policy's state when taking an action. :return: """ return list() def terminate(self): """ Clean up operation """ pass class StochasticPolicy(Policy): @property def distribution(self): """ :rtype Distribution """ raise NotImplementedError def dist_info_sym(self, obs_var, state_info_vars): """ Return the symbolic distribution information about the actions. :param obs_var: symbolic variable for observations :param state_info_vars: a dictionary whose values should contain information about the state of the policy at the time it received the observation :return: """ raise NotImplementedError def dist_info(self, obs, state_infos): """ Return the distribution information about the actions. :param obs_var: observation values :param state_info_vars: a dictionary whose values should contain information about the state of the policy at the time it received the observation :return: """ raise NotImplementedError
2,755
24.757009
117
py
rllab
rllab-master/sandbox/rocky/tf/policies/uniform_control_policy.py
from sandbox.rocky.tf.policies.base import Policy from rllab.core.serializable import Serializable class UniformControlPolicy(Policy, Serializable): def __init__( self, env_spec, ): Serializable.quick_init(self, locals()) super(UniformControlPolicy, self).__init__(env_spec=env_spec) @property def vectorized(self): return True def get_action(self, observation): return self.action_space.sample(), dict() def get_actions(self, observations): return self.action_space.sample_n(len(observations)), dict() def get_params_internal(self, **tags): return []
658
25.36
69
py
rllab
rllab-master/sandbox/rocky/tf/policies/categorical_gru_policy.py
import numpy as np import sandbox.rocky.tf.core.layers as L import tensorflow as tf from sandbox.rocky.tf.core.layers_powered import LayersPowered from sandbox.rocky.tf.core.network import GRUNetwork, MLP from sandbox.rocky.tf.distributions.recurrent_categorical import RecurrentCategorical from sandbox.rocky.tf.misc import tensor_utils from sandbox.rocky.tf.spaces.discrete import Discrete from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.core.serializable import Serializable from rllab.misc import special from rllab.misc.overrides import overrides class CategoricalGRUPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_dim=32, feature_network=None, state_include_action=True, hidden_nonlinearity=tf.tanh, gru_layer_cls=L.GRULayer, ): """ :param env_spec: A spec for the env. :param hidden_dim: dimension of hidden layer :param hidden_nonlinearity: nonlinearity used for each hidden layer :return: """ with tf.variable_scope(name): assert isinstance(env_spec.action_space, Discrete) Serializable.quick_init(self, locals()) super(CategoricalGRUPolicy, self).__init__(env_spec) obs_dim = env_spec.observation_space.flat_dim action_dim = env_spec.action_space.flat_dim if state_include_action: input_dim = obs_dim + action_dim else: input_dim = obs_dim l_input = L.InputLayer( shape=(None, None, input_dim), name="input" ) if feature_network is None: feature_dim = input_dim l_flat_feature = None l_feature = l_input else: feature_dim = feature_network.output_layer.output_shape[-1] l_flat_feature = feature_network.output_layer l_feature = L.OpLayer( l_flat_feature, extras=[l_input], name="reshape_feature", op=lambda flat_feature, input: tf.reshape( flat_feature, tf.stack([tf.shape(input)[0], tf.shape(input)[1], feature_dim]) ), shape_op=lambda _, input_shape: (input_shape[0], input_shape[1], feature_dim) ) prob_network = GRUNetwork( input_shape=(feature_dim,), input_layer=l_feature, output_dim=env_spec.action_space.n, hidden_dim=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=tf.nn.softmax, gru_layer_cls=gru_layer_cls, name="prob_network" ) self.prob_network = prob_network self.feature_network = feature_network self.l_input = l_input self.state_include_action = state_include_action flat_input_var = tf.placeholder(dtype=tf.float32, shape=(None, input_dim), name="flat_input") if feature_network is None: feature_var = flat_input_var else: feature_var = L.get_output(l_flat_feature, {feature_network.input_layer: flat_input_var}) self.f_step_prob = tensor_utils.compile_function( [ flat_input_var, prob_network.step_prev_hidden_layer.input_var ], L.get_output([ prob_network.step_output_layer, prob_network.step_hidden_layer ], {prob_network.step_input_layer: feature_var}) ) self.input_dim = input_dim self.action_dim = action_dim self.hidden_dim = hidden_dim self.prev_actions = None self.prev_hiddens = None self.dist = RecurrentCategorical(env_spec.action_space.n) out_layers = [prob_network.output_layer] if feature_network is not None: out_layers.append(feature_network.output_layer) LayersPowered.__init__(self, out_layers) @overrides def dist_info_sym(self, obs_var, state_info_vars): n_batches = tf.shape(obs_var)[0] n_steps = tf.shape(obs_var)[1] obs_var = tf.reshape(obs_var, tf.stack([n_batches, n_steps, -1])) obs_var = tf.cast(obs_var, tf.float32) if self.state_include_action: prev_action_var = tf.cast(state_info_vars["prev_action"], tf.float32) all_input_var = tf.concat(axis=2, values=[obs_var, prev_action_var]) else: all_input_var = obs_var if self.feature_network is None: return dict( prob=L.get_output( self.prob_network.output_layer, {self.l_input: all_input_var} ) ) else: flat_input_var = tf.reshape(all_input_var, (-1, self.input_dim)) return dict( prob=L.get_output( self.prob_network.output_layer, {self.l_input: all_input_var, self.feature_network.input_layer: flat_input_var} ) ) @property def vectorized(self): return True def reset(self, dones=None): if dones is None: dones = [True] dones = np.asarray(dones) if self.prev_actions is None or len(dones) != len(self.prev_actions): self.prev_actions = np.zeros((len(dones), self.action_space.flat_dim)) self.prev_hiddens = np.zeros((len(dones), self.hidden_dim)) self.prev_actions[dones] = 0. self.prev_hiddens[dones] = self.prob_network.hid_init_param.eval() # get_value() # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): actions, agent_infos = self.get_actions([observation]) return actions[0], {k: v[0] for k, v in agent_infos.items()} @overrides def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) if self.state_include_action: assert self.prev_actions is not None all_input = np.concatenate([ flat_obs, self.prev_actions ], axis=-1) else: all_input = flat_obs probs, hidden_vec = self.f_step_prob(all_input, self.prev_hiddens) actions = special.weighted_sample_n(probs, np.arange(self.action_space.n)) prev_actions = self.prev_actions self.prev_actions = self.action_space.flatten_n(actions) self.prev_hiddens = hidden_vec agent_info = dict(prob=probs) if self.state_include_action: agent_info["prev_action"] = np.copy(prev_actions) return actions, agent_info @property @overrides def recurrent(self): return True @property def distribution(self): return self.dist @property def state_info_specs(self): if self.state_include_action: return [ ("prev_action", (self.action_dim,)), ] else: return []
7,649
36.317073
105
py
rllab
rllab-master/sandbox/rocky/tf/policies/categorical_mlp_policy.py
from sandbox.rocky.tf.core.layers_powered import LayersPowered import sandbox.rocky.tf.core.layers as L from sandbox.rocky.tf.core.network import MLP from rllab.core.serializable import Serializable from sandbox.rocky.tf.distributions.categorical import Categorical from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.misc import ext from sandbox.rocky.tf.misc import tensor_utils from rllab.misc.overrides import overrides from sandbox.rocky.tf.spaces.discrete import Discrete import tensorflow as tf class CategoricalMLPPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_sizes=(32, 32), hidden_nonlinearity=tf.nn.tanh, prob_network=None, ): """ :param env_spec: A spec for the mdp. :param hidden_sizes: list of sizes for the fully connected hidden layers :param hidden_nonlinearity: nonlinearity used for each hidden layer :param prob_network: manually specified network for this policy, other network params are ignored :return: """ Serializable.quick_init(self, locals()) assert isinstance(env_spec.action_space, Discrete) with tf.variable_scope(name): if prob_network is None: prob_network = MLP( input_shape=(env_spec.observation_space.flat_dim,), output_dim=env_spec.action_space.n, hidden_sizes=hidden_sizes, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=tf.nn.softmax, name="prob_network", ) self._l_prob = prob_network.output_layer self._l_obs = prob_network.input_layer self._f_prob = tensor_utils.compile_function( [prob_network.input_layer.input_var], L.get_output(prob_network.output_layer) ) self._dist = Categorical(env_spec.action_space.n) super(CategoricalMLPPolicy, self).__init__(env_spec) LayersPowered.__init__(self, [prob_network.output_layer]) @property def vectorized(self): return True @overrides def dist_info_sym(self, obs_var, state_info_vars=None): return dict(prob=L.get_output(self._l_prob, {self._l_obs: tf.cast(obs_var, tf.float32)})) @overrides def dist_info(self, obs, state_infos=None): return dict(prob=self._f_prob(obs)) # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): flat_obs = self.observation_space.flatten(observation) prob = self._f_prob([flat_obs])[0] action = self.action_space.weighted_sample(prob) return action, dict(prob=prob) def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) probs = self._f_prob(flat_obs) actions = list(map(self.action_space.weighted_sample, probs)) return actions, dict(prob=probs) @property def distribution(self): return self._dist
3,395
36.733333
97
py
rllab
rllab-master/sandbox/rocky/tf/policies/categorical_lstm_policy.py
import numpy as np import sandbox.rocky.tf.core.layers as L import tensorflow as tf from sandbox.rocky.tf.core.layers_powered import LayersPowered from sandbox.rocky.tf.core.network import LSTMNetwork, MLP from sandbox.rocky.tf.distributions.recurrent_categorical import RecurrentCategorical from sandbox.rocky.tf.misc import tensor_utils from sandbox.rocky.tf.spaces.discrete import Discrete from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.core.serializable import Serializable from rllab.misc import special from rllab.misc.overrides import overrides class CategoricalLSTMPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_dim=32, feature_network=None, prob_network=None, state_include_action=True, hidden_nonlinearity=tf.tanh, forget_bias=1.0, use_peepholes=False, lstm_layer_cls=L.LSTMLayer ): """ :param env_spec: A spec for the env. :param hidden_dim: dimension of hidden layer :param hidden_nonlinearity: nonlinearity used for each hidden layer :return: """ with tf.variable_scope(name): assert isinstance(env_spec.action_space, Discrete) Serializable.quick_init(self, locals()) super(CategoricalLSTMPolicy, self).__init__(env_spec) obs_dim = env_spec.observation_space.flat_dim action_dim = env_spec.action_space.flat_dim if state_include_action: input_dim = obs_dim + action_dim else: input_dim = obs_dim l_input = L.InputLayer( shape=(None, None, input_dim), name="input" ) if feature_network is None: feature_dim = input_dim l_flat_feature = None l_feature = l_input else: feature_dim = feature_network.output_layer.output_shape[-1] l_flat_feature = feature_network.output_layer l_feature = L.OpLayer( l_flat_feature, extras=[l_input], name="reshape_feature", op=lambda flat_feature, input: tf.reshape( flat_feature, tf.stack([tf.shape(input)[0], tf.shape(input)[1], feature_dim]) ), shape_op=lambda _, input_shape: (input_shape[0], input_shape[1], feature_dim) ) if prob_network is None: prob_network = LSTMNetwork( input_shape=(feature_dim,), input_layer=l_feature, output_dim=env_spec.action_space.n, hidden_dim=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=tf.nn.softmax, forget_bias=forget_bias, use_peepholes=use_peepholes, lstm_layer_cls=lstm_layer_cls, name="prob_network" ) self.prob_network = prob_network self.feature_network = feature_network self.l_input = l_input self.state_include_action = state_include_action flat_input_var = tf.placeholder(dtype=tf.float32, shape=(None, input_dim), name="flat_input") if feature_network is None: feature_var = flat_input_var else: feature_var = L.get_output(l_flat_feature, {feature_network.input_layer: flat_input_var}) self.f_step_prob = tensor_utils.compile_function( [ flat_input_var, prob_network.step_prev_hidden_layer.input_var, prob_network.step_prev_cell_layer.input_var ], L.get_output([ prob_network.step_output_layer, prob_network.step_hidden_layer, prob_network.step_cell_layer ], {prob_network.step_input_layer: feature_var}) ) self.input_dim = input_dim self.action_dim = action_dim self.hidden_dim = hidden_dim self.prev_actions = None self.prev_hiddens = None self.prev_cells = None self.dist = RecurrentCategorical(env_spec.action_space.n) out_layers = [prob_network.output_layer] if feature_network is not None: out_layers.append(feature_network.output_layer) LayersPowered.__init__(self, out_layers) @overrides def dist_info_sym(self, obs_var, state_info_vars): n_batches = tf.shape(obs_var)[0] n_steps = tf.shape(obs_var)[1] obs_var = tf.reshape(obs_var, tf.stack([n_batches, n_steps, -1])) obs_var = tf.cast(obs_var, tf.float32) if self.state_include_action: prev_action_var = state_info_vars["prev_action"] prev_action_var = tf.cast(prev_action_var, tf.float32) all_input_var = tf.concat(axis=2, values=[obs_var, prev_action_var]) else: all_input_var = obs_var if self.feature_network is None: return dict( prob=L.get_output( self.prob_network.output_layer, {self.l_input: all_input_var} ) ) else: flat_input_var = tf.reshape(all_input_var, (-1, self.input_dim)) return dict( prob=L.get_output( self.prob_network.output_layer, {self.l_input: all_input_var, self.feature_network.input_layer: flat_input_var} ) ) @property def vectorized(self): return True def reset(self, dones=None): if dones is None: dones = [True] dones = np.asarray(dones) if self.prev_actions is None or len(dones) != len(self.prev_actions): self.prev_actions = np.zeros((len(dones), self.action_space.flat_dim)) self.prev_hiddens = np.zeros((len(dones), self.hidden_dim)) self.prev_cells = np.zeros((len(dones), self.hidden_dim)) self.prev_actions[dones] = 0. self.prev_hiddens[dones] = self.prob_network.hid_init_param.eval() self.prev_cells[dones] = self.prob_network.cell_init_param.eval() # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): actions, agent_infos = self.get_actions([observation]) return actions[0], {k: v[0] for k, v in agent_infos.items()} @overrides def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) if self.state_include_action: assert self.prev_actions is not None all_input = np.concatenate([ flat_obs, self.prev_actions ], axis=-1) else: all_input = flat_obs probs, hidden_vec, cell_vec = self.f_step_prob(all_input, self.prev_hiddens, self.prev_cells) actions = special.weighted_sample_n(probs, np.arange(self.action_space.n)) prev_actions = self.prev_actions self.prev_actions = self.action_space.flatten_n(actions) self.prev_hiddens = hidden_vec self.prev_cells = cell_vec agent_info = dict(prob=probs) if self.state_include_action: agent_info["prev_action"] = np.copy(prev_actions) return actions, agent_info @property @overrides def recurrent(self): return True @property def distribution(self): return self.dist @property def state_info_specs(self): if self.state_include_action: return [ ("prev_action", (self.action_dim,)), ] else: return []
8,307
37.110092
105
py
rllab
rllab-master/sandbox/rocky/tf/policies/gaussian_mlp_policy.py
import numpy as np from sandbox.rocky.tf.core.layers_powered import LayersPowered import sandbox.rocky.tf.core.layers as L from sandbox.rocky.tf.core.network import MLP from sandbox.rocky.tf.spaces.box import Box from rllab.core.serializable import Serializable from sandbox.rocky.tf.policies.base import StochasticPolicy from sandbox.rocky.tf.distributions.diagonal_gaussian import DiagonalGaussian from rllab.misc.overrides import overrides from rllab.misc import logger from sandbox.rocky.tf.misc import tensor_utils import tensorflow as tf class GaussianMLPPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_sizes=(32, 32), learn_std=True, init_std=1.0, adaptive_std=False, std_share_network=False, std_hidden_sizes=(32, 32), min_std=1e-6, std_hidden_nonlinearity=tf.nn.tanh, hidden_nonlinearity=tf.nn.tanh, output_nonlinearity=None, mean_network=None, std_network=None, std_parametrization='exp' ): """ :param env_spec: :param hidden_sizes: list of sizes for the fully-connected hidden layers :param learn_std: Is std trainable :param init_std: Initial std :param adaptive_std: :param std_share_network: :param std_hidden_sizes: list of sizes for the fully-connected layers for std :param min_std: whether to make sure that the std is at least some threshold value, to avoid numerical issues :param std_hidden_nonlinearity: :param hidden_nonlinearity: nonlinearity used for each hidden layer :param output_nonlinearity: nonlinearity for the output layer :param mean_network: custom network for the output mean :param std_network: custom network for the output log std :param std_parametrization: how the std should be parametrized. There are a few options: - exp: the logarithm of the std will be stored, and applied a exponential transformation - softplus: the std will be computed as log(1+exp(x)) :return: """ Serializable.quick_init(self, locals()) assert isinstance(env_spec.action_space, Box) with tf.variable_scope(name): obs_dim = env_spec.observation_space.flat_dim action_dim = env_spec.action_space.flat_dim # create network if mean_network is None: mean_network = MLP( name="mean_network", input_shape=(obs_dim,), output_dim=action_dim, hidden_sizes=hidden_sizes, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=output_nonlinearity, ) self._mean_network = mean_network l_mean = mean_network.output_layer obs_var = mean_network.input_layer.input_var if std_network is not None: l_std_param = std_network.output_layer else: if adaptive_std: std_network = MLP( name="std_network", input_shape=(obs_dim,), input_layer=mean_network.input_layer, output_dim=action_dim, hidden_sizes=std_hidden_sizes, hidden_nonlinearity=std_hidden_nonlinearity, output_nonlinearity=None, ) l_std_param = std_network.output_layer else: if std_parametrization == 'exp': init_std_param = np.log(init_std) elif std_parametrization == 'softplus': init_std_param = np.log(np.exp(init_std) - 1) else: raise NotImplementedError l_std_param = L.ParamLayer( mean_network.input_layer, num_units=action_dim, param=tf.constant_initializer(init_std_param), name="output_std_param", trainable=learn_std, ) self.std_parametrization = std_parametrization if std_parametrization == 'exp': min_std_param = np.log(min_std) elif std_parametrization == 'softplus': min_std_param = np.log(np.exp(min_std) - 1) else: raise NotImplementedError self.min_std_param = min_std_param # mean_var, log_std_var = L.get_output([l_mean, l_std_param]) # # if self.min_std_param is not None: # log_std_var = tf.maximum(log_std_var, np.log(min_std)) # # self._mean_var, self._log_std_var = mean_var, log_std_var self._l_mean = l_mean self._l_std_param = l_std_param self._dist = DiagonalGaussian(action_dim) LayersPowered.__init__(self, [l_mean, l_std_param]) super(GaussianMLPPolicy, self).__init__(env_spec) dist_info_sym = self.dist_info_sym(mean_network.input_layer.input_var, dict()) mean_var = dist_info_sym["mean"] log_std_var = dist_info_sym["log_std"] self._f_dist = tensor_utils.compile_function( inputs=[obs_var], outputs=[mean_var, log_std_var], ) @property def vectorized(self): return True def dist_info_sym(self, obs_var, state_info_vars=None): mean_var, std_param_var = L.get_output([self._l_mean, self._l_std_param], obs_var) if self.min_std_param is not None: std_param_var = tf.maximum(std_param_var, self.min_std_param) if self.std_parametrization == 'exp': log_std_var = std_param_var elif self.std_parametrization == 'softplus': log_std_var = tf.log(tf.log(1. + tf.exp(std_param_var))) else: raise NotImplementedError return dict(mean=mean_var, log_std=log_std_var) @overrides def get_action(self, observation): flat_obs = self.observation_space.flatten(observation) mean, log_std = [x[0] for x in self._f_dist([flat_obs])] rnd = np.random.normal(size=mean.shape) action = rnd * np.exp(log_std) + mean return action, dict(mean=mean, log_std=log_std) def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) means, log_stds = self._f_dist(flat_obs) rnd = np.random.normal(size=means.shape) actions = rnd * np.exp(log_stds) + means return actions, dict(mean=means, log_std=log_stds) def get_reparam_action_sym(self, obs_var, action_var, old_dist_info_vars): """ Given observations, old actions, and distribution of old actions, return a symbolically reparameterized representation of the actions in terms of the policy parameters :param obs_var: :param action_var: :param old_dist_info_vars: :return: """ new_dist_info_vars = self.dist_info_sym(obs_var, action_var) new_mean_var, new_log_std_var = new_dist_info_vars["mean"], new_dist_info_vars["log_std"] old_mean_var, old_log_std_var = old_dist_info_vars["mean"], old_dist_info_vars["log_std"] epsilon_var = (action_var - old_mean_var) / (tf.exp(old_log_std_var) + 1e-8) new_action_var = new_mean_var + epsilon_var * tf.exp(new_log_std_var) return new_action_var def log_diagnostics(self, paths): log_stds = np.vstack([path["agent_infos"]["log_std"] for path in paths]) logger.record_tabular('AveragePolicyStd', np.mean(np.exp(log_stds))) @property def distribution(self): return self._dist
8,050
40.076531
117
py
rllab
rllab-master/sandbox/rocky/tf/policies/gaussian_lstm_policy.py
import numpy as np import sandbox.rocky.tf.core.layers as L import tensorflow as tf from sandbox.rocky.tf.core.layers_powered import LayersPowered from sandbox.rocky.tf.core.network import LSTMNetwork from sandbox.rocky.tf.distributions.recurrent_diagonal_gaussian import RecurrentDiagonalGaussian from sandbox.rocky.tf.misc import tensor_utils from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.core.serializable import Serializable from rllab.misc.overrides import overrides class GaussianLSTMPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_dim=32, feature_network=None, state_include_action=True, hidden_nonlinearity=tf.tanh, learn_std=True, init_std=1.0, output_nonlinearity=None, lstm_layer_cls=L.LSTMLayer, use_peepholes=False, ): """ :param env_spec: A spec for the env. :param hidden_dim: dimension of hidden layer :param hidden_nonlinearity: nonlinearity used for each hidden layer :return: """ with tf.variable_scope(name): Serializable.quick_init(self, locals()) super(GaussianLSTMPolicy, self).__init__(env_spec) obs_dim = env_spec.observation_space.flat_dim action_dim = env_spec.action_space.flat_dim if state_include_action: input_dim = obs_dim + action_dim else: input_dim = obs_dim l_input = L.InputLayer( shape=(None, None, input_dim), name="input" ) if feature_network is None: feature_dim = input_dim l_flat_feature = None l_feature = l_input else: feature_dim = feature_network.output_layer.output_shape[-1] l_flat_feature = feature_network.output_layer l_feature = L.OpLayer( l_flat_feature, extras=[l_input], name="reshape_feature", op=lambda flat_feature, input: tf.reshape( flat_feature, tf.stack([tf.shape(input)[0], tf.shape(input)[1], feature_dim]) ), shape_op=lambda _, input_shape: (input_shape[0], input_shape[1], feature_dim) ) mean_network = LSTMNetwork( input_shape=(feature_dim,), input_layer=l_feature, output_dim=action_dim, hidden_dim=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=output_nonlinearity, lstm_layer_cls=lstm_layer_cls, name="mean_network", use_peepholes=use_peepholes, ) l_log_std = L.ParamLayer( mean_network.input_layer, num_units=action_dim, param=tf.constant_initializer(np.log(init_std)), name="output_log_std", trainable=learn_std, ) l_step_log_std = L.ParamLayer( mean_network.step_input_layer, num_units=action_dim, param=l_log_std.param, name="step_output_log_std", trainable=learn_std, ) self.mean_network = mean_network self.feature_network = feature_network self.l_input = l_input self.state_include_action = state_include_action flat_input_var = tf.placeholder(dtype=tf.float32, shape=(None, input_dim), name="flat_input") if feature_network is None: feature_var = flat_input_var else: feature_var = L.get_output(l_flat_feature, {feature_network.input_layer: flat_input_var}) self.f_step_mean_std = tensor_utils.compile_function( [ flat_input_var, mean_network.step_prev_state_layer.input_var, ], L.get_output([ mean_network.step_output_layer, l_step_log_std, mean_network.step_hidden_layer, mean_network.step_cell_layer ], {mean_network.step_input_layer: feature_var}) ) self.l_log_std = l_log_std self.input_dim = input_dim self.action_dim = action_dim self.hidden_dim = hidden_dim self.prev_actions = None self.prev_hiddens = None self.prev_cells = None self.dist = RecurrentDiagonalGaussian(action_dim) out_layers = [mean_network.output_layer, l_log_std] if feature_network is not None: out_layers.append(feature_network.output_layer) LayersPowered.__init__(self, out_layers) @overrides def dist_info_sym(self, obs_var, state_info_vars): n_batches = tf.shape(obs_var)[0] n_steps = tf.shape(obs_var)[1] obs_var = tf.reshape(obs_var, tf.stack([n_batches, n_steps, -1])) if self.state_include_action: prev_action_var = state_info_vars["prev_action"] all_input_var = tf.concat(axis=2, values=[obs_var, prev_action_var]) else: all_input_var = obs_var if self.feature_network is None: means, log_stds = L.get_output( [self.mean_network.output_layer, self.l_log_std], {self.l_input: all_input_var} ) else: flat_input_var = tf.reshape(all_input_var, (-1, self.input_dim)) means, log_stds = L.get_output( [self.mean_network.output_layer, self.l_log_std], {self.l_input: all_input_var, self.feature_network.input_layer: flat_input_var} ) return dict(mean=means, log_std=log_stds) @property def vectorized(self): return True def reset(self, dones=None): if dones is None: dones = [True] dones = np.asarray(dones) if self.prev_actions is None or len(dones) != len(self.prev_actions): self.prev_actions = np.zeros((len(dones), self.action_space.flat_dim)) self.prev_hiddens = np.zeros((len(dones), self.hidden_dim)) self.prev_cells = np.zeros((len(dones), self.hidden_dim)) self.prev_actions[dones] = 0. self.prev_hiddens[dones] = self.mean_network.hid_init_param.eval() self.prev_cells[dones] = self.mean_network.cell_init_param.eval() # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): actions, agent_infos = self.get_actions([observation]) return actions[0], {k: v[0] for k, v in agent_infos.items()} @overrides def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) if self.state_include_action: assert self.prev_actions is not None all_input = np.concatenate([ flat_obs, self.prev_actions ], axis=-1) else: all_input = flat_obs # probs, hidden_vec, cell_vec = self.f_step_prob(all_input, self.prev_hiddens, self.prev_cells) means, log_stds, hidden_vec, cell_vec = self.f_step_mean_std( all_input, np.hstack([self.prev_hiddens, self.prev_cells])) rnd = np.random.normal(size=means.shape) actions = rnd * np.exp(log_stds) + means prev_actions = self.prev_actions self.prev_actions = self.action_space.flatten_n(actions) self.prev_hiddens = hidden_vec self.prev_cells = cell_vec agent_info = dict(mean=means, log_std=log_stds) if self.state_include_action: agent_info["prev_action"] = np.copy(prev_actions) return actions, agent_info @property @overrides def recurrent(self): return True @property def distribution(self): return self.dist @property def state_info_specs(self): if self.state_include_action: return [ ("prev_action", (self.action_dim,)), ] else: return []
8,680
36.743478
105
py
rllab
rllab-master/sandbox/rocky/tf/policies/gaussian_gru_policy.py
import numpy as np import sandbox.rocky.tf.core.layers as L import tensorflow as tf from sandbox.rocky.tf.core.layers_powered import LayersPowered from sandbox.rocky.tf.core.network import GRUNetwork from sandbox.rocky.tf.distributions.recurrent_diagonal_gaussian import RecurrentDiagonalGaussian from sandbox.rocky.tf.misc import tensor_utils from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.core.serializable import Serializable from rllab.misc.overrides import overrides from rllab.misc import logger class GaussianGRUPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_dim=32, feature_network=None, state_include_action=True, hidden_nonlinearity=tf.tanh, gru_layer_cls=L.GRULayer, learn_std=True, init_std=1.0, output_nonlinearity=None, ): """ :param env_spec: A spec for the env. :param hidden_dim: dimension of hidden layer :param hidden_nonlinearity: nonlinearity used for each hidden layer :return: """ with tf.variable_scope(name): Serializable.quick_init(self, locals()) super(GaussianGRUPolicy, self).__init__(env_spec) obs_dim = env_spec.observation_space.flat_dim action_dim = env_spec.action_space.flat_dim if state_include_action: input_dim = obs_dim + action_dim else: input_dim = obs_dim l_input = L.InputLayer( shape=(None, None, input_dim), name="input" ) if feature_network is None: feature_dim = input_dim l_flat_feature = None l_feature = l_input else: feature_dim = feature_network.output_layer.output_shape[-1] l_flat_feature = feature_network.output_layer l_feature = L.OpLayer( l_flat_feature, extras=[l_input], name="reshape_feature", op=lambda flat_feature, input: tf.reshape( flat_feature, tf.stack([tf.shape(input)[0], tf.shape(input)[1], feature_dim]) ), shape_op=lambda _, input_shape: (input_shape[0], input_shape[1], feature_dim) ) mean_network = GRUNetwork( input_shape=(feature_dim,), input_layer=l_feature, output_dim=action_dim, hidden_dim=hidden_dim, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=output_nonlinearity, gru_layer_cls=gru_layer_cls, name="mean_network" ) l_log_std = L.ParamLayer( mean_network.input_layer, num_units=action_dim, param=tf.constant_initializer(np.log(init_std)), name="output_log_std", trainable=learn_std, ) l_step_log_std = L.ParamLayer( mean_network.step_input_layer, num_units=action_dim, param=l_log_std.param, name="step_output_log_std", trainable=learn_std, ) self.mean_network = mean_network self.feature_network = feature_network self.l_input = l_input self.state_include_action = state_include_action flat_input_var = tf.placeholder(dtype=tf.float32, shape=(None, input_dim), name="flat_input") if feature_network is None: feature_var = flat_input_var else: feature_var = L.get_output(l_flat_feature, {feature_network.input_layer: flat_input_var}) self.f_step_mean_std = tensor_utils.compile_function( [ flat_input_var, mean_network.step_prev_state_layer.input_var, ], L.get_output([ mean_network.step_output_layer, l_step_log_std, mean_network.step_hidden_layer, ], {mean_network.step_input_layer: feature_var}) ) self.l_log_std = l_log_std self.input_dim = input_dim self.action_dim = action_dim self.hidden_dim = hidden_dim self.prev_actions = None self.prev_hiddens = None self.dist = RecurrentDiagonalGaussian(action_dim) out_layers = [mean_network.output_layer, l_log_std, l_step_log_std] if feature_network is not None: out_layers.append(feature_network.output_layer) LayersPowered.__init__(self, out_layers) @overrides def dist_info_sym(self, obs_var, state_info_vars): n_batches = tf.shape(obs_var)[0] n_steps = tf.shape(obs_var)[1] obs_var = tf.reshape(obs_var, tf.stack([n_batches, n_steps, -1])) if self.state_include_action: prev_action_var = state_info_vars["prev_action"] all_input_var = tf.concat(axis=2, values=[obs_var, prev_action_var]) else: all_input_var = obs_var if self.feature_network is None: means, log_stds = L.get_output( [self.mean_network.output_layer, self.l_log_std], {self.l_input: all_input_var} ) else: flat_input_var = tf.reshape(all_input_var, (-1, self.input_dim)) means, log_stds = L.get_output( [self.mean_network.output_layer, self.l_log_std], {self.l_input: all_input_var, self.feature_network.input_layer: flat_input_var} ) return dict(mean=means, log_std=log_stds) @property def vectorized(self): return True def reset(self, dones=None): if dones is None: dones = [True] dones = np.asarray(dones) if self.prev_actions is None or len(dones) != len(self.prev_actions): self.prev_actions = np.zeros((len(dones), self.action_space.flat_dim)) self.prev_hiddens = np.zeros((len(dones), self.hidden_dim)) self.prev_actions[dones] = 0. self.prev_hiddens[dones] = self.mean_network.hid_init_param.eval() # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): actions, agent_infos = self.get_actions([observation]) return actions[0], {k: v[0] for k, v in agent_infos.items()} @overrides def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) if self.state_include_action: assert self.prev_actions is not None all_input = np.concatenate([ flat_obs, self.prev_actions ], axis=-1) else: all_input = flat_obs means, log_stds, hidden_vec = self.f_step_mean_std(all_input, self.prev_hiddens) rnd = np.random.normal(size=means.shape) actions = rnd * np.exp(log_stds) + means prev_actions = self.prev_actions self.prev_actions = self.action_space.flatten_n(actions) self.prev_hiddens = hidden_vec agent_info = dict(mean=means, log_std=log_stds) if self.state_include_action: agent_info["prev_action"] = np.copy(prev_actions) return actions, agent_info @property @overrides def recurrent(self): return True @property def distribution(self): return self.dist @property def state_info_specs(self): if self.state_include_action: return [ ("prev_action", (self.action_dim,)), ] else: return [] def log_diagnostics(self, paths): log_stds = np.vstack([path["agent_infos"]["log_std"] for path in paths]) logger.record_tabular('AveragePolicyStd', np.mean(np.exp(log_stds)))
8,416
36.243363
105
py
rllab
rllab-master/sandbox/rocky/tf/policies/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/policies/categorical_conv_policy.py
from sandbox.rocky.tf.core.layers_powered import LayersPowered import sandbox.rocky.tf.core.layers as L from sandbox.rocky.tf.core.network import ConvNetwork from rllab.core.serializable import Serializable from sandbox.rocky.tf.distributions.categorical import Categorical from sandbox.rocky.tf.policies.base import StochasticPolicy from rllab.misc import ext from sandbox.rocky.tf.misc import tensor_utils from rllab.misc.overrides import overrides from sandbox.rocky.tf.spaces.discrete import Discrete import tensorflow as tf class CategoricalConvPolicy(StochasticPolicy, LayersPowered, Serializable): def __init__( self, name, env_spec, conv_filters, conv_filter_sizes, conv_strides, conv_pads, hidden_sizes=[], hidden_nonlinearity=tf.nn.relu, output_nonlinearity=tf.nn.softmax, prob_network=None, ): """ :param env_spec: A spec for the mdp. :param hidden_sizes: list of sizes for the fully connected hidden layers :param hidden_nonlinearity: nonlinearity used for each hidden layer :param prob_network: manually specified network for this policy, other network params are ignored :return: """ Serializable.quick_init(self, locals()) assert isinstance(env_spec.action_space, Discrete) self._env_spec = env_spec # import pdb; pdb.set_trace() if prob_network is None: prob_network = ConvNetwork( input_shape=env_spec.observation_space.shape, output_dim=env_spec.action_space.n, conv_filters=conv_filters, conv_filter_sizes=conv_filter_sizes, conv_strides=conv_strides, conv_pads=conv_pads, hidden_sizes=hidden_sizes, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=output_nonlinearity, name="prob_network", ) self._l_prob = prob_network.output_layer self._l_obs = prob_network.input_layer self._f_prob = tensor_utils.compile_function( [prob_network.input_layer.input_var], L.get_output(prob_network.output_layer) ) self._dist = Categorical(env_spec.action_space.n) super(CategoricalConvPolicy, self).__init__(env_spec) LayersPowered.__init__(self, [prob_network.output_layer]) @property def vectorized(self): return True @overrides def dist_info_sym(self, obs_var, state_info_vars=None): return dict(prob=L.get_output(self._l_prob, {self._l_obs: tf.cast(obs_var, tf.float32)})) @overrides def dist_info(self, obs, state_infos=None): return dict(prob=self._f_prob(obs)) # The return value is a pair. The first item is a matrix (N, A), where each # entry corresponds to the action value taken. The second item is a vector # of length N, where each entry is the density value for that action, under # the current policy @overrides def get_action(self, observation): flat_obs = self.observation_space.flatten(observation) prob = self._f_prob([flat_obs])[0] action = self.action_space.weighted_sample(prob) return action, dict(prob=prob) def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) probs = self._f_prob(flat_obs) actions = list(map(self.action_space.weighted_sample, probs)) return actions, dict(prob=probs) @property def distribution(self): return self._dist
3,662
36.762887
97
py
rllab
rllab-master/sandbox/rocky/tf/policies/deterministic_mlp_policy.py
from rllab.core.serializable import Serializable from rllab.misc import ext from rllab.misc.overrides import overrides from sandbox.rocky.tf.core.layers_powered import LayersPowered from sandbox.rocky.tf.core.network import MLP from sandbox.rocky.tf.distributions.categorical import Categorical from sandbox.rocky.tf.policies.base import Policy from sandbox.rocky.tf.misc import tensor_utils import sandbox.rocky.tf.core.layers as L from sandbox.rocky.tf.core.layers import batch_norm from sandbox.rocky.tf.spaces.discrete import Discrete import tensorflow as tf class DeterministicMLPPolicy(Policy, LayersPowered, Serializable): def __init__( self, name, env_spec, hidden_sizes=(32, 32), hidden_nonlinearity=tf.nn.relu, output_nonlinearity=tf.nn.tanh, prob_network=None, bn=False): Serializable.quick_init(self, locals()) with tf.variable_scope(name): if prob_network is None: prob_network = MLP( input_shape=(env_spec.observation_space.flat_dim,), output_dim=env_spec.action_space.flat_dim, hidden_sizes=hidden_sizes, hidden_nonlinearity=hidden_nonlinearity, output_nonlinearity=output_nonlinearity, # batch_normalization=True, name="prob_network", ) self._l_prob = prob_network.output_layer self._l_obs = prob_network.input_layer self._f_prob = tensor_utils.compile_function( [prob_network.input_layer.input_var], L.get_output(prob_network.output_layer, deterministic=True) ) self.prob_network = prob_network # Note the deterministic=True argument. It makes sure that when getting # actions from single observations, we do not update params in the # batch normalization layers. # TODO: this doesn't currently work properly in the tf version so we leave out batch_norm super(DeterministicMLPPolicy, self).__init__(env_spec) LayersPowered.__init__(self, [prob_network.output_layer]) @property def vectorized(self): return True @overrides def get_action(self, observation): flat_obs = self.observation_space.flatten(observation) action = self._f_prob([flat_obs])[0] return action, dict() @overrides def get_actions(self, observations): flat_obs = self.observation_space.flatten_n(observations) actions = self._f_prob(flat_obs) return actions, dict() def get_action_sym(self, obs_var): return L.get_output(self.prob_network.output_layer, obs_var)
2,794
36.266667
97
py
rllab
rllab-master/sandbox/rocky/tf/algos/npo.py
from rllab.misc import ext from rllab.misc.overrides import overrides import rllab.misc.logger as logger from sandbox.rocky.tf.optimizers.penalty_lbfgs_optimizer import PenaltyLbfgsOptimizer from sandbox.rocky.tf.algos.batch_polopt import BatchPolopt from sandbox.rocky.tf.misc import tensor_utils import tensorflow as tf class NPO(BatchPolopt): """ Natural Policy Optimization. """ def __init__( self, optimizer=None, optimizer_args=None, step_size=0.01, **kwargs): if optimizer is None: if optimizer_args is None: optimizer_args = dict() optimizer = PenaltyLbfgsOptimizer(**optimizer_args) self.optimizer = optimizer self.step_size = step_size super(NPO, self).__init__(**kwargs) @overrides def init_opt(self): is_recurrent = int(self.policy.recurrent) obs_var = self.env.observation_space.new_tensor_variable( 'obs', extra_dims=1 + is_recurrent, ) action_var = self.env.action_space.new_tensor_variable( 'action', extra_dims=1 + is_recurrent, ) advantage_var = tensor_utils.new_tensor( 'advantage', ndim=1 + is_recurrent, dtype=tf.float32, ) dist = self.policy.distribution old_dist_info_vars = { k: tf.placeholder(tf.float32, shape=[None] * (1 + is_recurrent) + list(shape), name='old_%s' % k) for k, shape in dist.dist_info_specs } old_dist_info_vars_list = [old_dist_info_vars[k] for k in dist.dist_info_keys] state_info_vars = { k: tf.placeholder(tf.float32, shape=[None] * (1 + is_recurrent) + list(shape), name=k) for k, shape in self.policy.state_info_specs } state_info_vars_list = [state_info_vars[k] for k in self.policy.state_info_keys] if is_recurrent: valid_var = tf.placeholder(tf.float32, shape=[None, None], name="valid") else: valid_var = None dist_info_vars = self.policy.dist_info_sym(obs_var, state_info_vars) kl = dist.kl_sym(old_dist_info_vars, dist_info_vars) lr = dist.likelihood_ratio_sym(action_var, old_dist_info_vars, dist_info_vars) if is_recurrent: mean_kl = tf.reduce_sum(kl * valid_var) / tf.reduce_sum(valid_var) surr_loss = - tf.reduce_sum(lr * advantage_var * valid_var) / tf.reduce_sum(valid_var) else: mean_kl = tf.reduce_mean(kl) surr_loss = - tf.reduce_mean(lr * advantage_var) input_list = [ obs_var, action_var, advantage_var, ] + state_info_vars_list + old_dist_info_vars_list if is_recurrent: input_list.append(valid_var) self.optimizer.update_opt( loss=surr_loss, target=self.policy, leq_constraint=(mean_kl, self.step_size), inputs=input_list, constraint_name="mean_kl" ) return dict() @overrides def optimize_policy(self, itr, samples_data): all_input_values = tuple(ext.extract( samples_data, "observations", "actions", "advantages" )) agent_infos = samples_data["agent_infos"] state_info_list = [agent_infos[k] for k in self.policy.state_info_keys] dist_info_list = [agent_infos[k] for k in self.policy.distribution.dist_info_keys] all_input_values += tuple(state_info_list) + tuple(dist_info_list) if self.policy.recurrent: all_input_values += (samples_data["valids"],) logger.log("Computing loss before") loss_before = self.optimizer.loss(all_input_values) logger.log("Computing KL before") mean_kl_before = self.optimizer.constraint_val(all_input_values) logger.log("Optimizing") self.optimizer.optimize(all_input_values) logger.log("Computing KL after") mean_kl = self.optimizer.constraint_val(all_input_values) logger.log("Computing loss after") loss_after = self.optimizer.loss(all_input_values) logger.record_tabular('LossBefore', loss_before) logger.record_tabular('LossAfter', loss_after) logger.record_tabular('MeanKLBefore', mean_kl_before) logger.record_tabular('MeanKL', mean_kl) logger.record_tabular('dLoss', loss_before - loss_after) return dict() @overrides def get_itr_snapshot(self, itr, samples_data): return dict( itr=itr, policy=self.policy, baseline=self.baseline, env=self.env, )
4,814
35.755725
109
py
rllab
rllab-master/sandbox/rocky/tf/algos/vpg.py
from rllab.misc import logger from rllab.misc import ext from rllab.misc.overrides import overrides from sandbox.rocky.tf.algos.batch_polopt import BatchPolopt from sandbox.rocky.tf.optimizers.first_order_optimizer import FirstOrderOptimizer from sandbox.rocky.tf.misc import tensor_utils from rllab.core.serializable import Serializable import tensorflow as tf class VPG(BatchPolopt, Serializable): """ Vanilla Policy Gradient. """ def __init__( self, env, policy, baseline, optimizer=None, optimizer_args=None, **kwargs): Serializable.quick_init(self, locals()) if optimizer is None: default_args = dict( batch_size=None, max_epochs=1, ) if optimizer_args is None: optimizer_args = default_args else: optimizer_args = dict(default_args, **optimizer_args) optimizer = FirstOrderOptimizer(**optimizer_args) self.optimizer = optimizer self.opt_info = None super(VPG, self).__init__(env=env, policy=policy, baseline=baseline, **kwargs) @overrides def init_opt(self): is_recurrent = int(self.policy.recurrent) obs_var = self.env.observation_space.new_tensor_variable( 'obs', extra_dims=1 + is_recurrent, ) action_var = self.env.action_space.new_tensor_variable( 'action', extra_dims=1 + is_recurrent, ) advantage_var = tensor_utils.new_tensor( name='advantage', ndim=1 + is_recurrent, dtype=tf.float32, ) dist = self.policy.distribution old_dist_info_vars = { k: tf.placeholder(tf.float32, shape=[None] * (1 + is_recurrent) + list(shape), name='old_%s' % k) for k, shape in dist.dist_info_specs } old_dist_info_vars_list = [old_dist_info_vars[k] for k in dist.dist_info_keys] state_info_vars = { k: tf.placeholder(tf.float32, shape=[None] * (1 + is_recurrent) + list(shape), name=k) for k, shape in self.policy.state_info_specs } state_info_vars_list = [state_info_vars[k] for k in self.policy.state_info_keys] if is_recurrent: valid_var = tf.placeholder(tf.float32, shape=[None, None], name="valid") else: valid_var = None dist_info_vars = self.policy.dist_info_sym(obs_var, state_info_vars) logli = dist.log_likelihood_sym(action_var, dist_info_vars) kl = dist.kl_sym(old_dist_info_vars, dist_info_vars) # formulate as a minimization problem # The gradient of the surrogate objective is the policy gradient if is_recurrent: surr_obj = - tf.reduce_sum(logli * advantage_var * valid_var) / tf.reduce_sum(valid_var) mean_kl = tf.reduce_sum(kl * valid_var) / tf.reduce_sum(valid_var) max_kl = tf.reduce_max(kl * valid_var) else: surr_obj = - tf.reduce_mean(logli * advantage_var) mean_kl = tf.reduce_mean(kl) max_kl = tf.reduce_max(kl) input_list = [obs_var, action_var, advantage_var] + state_info_vars_list if is_recurrent: input_list.append(valid_var) self.optimizer.update_opt(loss=surr_obj, target=self.policy, inputs=input_list) f_kl = tensor_utils.compile_function( inputs=input_list + old_dist_info_vars_list, outputs=[mean_kl, max_kl], ) self.opt_info = dict( f_kl=f_kl, ) @overrides def optimize_policy(self, itr, samples_data): logger.log("optimizing policy") inputs = ext.extract( samples_data, "observations", "actions", "advantages" ) agent_infos = samples_data["agent_infos"] state_info_list = [agent_infos[k] for k in self.policy.state_info_keys] inputs += tuple(state_info_list) if self.policy.recurrent: inputs += (samples_data["valids"],) dist_info_list = [agent_infos[k] for k in self.policy.distribution.dist_info_keys] loss_before = self.optimizer.loss(inputs) self.optimizer.optimize(inputs) loss_after = self.optimizer.loss(inputs) logger.record_tabular("LossBefore", loss_before) logger.record_tabular("LossAfter", loss_after) mean_kl, max_kl = self.opt_info['f_kl'](*(list(inputs) + dist_info_list)) logger.record_tabular('MeanKL', mean_kl) logger.record_tabular('MaxKL', max_kl) @overrides def get_itr_snapshot(self, itr, samples_data): return dict( itr=itr, policy=self.policy, baseline=self.baseline, env=self.env, )
4,899
34.766423
109
py
rllab
rllab-master/sandbox/rocky/tf/algos/trpo.py
from sandbox.rocky.tf.algos.npo import NPO from sandbox.rocky.tf.optimizers.conjugate_gradient_optimizer import ConjugateGradientOptimizer class TRPO(NPO): """ Trust Region Policy Optimization """ def __init__( self, optimizer=None, optimizer_args=None, **kwargs): if optimizer is None: if optimizer_args is None: optimizer_args = dict() optimizer = ConjugateGradientOptimizer(**optimizer_args) super(TRPO, self).__init__(optimizer=optimizer, **kwargs)
578
25.318182
95
py
rllab
rllab-master/sandbox/rocky/tf/algos/__init__.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/algos/npg.py
1
0
0
py
rllab
rllab-master/sandbox/rocky/tf/algos/batch_polopt.py
import time from rllab.algos.base import RLAlgorithm import rllab.misc.logger as logger from sandbox.rocky.tf.policies.base import Policy import tensorflow as tf from sandbox.rocky.tf.samplers.batch_sampler import BatchSampler from sandbox.rocky.tf.samplers.vectorized_sampler import VectorizedSampler from rllab.sampler.utils import rollout class BatchPolopt(RLAlgorithm): """ Base class for batch sampling-based policy optimization methods. This includes various policy gradient methods like vpg, npg, ppo, trpo, etc. """ def __init__( self, env, policy, baseline, scope=None, n_itr=500, start_itr=0, batch_size=5000, max_path_length=500, discount=0.99, gae_lambda=1, plot=False, pause_for_plot=False, center_adv=True, positive_adv=False, store_paths=False, whole_paths=True, fixed_horizon=False, sampler_cls=None, sampler_args=None, force_batch_sampler=False, **kwargs ): """ :param env: Environment :param policy: Policy :type policy: Policy :param baseline: Baseline :param scope: Scope for identifying the algorithm. Must be specified if running multiple algorithms simultaneously, each using different environments and policies :param n_itr: Number of iterations. :param start_itr: Starting iteration. :param batch_size: Number of samples per iteration. :param max_path_length: Maximum length of a single rollout. :param discount: Discount. :param gae_lambda: Lambda used for generalized advantage estimation. :param plot: Plot evaluation run after each iteration. :param pause_for_plot: Whether to pause before contiuing when plotting. :param center_adv: Whether to rescale the advantages so that they have mean 0 and standard deviation 1. :param positive_adv: Whether to shift the advantages so that they are always positive. When used in conjunction with center_adv the advantages will be standardized before shifting. :param store_paths: Whether to save all paths data to the snapshot. :return: """ self.env = env self.policy = policy self.baseline = baseline self.scope = scope self.n_itr = n_itr self.start_itr = start_itr self.batch_size = batch_size self.max_path_length = max_path_length self.discount = discount self.gae_lambda = gae_lambda self.plot = plot self.pause_for_plot = pause_for_plot self.center_adv = center_adv self.positive_adv = positive_adv self.store_paths = store_paths self.whole_paths = whole_paths self.fixed_horizon = fixed_horizon if sampler_cls is None: if self.policy.vectorized and not force_batch_sampler: sampler_cls = VectorizedSampler else: sampler_cls = BatchSampler if sampler_args is None: sampler_args = dict() self.sampler = sampler_cls(self, **sampler_args) self.init_opt() def start_worker(self): self.sampler.start_worker() def shutdown_worker(self): self.sampler.shutdown_worker() def obtain_samples(self, itr): return self.sampler.obtain_samples(itr) def process_samples(self, itr, paths): return self.sampler.process_samples(itr, paths) def train(self, sess=None): created_session = True if (sess is None) else False if sess is None: sess = tf.Session() sess.__enter__() sess.run(tf.global_variables_initializer()) self.start_worker() start_time = time.time() for itr in range(self.start_itr, self.n_itr): itr_start_time = time.time() with logger.prefix('itr #%d | ' % itr): logger.log("Obtaining samples...") paths = self.obtain_samples(itr) logger.log("Processing samples...") samples_data = self.process_samples(itr, paths) logger.log("Logging diagnostics...") self.log_diagnostics(paths) logger.log("Optimizing policy...") self.optimize_policy(itr, samples_data) logger.log("Saving snapshot...") params = self.get_itr_snapshot(itr, samples_data) # , **kwargs) if self.store_paths: params["paths"] = samples_data["paths"] logger.save_itr_params(itr, params) logger.log("Saved") logger.record_tabular('Time', time.time() - start_time) logger.record_tabular('ItrTime', time.time() - itr_start_time) logger.dump_tabular(with_prefix=False) if self.plot: rollout(self.env, self.policy, animated=True, max_path_length=self.max_path_length) if self.pause_for_plot: input("Plotting evaluation run: Press Enter to " "continue...") self.shutdown_worker() if created_session: sess.close() def log_diagnostics(self, paths): self.env.log_diagnostics(paths) self.policy.log_diagnostics(paths) self.baseline.log_diagnostics(paths) def init_opt(self): """ Initialize the optimization procedure. If using tensorflow, this may include declaring all the variables and compiling functions """ raise NotImplementedError def get_itr_snapshot(self, itr, samples_data): """ Returns all the data that should be saved in the snapshot for this iteration. """ raise NotImplementedError def optimize_policy(self, itr, samples_data): raise NotImplementedError
6,100
36.89441
111
py
rllab
rllab-master/sandbox/rocky/tf/spaces/box.py
from rllab.spaces.box import Box as TheanoBox import tensorflow as tf class Box(TheanoBox): def new_tensor_variable(self, name, extra_dims, flatten=True): if flatten: return tf.placeholder(tf.float32, shape=[None] * extra_dims + [self.flat_dim], name=name) return tf.placeholder(tf.float32, shape=[None] * extra_dims + list(self.shape), name=name) @property def dtype(self): return tf.float32
444
30.785714
101
py
rllab
rllab-master/sandbox/rocky/tf/spaces/discrete.py
from rllab.spaces.base import Space import numpy as np from rllab.misc import special from rllab.misc import ext import tensorflow as tf class Discrete(Space): """ {0,1,...,n-1} """ def __init__(self, n): self._n = n @property def n(self): return self._n def sample(self): return np.random.randint(self.n) def sample_n(self, n): return np.random.randint(low=0, high=self.n, size=n) def contains(self, x): x = np.asarray(x) return x.shape == () and x.dtype.kind == 'i' and x >= 0 and x < self.n def __repr__(self): return "Discrete(%d)" % self.n def __eq__(self, other): return self.n == other.n def flatten(self, x): return special.to_onehot(x, self.n) def unflatten(self, x): return special.from_onehot(x) def flatten_n(self, x): return special.to_onehot_n(x, self.n) def unflatten_n(self, x): return special.from_onehot_n(x) @property def default_value(self): return 0 @property def flat_dim(self): return self.n def weighted_sample(self, weights): return special.weighted_sample(weights, range(self.n)) def new_tensor_variable(self, name, extra_dims): # needed for safe conversion to float32 return tf.placeholder(dtype=tf.uint8, shape=[None] * extra_dims + [self.flat_dim], name=name) @property def dtype(self): return tf.uint8 def __eq__(self, other): if not isinstance(other, Discrete): return False return self.n == other.n def __hash__(self): return hash(self.n)
1,672
21.306667
101
py
rllab
rllab-master/sandbox/rocky/tf/spaces/__init__.py
from .product import Product from .discrete import Discrete from .box import Box __all__ = ["Product", "Discrete", "Box"]
123
19.666667
40
py
rllab
rllab-master/sandbox/rocky/tf/spaces/product.py
from rllab.spaces.base import Space import tensorflow as tf import numpy as np class Product(Space): def __init__(self, *components): if isinstance(components[0], (list, tuple)): assert len(components) == 1 components = components[0] self._components = tuple(components) dtypes = [c.dtype for c in components] if len(dtypes) > 0 and hasattr(dtypes[0], "as_numpy_dtype"): dtypes = [d.as_numpy_dtype for d in dtypes] self._common_dtype = np.core.numerictypes.find_common_type([], dtypes) def sample(self): return tuple(x.sample() for x in self._components) @property def components(self): return self._components def contains(self, x): return isinstance(x, tuple) and all(c.contains(xi) for c, xi in zip(self._components, x)) def new_tensor_variable(self, name, extra_dims): return tf.placeholder( dtype=self._common_dtype, shape=[None] * extra_dims + [self.flat_dim], name=name, ) @property def dtype(self): return self._common_dtype @property def flat_dim(self): return int(np.sum([c.flat_dim for c in self._components])) def flatten(self, x): return np.concatenate([c.flatten(xi) for c, xi in zip(self._components, x)]) def flatten_n(self, xs): xs_regrouped = [[x[i] for x in xs] for i in range(len(xs[0]))] flat_regrouped = [c.flatten_n(xi) for c, xi in zip(self.components, xs_regrouped)] return np.concatenate(flat_regrouped, axis=-1) def unflatten(self, x): dims = [c.flat_dim for c in self._components] flat_xs = np.split(x, np.cumsum(dims)[:-1]) return tuple(c.unflatten(xi) for c, xi in zip(self._components, flat_xs)) def unflatten_n(self, xs): dims = [c.flat_dim for c in self._components] flat_xs = np.split(xs, np.cumsum(dims)[:-1], axis=-1) unflat_xs = [c.unflatten_n(xi) for c, xi in zip(self.components, flat_xs)] unflat_xs_grouped = list(zip(*unflat_xs)) return unflat_xs_grouped def __eq__(self, other): if not isinstance(other, Product): return False return tuple(self.components) == tuple(other.components) def __hash__(self): return hash(tuple(self.components))
2,360
33.217391
97
py
rllab
rllab-master/sandbox/rocky/tf/misc/tensor_utils.py
import tensorflow as tf import numpy as np def compile_function(inputs, outputs, log_name=None): def run(*input_vals): sess = tf.get_default_session() return sess.run(outputs, feed_dict=dict(list(zip(inputs, input_vals)))) return run def flatten_tensor_variables(ts): return tf.concat(axis=0, values=[tf.reshape(x, [-1]) for x in ts]) def unflatten_tensor_variables(flatarr, shapes, symb_arrs): arrs = [] n = 0 for (shape, symb_arr) in zip(shapes, symb_arrs): size = np.prod(list(shape)) arr = tf.reshape(flatarr[n:n + size], shape) arrs.append(arr) n += size return arrs def new_tensor(name, ndim, dtype): return tf.placeholder(dtype=dtype, shape=[None] * ndim, name=name) def new_tensor_like(name, arr_like): return new_tensor(name, arr_like.get_shape().ndims, arr_like.dtype.base_dtype) def concat_tensor_list(tensor_list): return np.concatenate(tensor_list, axis=0) def concat_tensor_dict_list(tensor_dict_list): keys = list(tensor_dict_list[0].keys()) ret = dict() for k in keys: example = tensor_dict_list[0][k] if isinstance(example, dict): v = concat_tensor_dict_list([x[k] for x in tensor_dict_list]) else: v = concat_tensor_list([x[k] for x in tensor_dict_list]) ret[k] = v return ret def stack_tensor_list(tensor_list): return np.array(tensor_list) # tensor_shape = np.array(tensor_list[0]).shape # if tensor_shape is tuple(): # return np.array(tensor_list) # return np.vstack(tensor_list) def stack_tensor_dict_list(tensor_dict_list): """ Stack a list of dictionaries of {tensors or dictionary of tensors}. :param tensor_dict_list: a list of dictionaries of {tensors or dictionary of tensors}. :return: a dictionary of {stacked tensors or dictionary of stacked tensors} """ keys = list(tensor_dict_list[0].keys()) ret = dict() for k in keys: example = tensor_dict_list[0][k] if isinstance(example, dict): v = stack_tensor_dict_list([x[k] for x in tensor_dict_list]) else: v = stack_tensor_list([x[k] for x in tensor_dict_list]) ret[k] = v return ret def split_tensor_dict_list(tensor_dict): keys = list(tensor_dict.keys()) ret = None for k in keys: vals = tensor_dict[k] if isinstance(vals, dict): vals = split_tensor_dict_list(vals) if ret is None: ret = [{k: v} for v in vals] else: for v, cur_dict in zip(vals, ret): cur_dict[k] = v return ret def to_onehot_sym(inds, dim): return tf.one_hot(inds, depth=dim, on_value=1, off_value=0) def pad_tensor(x, max_len): return np.concatenate([ x, np.tile(np.zeros_like(x[0]), (max_len - len(x),) + (1,) * np.ndim(x[0])) ]) def pad_tensor_n(xs, max_len): ret = np.zeros((len(xs), max_len) + xs[0].shape[1:], dtype=xs[0].dtype) for idx, x in enumerate(xs): ret[idx][:len(x)] = x return ret def pad_tensor_dict(tensor_dict, max_len): keys = list(tensor_dict.keys()) ret = dict() for k in keys: if isinstance(tensor_dict[k], dict): ret[k] = pad_tensor_dict(tensor_dict[k], max_len) else: ret[k] = pad_tensor(tensor_dict[k], max_len) return ret
3,406
27.157025
90
py