File size: 14,992 Bytes
714cf46 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 | """
Embedding Test Suite CLI
Tests embedding quality by sampling sequences from EC dataset (default),
embedding them with various pooling methods, and reporting statistics
on distribution, NaNs, and sparsity.
"""
import json
import argparse
import math
import random
import numpy as np
import torch
from typing import Dict, List, Optional
try:
from data.data_mixin import DataMixin, DataArguments
from embedder import Embedder, EmbeddingArguments
from base_models.get_base_models import standard_models
from seed_utils import set_global_seed, get_global_seed
from utils import print_message
except ImportError:
from ..data.data_mixin import DataMixin, DataArguments
from ..embedder import Embedder, EmbeddingArguments
from ..base_models.get_base_models import standard_models
from ..seed_utils import set_global_seed, get_global_seed
from ..utils import print_message
# Default test datasets
DEFAULT_TEST_DATASETS = [
'EC', # multilabel
]
seed = get_global_seed()
if seed is not None:
random.seed(seed)
np.random.seed(seed)
def load_and_sample_sequences(
dataset_names: List[str],
sample_frac: float = 0.1,
max_length: int = 1024,
trim: bool = False
) -> Dict[str, List[str]]:
"""
Load datasets and sample sequences from them.
Args:
dataset_names: List of dataset names to load
sample_frac: Fraction of sequences to sample (default 0.1 = 10%)
max_length: Maximum sequence length
trim: Whether to trim sequences to max_length
Returns:
Dictionary mapping dataset names to lists of sampled sequences
"""
dataset_seqs = {}
for dataset_name in dataset_names:
print_message(f"Loading dataset: {dataset_name}")
try:
# Load dataset using DataMixin
data_args = DataArguments(
data_names=[dataset_name],
max_length=max_length,
trim=trim
)
data_mixin = DataMixin(data_args)
datasets, all_seqs = data_mixin.get_data()
# Get sequences from all splits
sequences = []
if dataset_name in datasets:
train_set, valid_set, test_set, _, _, ppi = datasets[dataset_name]
if ppi:
# For PPI datasets, combine SeqA and SeqB
sequences.extend(list(train_set['SeqA']))
sequences.extend(list(train_set['SeqB']))
sequences.extend(list(valid_set['SeqA']))
sequences.extend(list(valid_set['SeqB']))
sequences.extend(list(test_set['SeqA']))
sequences.extend(list(test_set['SeqB']))
else:
sequences.extend(list(train_set['seqs']))
sequences.extend(list(valid_set['seqs']))
sequences.extend(list(test_set['seqs']))
else:
# Use all sequences if dataset processing failed
sequences = list(all_seqs)
# Sample
sequences = list(set(sequences))
n_samples = max(1, math.ceil(len(sequences) * sample_frac))
sampled = random.sample(sequences, min(n_samples, len(sequences)))
dataset_seqs[dataset_name] = sampled
print_message(f"Sampled {len(sampled)} sequences from {len(sequences)} total")
except Exception as e:
print_message(f"Error loading dataset {dataset_name}: {e}")
continue
return dataset_seqs
def compute_diagnostics(embeddings: torch.Tensor, zero_eps: float = 1e-8) -> Dict[str, float]:
emb = embeddings.detach().float().cpu().numpy()
flat = emb.ravel()
is_nan = np.isnan(flat)
is_inf = np.isinf(flat)
is_finite = np.isfinite(flat)
finite = flat[is_finite]
if finite.size == 0:
# If everything is NaN/Inf
return {
"n_samples": int(emb.shape[0]),
"embedding_dim": int(emb.shape[1]),
"finite_count": 0,
"nan_count": int(is_nan.sum()),
"inf_count": int(is_inf.sum()),
}
near_zero = np.abs(finite) < zero_eps
sample_l2 = np.linalg.norm(emb, axis=1)
return {
"n_samples": int(emb.shape[0]),
"embedding_dim": int(emb.shape[1]),
"finite_count": int(finite.size),
"finite_fraction": float(finite.size / flat.size),
"nan_count": int(is_nan.sum()),
"nan_fraction": float(is_nan.mean()),
"inf_count": int(is_inf.sum()),
"inf_fraction": float(is_inf.mean()),
"zero_eps": float(zero_eps),
"near_zero_count": int(near_zero.sum()),
"near_zero_fraction": float(near_zero.mean()),
"mean": float(np.mean(finite)),
"std": float(np.std(finite)),
"min": float(np.min(finite)),
"max": float(np.max(finite)),
"p25": float(np.percentile(finite, 25)),
"p50": float(np.percentile(finite, 50)),
"p75": float(np.percentile(finite, 75)),
"p95": float(np.percentile(finite, 95)),
"p99": float(np.percentile(finite, 99)),
"mean_l2": float(np.mean(sample_l2)),
"std_l2": float(np.std(sample_l2)),
"p95_l2": float(np.percentile(sample_l2, 95)),
}
def embed_and_diagnose(
sequences: List[str],
model_name: str,
pooling_types: List[str],
batch_size: int = 16,
num_workers: int = 0
) -> Dict[str, Dict[str, float]]:
"""
Embed sequences and compute diagnostics for each pooling type.
Args:
sequences: List of sequences to embed
model_name: Name of the model to use
pooling_types: List of pooling types to test
batch_size: Batch size for embedding
num_workers: Number of workers for data loading
Returns:
Dictionary mapping pooling types to their diagnostics
"""
print_message(f"Embedding {len(sequences)} sequences with {model_name}")
# Parse pooling types so combinations can be tested
pooling_list = {}
for pool_type in pooling_types:
# Check if it's a combination
if ',' in pool_type:
# Split and create a list for the combination
pool_list = [p.strip() for p in pool_type.split(',')]
pooling_list[pool_type] = pool_list
else:
# Single pooling type
pooling_list[pool_type] = [pool_type]
results = {}
# Load model once and reuse for all pooling types
print_message(f"Loading model: {model_name}")
from base_models.get_base_models import get_base_model
model, tokenizer = get_base_model(model_name)
for pool_type, pool_list in pooling_list.items():
print_message(f"Testing pooling: {pool_type} (types: {pool_list})")
# Set up embedder for this pooling type
embedder_args = EmbeddingArguments(
embedding_batch_size=batch_size,
embedding_num_workers=num_workers,
download_embeddings=False,
matrix_embed=False,
embedding_pooling_types=pool_list,
save_embeddings=False,
embed_dtype=torch.float32,
sql=False,
embedding_save_dir='embeddings'
)
embedder = Embedder(embedder_args, sequences)
try:
# read embeddings from disk if they exist
to_embed, save_path, embeddings_dict = embedder._read_embeddings_from_disk(model_name)
if len(to_embed) > 0:
result = embedder._embed_sequences(
to_embed, save_path, model, tokenizer, embeddings_dict
)
if result is not None:
embeddings_dict = result
if embeddings_dict is None or len(embeddings_dict) == 0:
print_message(f"Warning: No embeddings returned for {model_name} with {pool_type}")
continue
embedding_tensors = []
for seq in sequences:
if seq in embeddings_dict:
embedding_tensors.append(embeddings_dict[seq])
if len(embedding_tensors) == 0:
print_message(f"Error: No embeddings found for {pool_type}")
continue
embeddings = torch.stack(embedding_tensors)
diagnostics = compute_diagnostics(embeddings)
results[pool_type] = diagnostics
except Exception as e:
print_message(f"Error embedding with {model_name} using {pool_type}: {e}")
import traceback
traceback.print_exc()
continue
return results
def run_test_suite(
dataset_names: Optional[List[str]] = None,
model_names: Optional[List[str]] = None,
pooling_methods: List[str] = ['cls', 'mean,var'],
sample_frac: float = 0.1,
batch_size: int = 16,
num_workers: int = 0
) -> Dict:
"""
Run the embedding test suite.
"""
if dataset_names is None:
dataset_names = DEFAULT_TEST_DATASETS
if model_names is None:
model_names = standard_models
print_message(f"Running embedding test suite")
print_message(f"Datasets: {dataset_names}")
print_message(f"Models: {model_names}")
print_message(f"Pooling methods: {pooling_methods}")
print_message(f"Sample fraction: {sample_frac}")
dataset_seqs = load_and_sample_sequences(dataset_names, sample_frac=sample_frac)
if len(dataset_seqs) == 0:
print_message("Error: No sequences loaded")
return {}
all_results = {}
for dataset_name, sequences in dataset_seqs.items():
print_message(f"\nProcessing dataset: {dataset_name}")
all_results[dataset_name] = {}
for model_name in model_names:
print_message(f"Model: {model_name}")
model_results = embed_and_diagnose(
sequences,
model_name,
pooling_methods,
batch_size=batch_size,
num_workers=num_workers
)
if model_results:
all_results[dataset_name][model_name] = model_results
print_table_results(all_results)
print_json_results(all_results)
return all_results
def print_table_results(results: Dict):
"""Print results in table format."""
print("\n" + "="*100)
print("EMBEDDING TEST SUITE RESULTS")
print("="*100)
for dataset_name, dataset_results in results.items():
print(f"\nDataset: {dataset_name}")
print("-" * 100)
for model_name, model_results in dataset_results.items():
print(f"\n Model: {model_name}")
for pool_type, diagnostics in model_results.items():
print(f"\nPooling: {pool_type}")
print(f"Samples: {diagnostics['n_samples']}, Dim: {diagnostics['embedding_dim']}")
print(f"Mean: {diagnostics['mean']:.6f}, Std: {diagnostics['std']:.6f}")
print(f"Min: {diagnostics['min']:.6f}, Max: {diagnostics['max']:.6f}")
print(f"Percentiles: P25={diagnostics['p25']:.6f}, P50={diagnostics['p50']:.6f}, "
f"P75={diagnostics['p75']:.6f}, P95={diagnostics['p95']:.6f}, P99={diagnostics['p99']:.6f}")
print(f"NaN: {diagnostics['nan_count']} ({diagnostics['nan_fraction']*100:.2f}%)")
if 'near_zero_count' in diagnostics:
print(f"Near zeros: {diagnostics['near_zero_count']} ({diagnostics['near_zero_fraction']*100:.2f}%)")
print(f"Inf: {diagnostics['inf_count']} ({diagnostics['inf_fraction']*100:.2f}%)")
# Flag anomalies
anomalies = []
if diagnostics['nan_fraction'] > 0:
anomalies.append(f"NaNs detected ({diagnostics['nan_fraction']*100:.2f}%)")
if 'near_zero_fraction' in diagnostics and diagnostics['near_zero_fraction'] > 0.2:
anomalies.append(f"High sparsity ({diagnostics['near_zero_fraction']*100:.2f}%)")
if diagnostics['inf_fraction'] > 0:
anomalies.append(f"Infs detected ({diagnostics['inf_fraction']*100:.2f}%)")
if abs(diagnostics['mean']) > 100:
anomalies.append(f"Extreme mean ({diagnostics['mean']:.2f})")
if diagnostics['std'] > 100:
anomalies.append(f"Extreme std ({diagnostics['std']:.2f})")
if anomalies:
print(f"Anomalies: {', '.join(anomalies)}")
else:
print(f"No anomalies detected")
def print_json_results(results: Dict):
"""Print results in JSON format."""
print("\n" + "="*50)
print("JSON RESULTS")
print("="*50)
print(json.dumps(results, indent=2))
def main():
parser = argparse.ArgumentParser(
description='Embedding Test Suite - Test embedding quality across datasets and models'
)
parser.add_argument(
'--datasets',
nargs='+',
default=None,
help=f'List of dataset names to test (default: EC)'
)
parser.add_argument(
'--model_names',
nargs='+',
default=None,
help='List of model names to test (default: all currently_supported_models)'
)
parser.add_argument(
'--pooling_methods',
nargs='+',
default=['cls', 'mean,var'],
help='List of pooling methods to test (default: mean, var, cls, parti, mean,var)'
)
parser.add_argument(
'--sample_frac',
type=float,
default=0.1,
help='Fraction of sequences to sample from each dataset (default: 0.1)'
)
parser.add_argument(
'--batch_size',
type=int,
default=16,
help='Batch size for embedding (default: 16)'
)
parser.add_argument(
'--num_workers',
type=int,
default=0,
help='Number of workers for data loading (default: 0)'
)
parser.add_argument(
'--seed',
type=int,
default=None,
help='Random seed for reproducibility'
)
args = parser.parse_args()
# Set seed if provided
if args.seed is not None:
set_global_seed(args.seed)
# Run test suite
results = run_test_suite(
dataset_names=args.datasets,
model_names=args.model_names,
pooling_methods=args.pooling_methods,
sample_frac=args.sample_frac,
batch_size=args.batch_size,
num_workers=args.num_workers
)
return results
if __name__ == '__main__':
main()
|