File size: 17,333 Bytes
196bee3 | 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 | # summarize_results.py
import re
import os
import glob
import pickle
import numpy as np
import pandas as pd
from scipy.special import expit # Sigmoid function
from sklearn.metrics import roc_auc_score, accuracy_score, f1_score
PREFIX = 'B1'
LOG_DIR = './server_logs/'
RESULTS_DIR = './results/'
# Define all the experiments
MODELS_ORDER = ['LDA', 'LR', 'MLP', 'EEGNet', 'EEGPT']
STRATEGIES = ['single_subject', 'cross_subject', 'subject_adapted']
EXPERIMENTS = ['relevance']#, 'task', 'joint']
TASKS_ORDER = [
'facecat/female', 'facecat/male', 'facecat/blond', 'facecat/darkhaired',
'facecat/smiles', 'facecat/nosmile', 'facecat/old', 'facecat/young'
]
def parse_summary_line(line):
"""
Uses regex to pull mean/std from a SUMMARY line.
"""
metrics = {}
try:
metrics['acc_mean'] = float(re.search(r"acc:([\d\.]+) \+\/-", line).group(1))
metrics['acc_std'] = float(re.search(r"acc:[\d\.]+ \+\/- ([\d\.]+)", line).group(1))
metrics['auc_mean'] = float(re.search(r"auc:([\d\.]+) \+\/-", line).group(1))
metrics['auc_std'] = float(re.search(r"auc:[\d\.]+ \+\/- ([\d\.]+)", line).group(1))
metrics['f1_mean'] = float(re.search(r"f1:([\d\.]+) \+\/-", line).group(1))
metrics['f1_std'] = float(re.search(r"f1:[\d\.]+ \+\/- ([\d\.]+)", line).group(1))
metrics['task'] = int(re.search(r"task_id:(\d+),", line).group(1))
except Exception:
return None # Silently fail
return metrics
def parse_result_line(line):
"""
Uses regex to pull mean/std from a RESULT line.
"""
metrics = {}
try:
metrics['acc_mean'] = float(re.search(r"acc:([\d\.]+)", line).group(1))
metrics['auc_mean'] = float(re.search(r"auc:([\d\.]+)", line).group(1))
metrics['f1_mean'] = float(re.search(r"f1:([\d\.]+)", line).group(1))
metrics['task'] = int(re.search(r"task_id:(\d+),", line).group(1))
metrics['subject'] = int(re.search(r"subject:(\d+),", line).group(1))
except Exception:
return None # Silently fail
return metrics
prefix='B1'
exp='relevance'
strat='single_subject'
model='LDA'
log_file = f"log_{prefix}_{exp}_{strat}_{model}.log"
log_path = os.path.join(LOG_DIR, log_file)
def load_all_summaries(prefix=PREFIX, experiments=EXPERIMENTS, strategies=STRATEGIES, models=MODELS_ORDER):
"""
Parses all log files and returns a master DataFrame.
"""
print(f"Loading all summaries for prefix='{prefix}'...")
results_list = []
summary_list = []
for prefix in ['A1', 'B1']:
for exp in experiments:
for strat in strategies:
for model in models:
# 'relevance' experiment is special, it has 8 task files
if exp == 'relevance':
log_file = f"log_{prefix}_{exp}_{strat}_{model}.log"
log_path = os.path.join(LOG_DIR, log_file)
summary = None
if os.path.exists(log_path):
with open(log_path, 'r') as f:
for line in reversed(f.readlines()):
if "SUMMARY" in line:
summary = parse_summary_line(line)
if summary is None:
continue
summary.update({
'experiment': exp,
'strategy': strat,
'model': model,
})
summary_list.append(summary)
# RESULT, exp:relevance, strat:subject_adapted, task_id:5, subject:27, model:EEGNet, acc:0.8491, auc:0.8953, f1:0.7870
if "RESULT" in line and "fold:" not in line:
result = parse_result_line(line)
if result is None:
continue
result.update({
'experiment': exp,
'strategy': strat,
'model': model,
})
results_list.append(result)
else:
# 'task' and 'joint' experiments
log_file = f"log_{prefix}_{exp}_{strat}_{model}.log"
log_path = os.path.join(LOG_DIR, log_file)
summary = None
if os.path.exists(log_path):
with open(log_path, 'r') as f:
for line in reversed(f.readlines()):
if line.startswith("SUMMARY"):
summary = parse_summary_line(line)
break
if summary:
summary.update({
'experiment': exp,
'strategy': strat,
'model': model,
'task': 'all_tasks_combined' # Special task name
})
summary_list.append(summary)
print(f"Found {len(summary_list)} summary lines, {len(results_list)} result lines.")
return pd.DataFrame(summary_list), pd.DataFrame(results_list)
def format_table_rows(
df, results_df,
experiment,
strategy,
metric='auc',
precision=4,
models_order=MODELS_ORDER,
tasks_order=TASKS_ORDER
):
"""
Generates and prints only the inner LaTeX rows for a table.
It's robust to missing data (e.g., LDA for subject_adapted).
"""
print("\n" + "="*80)
print(f"Generating LaTeX Rows: Exp='{experiment}', Strat='{strategy}', Metric='{metric.upper()}'")
print("="*80 + "\n")
df_filt = df[(df['experiment'] == experiment) & (df['strategy'] == strategy)]
if df_filt.empty:
print(f"% No data found for: {experiment} / {strategy}")
return
metric_mean = f'{metric}_mean'
metric_std = f'{metric}_std'
if metric_mean not in df_filt.columns:
print(f"% No data for metric '{metric}' in {experiment} / {strategy}")
return
pivot = pd.pivot_table(
df_filt,
values=[metric_mean, metric_std],
index='task',
columns='model'
)
available_models = pivot[metric_mean].columns.tolist()
models_to_print = [m for m in models_order if m in available_models]
if 'EEGPT' not in available_models:
models_to_print += ['EEGPT']
tasks_to_print = tasks_order
task_means = pivot[metric_mean] # Shape: (n_tasks, n_models)
all_task_mean = task_means.mean(axis=0) # Shape: (n_models,)
all_task_std = task_means.std(axis=0)
latex_str = []
# Header
col_str = " & ".join(models_to_print)
latex_str.append(f"% Table for {experiment} / {strategy} / {metric}")
latex_str.append(f"% Columns: Task & {col_str} \\\\")
latex_str.append("\\midrule")
tmp = {}
# Body
for tid, task_name in enumerate(tasks_to_print):
# Shorten the task name, e.g., 'facecat/blond' -> 'Blond'
short_name = task_name.split('/')[-1].title()
row = f"{short_name:15s} & "
cells = []
for model in models_to_print:
if (metric_mean, model) not in pivot.columns:
# try to recover from results_df
results_filt = results_df[
(results_df['experiment'] == experiment) &
(results_df['strategy'] == strategy) &
(results_df['model'] == model) &
(results_df['task'] == tid)
]
# average all subjects for this task
mean_val = results_filt[metric_mean].mean()
std_val = results_filt[metric_mean].std()
else:
mean_val = pivot.loc[tid][(metric_mean, model)]
std_val = pivot.loc[tid][(metric_std, model)]
tmp[(tid, model)] = (mean_val, std_val)
cell_str = f"${mean_val:.{precision}f} \\pm {std_val:.{precision}f}$"
cells.append(cell_str)
row += " & ".join(cells) + " \\\\"
latex_str.append(row)
# Footer (All-Task Mean)
latex_str.append("\\midrule")
row = f"{'Mean':15s} & "
cells = []
for model in models_to_print:
mean_val = all_task_mean.loc[model]
std_val = all_task_std.loc[model] # std of the 8 task means
tmp_val = np.array([tmp[(tid, model)][0] for tid in range(len(tasks_to_print))])
tmp_mean = np.mean(tmp_val)
tmp_std = np.std(tmp_val)
print(f"{model}: {tmp_mean:.4f} +/- {tmp_std:.4f}, {mean_val:.4f} +/- {std_val:.4f}, {tmp_val.shape}")
mean_val, std_val = tmp_mean, tmp_std
cell_str = f"${mean_val:.{precision}f} \\pm {std_val:.{precision}f}$"
cells.append(cell_str)
row += " & ".join(cells) + " \\\\"
latex_str.append(row)
print("\n".join(latex_str))
def run_voting_analysis(
prefix=PREFIX,
model_name='LDA',
experiment_name='relevance',
strategy_name='cross_subject',
tasks=TASKS_ORDER,
results_dir=RESULTS_DIR
):
"""
Loads all pickle files for a model and performs multi-subject voting.
This is designed for 'cross_subject' data, where predictions
for each subject are independent.
"""
print("\n" + "="*80)
print(f"Running Voting Analysis: Model='{model_name}', Strat='{strategy_name}'")
print("="*80)
all_predictions = []
# check all files inside "results_all/archive/"
archive_dir = "results_all/archive/"
for pkl_path in glob.glob(os.path.join(archive_dir, "*.pkl")):
# try parse filename
# example: results_A1_relevance_cross_subject_EEGPT_task0.pkl
s = pkl_path.split('/')[-1].split('.')[0].split('_')
if len(s) < 8:
continue
for experiment in EXPERIMENTS:
if experiment in pkl_path:
break
for strategy in STRATEGIES:
if strategy in pkl_path:
break
for model in MODELS_ORDER:
if model in pkl_path:
break
for task_id, task_name in enumerate(tasks):
if f"task{task_id}" in pkl_path:
break
if model_name != model or experiment_name != experiment or strategy_name != strategy:
continue
results = pickle.load(open(pkl_path, 'rb'))
# results.values() = (y_prob, test_mask, ids_test, (acc, auc, f1))
for key, (y_prob, _, ids_test, _) in results.items():
# key is test_subject_id
for i in range(len(y_prob)):
prob_class_1 = y_prob[i, 1]
true_label = ids_test[i, 3]
image_idx = ids_test[i, 4]
all_predictions.append({
'task_id': task_id,
'image_idx': image_idx,
'true_label': true_label,
'y_prob_1': prob_class_1,
'test_subject': key
})
if not all_predictions:
print("No predictions found. Aborting voting analysis.")
return
df = pd.DataFrame(all_predictions)
print(f"Loaded {len(df)} single-trial predictions from {model}...")
# Group by unique image and average probabilities
voted_df = df.groupby(['task_id', 'image_idx']).agg(
voted_prob=('y_prob_1', 'mean'), # Average the probabilities
true_label=('true_label', 'first'),
n_votes=('true_label', 'count')
)
print(f"Aggregated into {len(voted_df)} unique image-task pairs.")
# Calculate final voted metrics
y_true = voted_df['true_label']
y_prob = voted_df['voted_prob']
y_pred = (y_prob > 0.5).astype(int)
voted_auc = roc_auc_score(y_true, y_prob)
voted_acc = accuracy_score(y_true, y_pred)
voted_f1 = f1_score(y_true, y_pred, average='macro')
print("\n--- Voting Results ---")
print(f"Model: {model}")
print(f"Strategy: {strategy} (Multi-Subject Voted)")
print(f"Voted AUC: {voted_auc:.4f}")
print(f"Voted Acc: {voted_acc:.4f}")
print(f"Voted F1: {voted_f1:.4f}")
def bootstrap_metric(y_true, y_prob, metric, n_bootstraps=1000, random_state=42):
results = []
y_prob = np.array(y_prob)
y_true = np.array(y_true)
for i in range(n_bootstraps):
idx = np.random.choice(len(y_true), size=len(y_true), replace=True)
y_true_boot = y_true[idx]
y_prob_boot = y_prob[idx]
if metric == 'auc':
score = roc_auc_score(y_true_boot, y_prob_boot)
elif metric == 'acc':
y_pred_boot = (y_prob_boot > 0.5).astype(int)
score = accuracy_score(y_true_boot, y_pred_boot)
elif metric == 'f1':
y_pred_boot = (y_prob_boot > 0.5).astype(int)
score = f1_score(y_true_boot, y_pred_boot, average='macro')
results.append(score)
return np.mean(results), np.std(results)
# print by task
task_metrics = {}
for task_id, task_name in enumerate(tasks):
task_df = voted_df[voted_df.index.get_level_values('task_id') == task_id]
y_true = task_df['true_label']
y_prob = task_df['voted_prob']
y_pred = (y_prob > 0.5).astype(int)
# task_auc = roc_auc_score(y_true, y_prob)
# task_acc = accuracy_score(y_true, y_pred)
# task_f1 = f1_score(y_true, y_pred, average='macro')
task_auc, task_auc_std = bootstrap_metric(y_true, y_prob, 'auc')
task_acc, task_acc_std = bootstrap_metric(y_true, y_prob, 'acc')
task_f1, task_f1_std = bootstrap_metric(y_true, y_prob, 'f1')
print(f"\nTask {task_id}: {task_name:<20s}, AUC: {task_auc:.4f} +/- {task_auc_std:.4f}, Acc: {task_acc:.4f} +/- {task_acc_std:.4f}, F1: {task_f1:.4f} +/- {task_f1_std:.4f}")
task_metrics[task_id] = {'auc': task_auc, 'acc': task_acc, 'f1': task_f1, 'task': task_id, 'auc_std': task_auc_std, 'acc_std': task_acc_std, 'f1_std': task_f1_std}
return voted_df, {'auc': voted_auc, 'acc': voted_acc, 'f1': voted_f1}, task_metrics
# --- 5. Main Execution ---
if __name__ == '__main__':
master_df, results_df = load_all_summaries()
print("\n--- Master DataFrame Head ---")
print(master_df.head())
# --- Single Subject ---
format_table_rows(
master_df, results_df,
experiment='relevance',
strategy='single_subject',
metric='auc'
)
# --- Cross Subject ---
format_table_rows(
master_df, results_df,
experiment='relevance',
strategy='cross_subject',
metric='auc'
)
# --- Subject Adapted ---
format_table_rows(
master_df, results_df,
experiment='relevance',
strategy='subject_adapted',
metric='auc'
)
# Run the voting analysis
print("\n--- Running Voting Analysis ---")
all_voted_results = []
all_task_metrics = {}
for model in MODELS_ORDER:
if model in ['LDA', 'LR', 'MLP', 'EEGNet', 'EEGPT']:
voted_df, voted_metrics, task_metrics = run_voting_analysis(
model_name=model,
strategy_name='cross_subject'
)
voted_metrics['model'] = model
all_voted_results.append(voted_metrics)
all_task_metrics[model] = task_metrics
print("\n--- Final Voting Summary Table ---")
voted_summary_df = pd.DataFrame(all_voted_results).set_index('model')
print(voted_summary_df)
print("\n--- Final Voting Task Table AUC latex table ---")
for task_id, task_name in enumerate(TASKS_ORDER):
task_df = pd.DataFrame(all_task_metrics).T
line = f"{task_name.split('/')[-1].title():<20s} "
for model in MODELS_ORDER:
if model in task_df.index:
mean_val = task_df.loc[model, task_id]['auc']
std_val = task_df.loc[model, task_id]['auc_std']
line += f"& {mean_val:.4f} $\\pm$ {std_val:.4f}"
else:
line += "& "
line += " \\\\"
print(line)
print("\\midrule")
line = f"{'Mean':<20s} "
for model in MODELS_ORDER:
if model in task_df.index:
tmp_val = [task_df.loc[model, task_id]['auc'] for task_id in range(len(TASKS_ORDER))]
mean_val = np.mean(tmp_val)
std_val = np.std(tmp_val)
tmp_val = voted_summary_df.loc[model, 'auc']
# print(f"{model}: {tmp_val:.4f} +/- {mean_val:.4f}")
line += f"& {mean_val:.4f} $\\pm$ {std_val:.4f}"
else:
line += "& "
line += " \\\\"
print(line) |