File size: 12,231 Bytes
3549219 2ad523a 3549219 95974bc 3549219 2ad523a 3549219 95974bc 3549219 2ad523a 3549219 95974bc 3549219 95974bc 3549219 | 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 | """
DAPT corpus validation β run BEFORE training on collected Reddit data.
Checks:
1. Data leakage: DAPT corpus must NOT overlap with ReDSM5 labeled data
2. Deduplication: exact and near-duplicate removal
3. Language quality: filter non-English, bot/spam, too-short posts
4. Corpus statistics: length distribution, vocabulary
5. Provenance documentation
Usage:
python validate_dapt_corpus.py
"""
import hashlib
import json
import re
from collections import Counter
from pathlib import Path
import pandas as pd
def load_redsm5_texts(redsm5_dir: Path) -> set[str]:
"""Load all labeled ReDSM5 text for leakage detection."""
posts = pd.read_csv(redsm5_dir / "redsm5_posts.csv")
annotations = pd.read_csv(redsm5_dir / "redsm5_annotations.csv")
# Collect all text that appears in the labeled dataset
texts = set()
for t in posts["text"].dropna():
texts.add(t.strip().lower())
for t in annotations["sentence_text"].dropna():
texts.add(t.strip().lower())
return texts
def normalize_for_dedup(text: str) -> str:
"""Normalize text for near-duplicate detection."""
text = text.lower().strip()
text = re.sub(r"\s+", " ", text)
text = re.sub(r"[^\w\s]", "", text) # remove punctuation
return text
def detect_bot_spam(text: str) -> bool:
"""Heuristic bot/spam detection."""
# Repetitive character patterns
if re.search(r"(.)\1{10,}", text):
return True
# All caps posts (shouting/spam)
alpha_chars = [c for c in text if c.isalpha()]
if len(alpha_chars) > 20 and sum(1 for c in alpha_chars if c.isupper()) / len(alpha_chars) > 0.8:
return True
# Promotional patterns
promo_patterns = [
r"check out my",
r"subscribe to",
r"use code",
r"discount",
r"buy now",
r"click here",
r"free trial",
r"www\.\S+\.(com|org|net)",
]
return any(re.search(pattern, text, re.IGNORECASE) for pattern in promo_patterns)
def detect_non_english(text: str) -> bool:
"""Simple heuristic: if <60% of words are common English words, flag it.
Uses a small set of the most common English words as a proxy.
"""
common_words = {
"the",
"be",
"to",
"of",
"and",
"a",
"in",
"that",
"have",
"i",
"it",
"for",
"not",
"on",
"with",
"he",
"as",
"you",
"do",
"at",
"this",
"but",
"his",
"by",
"from",
"they",
"we",
"say",
"her",
"she",
"or",
"an",
"will",
"my",
"one",
"all",
"would",
"there",
"their",
"what",
"so",
"up",
"out",
"if",
"about",
"who",
"get",
"which",
"go",
"me",
"when",
"make",
"can",
"like",
"time",
"no",
"just",
"him",
"know",
"take",
"people",
"into",
"year",
"your",
"good",
"some",
"could",
"them",
"see",
"other",
"than",
"then",
"now",
"look",
"only",
"come",
"its",
"over",
"think",
"also",
"back",
"after",
"use",
"two",
"how",
"our",
"work",
"first",
"well",
"way",
"even",
"new",
"want",
"because",
"any",
"these",
"give",
"day",
"most",
"us",
"am",
"been",
"was",
"is",
"are",
"don",
"feel",
"really",
"life",
"much",
"very",
"still",
"never",
"going",
"help",
"being",
"too",
"need",
"had",
"did",
"has",
}
words = text.lower().split()
if len(words) < 5:
return False # too short to judge
common_count = sum(1 for w in words if w in common_words)
ratio = common_count / len(words)
return ratio < 0.15 # very low overlap with English
def main():
project_root = Path(__file__).parent.parent.parent.parent
base_dir = Path(__file__).parent.parent
corpus_dir = base_dir / "data" / "dapt_corpus"
redsm5_dir = project_root / "redsm5"
corpus_file = corpus_dir / "dapt_corpus.txt"
print("=" * 60)
print("DAPT CORPUS VALIDATION")
print("=" * 60)
# Load corpus
with open(corpus_file) as f:
raw = f.read()
chunks = [c.strip() for c in raw.split("\n\n") if len(c.strip()) > 0]
print(f"\nRaw chunks loaded: {len(chunks)}")
# ββ CHECK 1: Data Leakage ββββββββββββββββββββββββββββββββββββββββββββββ
print("\nββ CHECK 1: Data Leakage Against ReDSM5 ββ")
redsm5_texts = load_redsm5_texts(redsm5_dir)
print(f" ReDSM5 reference texts: {len(redsm5_texts)}")
leaked = []
for i, chunk in enumerate(chunks):
chunk_lower = chunk.strip().lower()
# Exact match
if chunk_lower in redsm5_texts:
leaked.append((i, chunk[:100]))
continue
# Substring match β check if any ReDSM5 sentence appears in the DAPT chunk
for ref in redsm5_texts:
if len(ref) >= 30 and ref in chunk_lower:
leaked.append((i, f"contains: '{ref[:80]}...'"))
break
if leaked:
print(f" LEAKAGE DETECTED: {len(leaked)} chunks overlap with ReDSM5!")
for idx, preview in leaked[:5]:
print(f" chunk {idx}: {preview}")
if len(leaked) > 5:
print(f" ... and {len(leaked) - 5} more")
else:
print(" NO LEAKAGE β zero overlap with ReDSM5 labeled data")
# ββ CHECK 2: Exact Deduplication βββββββββββββββββββββββββββββββββββββββ
print("\nββ CHECK 2: Deduplication ββ")
seen_hashes = {}
exact_dupes = 0
unique_chunks = []
dupe_indices = set()
for i, chunk in enumerate(chunks):
h = hashlib.md5(chunk.encode()).hexdigest()
if h in seen_hashes:
exact_dupes += 1
dupe_indices.add(i)
else:
seen_hashes[h] = i
unique_chunks.append(chunk)
print(f" Exact duplicates: {exact_dupes}")
print(f" Unique chunks: {len(unique_chunks)}")
# Near-duplicate detection (normalized text)
seen_normalized = {}
near_dupes = 0
final_chunks = []
for chunk in unique_chunks:
norm = normalize_for_dedup(chunk)
if norm in seen_normalized:
near_dupes += 1
else:
seen_normalized[norm] = True
final_chunks.append(chunk)
print(f" Near-duplicates (after normalization): {near_dupes}")
print(f" After dedup: {len(final_chunks)} chunks")
# ββ CHECK 3: Bot/Spam Detection ββββββββββββββββββββββββββββββββββββββββ
print("\nββ CHECK 3: Bot/Spam Detection ββ")
spam_count = 0
spam_indices = []
clean_chunks = []
for i, chunk in enumerate(final_chunks):
if detect_bot_spam(chunk):
spam_count += 1
spam_indices.append(i)
else:
clean_chunks.append(chunk)
print(f" Bot/spam detected: {spam_count}")
if spam_indices[:3]:
for idx in spam_indices[:3]:
print(f" example: '{final_chunks[idx][:100]}...'")
print(f" After spam removal: {len(clean_chunks)} chunks")
# ββ CHECK 4: Language Quality ββββββββββββββββββββββββββββββββββββββββββ
print("\nββ CHECK 4: Language Quality ββ")
non_english = 0
too_short = 0
quality_chunks = []
for chunk in clean_chunks:
if len(chunk) < 30:
too_short += 1
continue
if detect_non_english(chunk):
non_english += 1
continue
quality_chunks.append(chunk)
print(f" Too short (<30 chars): {too_short}")
print(f" Non-English (heuristic): {non_english}")
print(f" After quality filter: {len(quality_chunks)} chunks")
# ββ CHECK 5: Corpus Statistics βββββββββββββββββββββββββββββββββββββββββ
print("\nββ CHECK 5: Corpus Statistics ββ")
lengths = [len(c) for c in quality_chunks]
word_counts = [len(c.split()) for c in quality_chunks]
total_chars = sum(lengths)
total_words = sum(word_counts)
print(f" Final chunks: {len(quality_chunks)}")
print(f" Total characters: {total_chars:,}")
print(f" Total words: {total_words:,}")
print(f" Estimated tokens (~0.75 words/token): {int(total_words / 0.75):,}")
print(f" Avg chars/chunk: {total_chars / len(quality_chunks):.0f}")
print(f" Avg words/chunk: {total_words / len(quality_chunks):.0f}")
print(f" Median chars: {sorted(lengths)[len(lengths) // 2]}")
print(f" Min chars: {min(lengths)}, Max chars: {max(lengths)}")
# Length distribution buckets
buckets = Counter()
for l in lengths:
if l < 100:
buckets["<100"] += 1
elif l < 300:
buckets["100-300"] += 1
elif l < 500:
buckets["300-500"] += 1
elif l < 1000:
buckets["500-1000"] += 1
else:
buckets["1000+"] += 1
print("\n Length distribution:")
for bucket in ["<100", "100-300", "300-500", "500-1000", "1000+"]:
count = buckets.get(bucket, 0)
pct = count / len(quality_chunks) * 100
bar = "#" * int(pct / 2)
print(f" {bucket:>10}: {count:>6} ({pct:>5.1f}%) {bar}")
# ββ Save cleaned corpus ββββββββββββββββββββββββββββββββββββββββββββββββ
print("\nββ Saving cleaned corpus ββ")
cleaned_file = corpus_dir / "dapt_corpus_cleaned.txt"
with open(cleaned_file, "w") as f:
f.write("\n\n".join(quality_chunks))
# Save validation report
report = {
"raw_chunks": len(chunks),
"leakage_detected": len(leaked),
"exact_duplicates_removed": exact_dupes,
"near_duplicates_removed": near_dupes,
"spam_removed": spam_count,
"too_short_removed": too_short,
"non_english_removed": non_english,
"final_chunks": len(quality_chunks),
"total_characters": total_chars,
"total_words": total_words,
"estimated_tokens": int(total_words / 0.75),
"avg_chars_per_chunk": round(total_chars / len(quality_chunks)),
"avg_words_per_chunk": round(total_words / len(quality_chunks)),
"validation_passed": len(leaked) == 0,
}
with open(corpus_dir / "validation_report.json", "w") as f:
json.dump(report, f, indent=2)
print(f" Cleaned corpus: {cleaned_file}")
print(f" Validation report: {corpus_dir / 'validation_report.json'}")
# ββ VERDICT ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
print(f"\n{'=' * 60}")
if len(leaked) > 0:
print("VERDICT: FAIL β Data leakage detected. Must remove leaked chunks before DAPT.")
else:
removed = len(chunks) - len(quality_chunks)
print(f"VERDICT: PASS β No leakage. Removed {removed} chunks ({removed / len(chunks) * 100:.1f}%) for quality.")
print(
f"Cleaned corpus: {len(quality_chunks)} chunks, {total_words:,} words, ~{int(total_words / 0.75):,} tokens"
)
print("=" * 60)
if __name__ == "__main__":
main()
|