Spaces:
Running
Running
File size: 17,684 Bytes
4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac 9794ac3 4829aac | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 | #!/usr/bin/env python3
"""
Batch re-evaluate all submissions with the new Semantic Accuracy metric.
This script downloads all prediction files from HuggingFace Hub and re-evaluates
them with the ANLS* + LLM judge metric.
Usage:
# Dry run - list files only
python batch_reevaluate.py --dry-run
# Re-evaluate all files
python batch_reevaluate.py
# Re-evaluate specific organization
python batch_reevaluate.py --org OpenAI
# Upload results after review
python batch_reevaluate.py --upload
"""
import json
import os
import sys
import time
from collections import defaultdict
from concurrent.futures import ThreadPoolExecutor, as_completed
from datetime import datetime, timezone
from pathlib import Path
from huggingface_hub import HfApi, hf_hub_download, list_repo_files
from datasets import load_dataset
# Add parent for imports
sys.path.insert(0, str(Path(__file__).parent))
from metrics import (
anls_star,
anls_star_llm,
aggregate_anls_star_llm,
citation_f1,
kuiper_statistic
)
# Parallelization config
MAX_WORKERS = 24
# Config
RESULTS_REPO = "agentic-document-ai/backend-results"
TOKEN = os.environ.get("HF_TOKEN")
OUTPUT_DIR = Path(__file__).parent / "reevaluated_results"
def load_gold_data():
"""Load gold standard from HuggingFace."""
print("Loading gold standard...")
dataset = load_dataset("agentic-document-ai/dataset-PRIVATE", split="test")
gold_by_id = {}
gold_by_text = {}
def _derive_hop_type(evidence: list) -> str:
if not evidence:
return 'single'
documents = set()
pages = set()
for ev in evidence:
doc = ev.get('document')
page = ev.get('page')
if doc is not None:
documents.add(doc)
if page is not None:
pages.add(page)
if len(documents) > 1:
return 'cross_doc'
if len(pages) > 1:
return 'cross_page'
return 'single'
for ex in dataset:
qid = ex.get('id', '')
question = ex['question'].strip()
evidence = ex.get('evidence', [])
data = {
'question': question,
'answers': ex.get('answer_variants', []),
'evidence': evidence,
'category': ex.get('document_category', ''),
'domain': ex.get('domain', ''),
'hop_type': _derive_hop_type(evidence),
}
gold_by_id[qid] = data
gold_by_text[question] = data
return gold_by_id, gold_by_text
def find_prediction_files(org_filter: str = None):
"""Find all prediction JSONL files in the results repo."""
files = list_repo_files(RESULTS_REPO, repo_type="dataset", token=TOKEN)
pred_files = [f for f in files if '_predictions' in f and f.endswith('.jsonl')]
if org_filter:
pred_files = [f for f in pred_files if f.startswith(org_filter + '/')]
return pred_files
def find_result_file(pred_file: str):
"""Find the corresponding results JSON file for a predictions file."""
# Pattern: {org}/{model}_predictions_{timestamp}.jsonl -> {org}/{model}_results_{timestamp}.json
parts = pred_file.rsplit('_predictions_', 1)
if len(parts) == 2:
result_file = parts[0] + '_results_' + parts[1].replace('.jsonl', '.json')
return result_file
return None
def download_file(filepath: str) -> str:
"""Download a file from HuggingFace Hub."""
return hf_hub_download(
repo_id=RESULTS_REPO,
filename=filepath,
repo_type="dataset",
token=TOKEN
)
def _evaluate_single_prediction(args, max_retries=3):
"""Evaluate a single prediction (for parallel processing)."""
idx, pred, gold_data = args
answer = pred.get('answer', '')
question = pred.get('question', '').strip()
citations = pred.get('citations', [])
search_history = pred.get('search_history', [])
steps = len(search_history) if search_history else pred.get('iterations', 0)
# Calculate non-LLM metrics first
anls = anls_star(answer, gold_data['answers'])
doc_f1 = citation_f1(citations, gold_data['evidence'], level='document')
page_f1 = citation_f1(citations, gold_data['evidence'], level='page')
# Retry LLM call on failure
for attempt in range(max_retries):
try:
llm_result = anls_star_llm(answer, gold_data['answers'], question)
semantic_score = llm_result['score']
break
except Exception as e:
if attempt < max_retries - 1:
print(f" Item {idx} attempt {attempt+1} failed: {e}, retrying...")
time.sleep(2 ** attempt) # Exponential backoff
else:
print(f" Failed item {idx} after {max_retries} retries: {e}")
raise
return {
'idx': idx,
'anls': anls,
'semantic_score': semantic_score,
'correct': semantic_score >= 0.5,
'doc_f1': doc_f1['f1'],
'page_f1': page_f1['f1'],
'steps': steps,
'hop_type': gold_data.get('hop_type', 'single'),
'category': gold_data['category'],
'domain': gold_data['domain']
}
def evaluate_with_semantic(predictions: list, gold_by_id: dict, gold_by_text: dict) -> dict:
"""Evaluate predictions with semantic accuracy metric (parallelized)."""
# First, filter predictions to only those in test set
matched_predictions = []
for pred in predictions:
question = pred.get('question', '').strip()
qid = pred.get('id', '')
gold_data = None
if question in gold_by_text:
gold_data = gold_by_text[question]
elif qid and qid in gold_by_id:
gold_data = gold_by_id[qid]
if gold_data:
matched_predictions.append((pred, gold_data))
unmatched = len(predictions) - len(matched_predictions)
print(f" Matched {len(matched_predictions)}/{len(predictions)} predictions to test set (skipping {unmatched})")
total = len(matched_predictions)
evals = []
completed = 0
# Prepare items with index for tracking
items_with_idx = [(i, pred, gold) for i, (pred, gold) in enumerate(matched_predictions)]
# Parallel evaluation with ThreadPoolExecutor
print(f" Evaluating with {MAX_WORKERS} parallel workers...")
with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
futures = {executor.submit(_evaluate_single_prediction, item): item[0]
for item in items_with_idx}
completed_indices = set()
try:
for future in as_completed(futures, timeout=600): # 10 min overall timeout
try:
result = future.result(timeout=120) # 2 min per item max
evals.append(result)
completed_indices.add(result['idx'])
completed += 1
if completed % 50 == 0 or completed == total:
print(f" Progress: {completed}/{total}")
except TimeoutError:
idx = futures[future]
print(f" TIMEOUT: Item {idx} took too long, skipping")
completed += 1
except TimeoutError:
# Find which items are still pending
pending = set(range(total)) - completed_indices
print(f" OVERALL TIMEOUT: {len(pending)} items still pending: {sorted(pending)[:10]}...")
# Cancel remaining futures
for future in futures:
future.cancel()
if not evals:
return None
# Aggregate
n = len(evals)
semantic_scores = [e['semantic_score'] for e in evals]
agg = aggregate_anls_star_llm(semantic_scores, apply_bias_correction=True)
mean_anls = sum(e['anls'] for e in evals) / n * 100
mean_doc_f1 = sum(e['doc_f1'] for e in evals) / n * 100
mean_page_f1 = sum(e['page_f1'] for e in evals) / n * 100
kuiper = kuiper_statistic(evals)
# By hop type
single_hop = [e for e in evals if e['hop_type'] == 'single']
cross_page = [e for e in evals if e['hop_type'] == 'cross_page']
cross_doc = [e for e in evals if e['hop_type'] == 'cross_doc']
# By domain
by_domain = defaultdict(list)
for e in evals:
domain = e['domain'] or 'Other'
by_domain[domain].append(e)
domain_scores = {}
for domain, domain_evals in sorted(by_domain.items()):
domain_semantic_scores = [e['semantic_score'] for e in domain_evals]
domain_agg = aggregate_anls_star_llm(domain_semantic_scores, apply_bias_correction=True)
domain_scores[domain] = {
'semantic': domain_agg['adjusted_score'] * 100,
'anls': sum(e['anls'] for e in domain_evals) / len(domain_evals) * 100,
'n': len(domain_evals)
}
return {
'overall': {
'semantic': agg['adjusted_score'] * 100,
'semantic_ci': (agg['ci_lower'] * 100, agg['ci_upper'] * 100), # 95% CI
'anls': mean_anls,
'page_f1': mean_page_f1,
'doc_f1': mean_doc_f1,
'kuiper': kuiper['kuiper_stat'] if not kuiper.get('degenerate') else None,
},
'single_evidence': {
'semantic': aggregate_anls_star_llm([e['semantic_score'] for e in single_hop], apply_bias_correction=True)['adjusted_score'] * 100 if single_hop else 0,
'anls': sum(e['anls'] for e in single_hop) / len(single_hop) * 100 if single_hop else 0,
'n': len(single_hop)
},
'multi_evidence_same_doc': {
'semantic': aggregate_anls_star_llm([e['semantic_score'] for e in cross_page], apply_bias_correction=True)['adjusted_score'] * 100 if cross_page else 0,
'anls': sum(e['anls'] for e in cross_page) / len(cross_page) * 100 if cross_page else 0,
'n': len(cross_page)
},
'multi_evidence_multi_doc': {
'semantic': aggregate_anls_star_llm([e['semantic_score'] for e in cross_doc], apply_bias_correction=True)['adjusted_score'] * 100 if cross_doc else 0,
'anls': sum(e['anls'] for e in cross_doc) / len(cross_doc) * 100 if cross_doc else 0,
'n': len(cross_doc)
},
'by_domain': domain_scores,
'n_evaluated': n,
'n_unmatched': unmatched
}
def main():
import argparse
parser = argparse.ArgumentParser(description="Batch re-evaluate submissions")
parser.add_argument('--dry-run', action='store_true', help="List files only, don't evaluate")
parser.add_argument('--org', type=str, help="Filter by organization (e.g., 'OpenAI')")
parser.add_argument('--upload', action='store_true', help="Upload already processed results to HuggingFace Hub (no re-evaluation)")
parser.add_argument('--skip-existing', action='store_true', help="Skip already evaluated files")
args = parser.parse_args()
OUTPUT_DIR.mkdir(exist_ok=True)
# Upload-only mode: just upload existing files
if args.upload:
print("Uploading existing results to HuggingFace Hub...")
api = HfApi()
result_files = list(OUTPUT_DIR.glob("**/*.json"))
print(f"Found {len(result_files)} result files to upload")
for result_file in result_files:
rel_path = result_file.relative_to(OUTPUT_DIR)
print(f" Uploading: {rel_path}")
try:
api.upload_file(
path_or_fileobj=str(result_file),
path_in_repo=str(rel_path),
repo_id=RESULTS_REPO,
repo_type="dataset",
token=TOKEN,
commit_message=f"Re-evaluate with semantic accuracy: {rel_path.stem}"
)
print(f" ✓ Done")
except Exception as e:
print(f" ✗ Error: {e}")
print("\nUpload complete!")
return
# Find prediction files
print("Finding prediction files...")
pred_files = find_prediction_files(args.org)
print(f"Found {len(pred_files)} prediction files")
if args.dry_run:
for f in pred_files:
print(f" - {f}")
return
# Load gold standard
gold_by_id, gold_by_text = load_gold_data()
print(f"Loaded {len(gold_by_id)} gold examples")
# Process each file
for i, pred_file in enumerate(pred_files):
print(f"\n{'='*60}")
print(f"[{i+1}/{len(pred_files)}] Processing: {pred_file}")
print('='*60)
# Check if already processed
output_file = OUTPUT_DIR / (Path(pred_file).stem.replace('_predictions', '_results') + '_reevaluated.json')
if args.skip_existing and output_file.exists():
print(" Skipping (already processed)")
continue
try:
# Download predictions
print(" Downloading predictions...")
local_pred = download_file(pred_file)
predictions = []
with open(local_pred) as f:
for line in f:
if line.strip():
predictions.append(json.loads(line))
print(f" Loaded {len(predictions)} predictions")
# Download original results to preserve metadata
result_file = find_result_file(pred_file)
original_metadata = {}
if result_file:
try:
local_result = download_file(result_file)
with open(local_result) as f:
original_data = json.load(f)
original_metadata = {
'model_name': original_data.get('model_name'),
'organization': original_data.get('organization'),
'description': original_data.get('description'),
'link': original_data.get('link'),
'tags': original_data.get('tags'),
'submitted_by': original_data.get('submitted_by'),
'metadata': original_data.get('metadata'),
'submission_date': original_data.get('submission_date'),
}
print(f" Loaded metadata: model_name={original_metadata.get('model_name')}")
except Exception as e:
print(f" Warning: Could not load original results: {e}")
# Fallback: extract metadata from filename if not found
if not original_metadata.get('model_name'):
# Pattern: Org/Model_Name_with_Stuff_predictions_timestamp.jsonl
filename = Path(pred_file).stem # e.g., GPT-5_(2025-08-07)_with_BM25_Search_Tool_predictions_20260109_152104
parts = filename.rsplit('_predictions_', 1)
if parts:
model_name = parts[0].replace('_', ' ') # Convert underscores to spaces
org = Path(pred_file).parts[0] if '/' in pred_file else 'Unknown'
original_metadata = {
'model_name': model_name,
'organization': org.replace('_', ' '),
'description': '',
'tags': ['Agentic'],
'metadata': {'model_type': 'unknown'},
}
print(f" Using fallback metadata: model_name={model_name}, org={org}")
# Evaluate
print(" Evaluating with semantic accuracy...")
start_time = time.time()
results = evaluate_with_semantic(predictions, gold_by_id, gold_by_text)
elapsed = time.time() - start_time
if results:
print(f"\n Results (took {elapsed:.1f}s):")
print(f" Semantic Accuracy: {results['overall']['semantic']:.1f}")
print(f" ANLS*: {results['overall']['anls']:.1f}")
print(f" Page F1: {results['overall']['page_f1']:.1f}")
# Save with original metadata
org = Path(pred_file).parts[0] if '/' in pred_file else 'Unknown'
output_filename = Path(pred_file).name.replace('_predictions', '_results').replace('.jsonl', '.json')
full_result = {
**original_metadata,
'results': results,
'reevaluated_date': datetime.now(timezone.utc).isoformat(),
'source_predictions_file': pred_file,
'result_file_path': f"{org}/{output_filename}",
}
# Create org subfolder
org_dir = OUTPUT_DIR / org
org_dir.mkdir(exist_ok=True)
output_file = org_dir / output_filename
with open(output_file, 'w') as f:
json.dump(full_result, f, indent=2)
print(f" Saved to: {output_file}")
else:
print(" No valid evaluations")
except Exception as e:
print(f" Error: {e}")
import traceback
traceback.print_exc()
continue
print(f"\n{'='*60}")
print("DONE!")
print(f"Results saved to: {OUTPUT_DIR}")
print(f"\nTo upload results, run: python batch_reevaluate.py --upload")
if __name__ == "__main__":
main()
|