Spaces:
Running
Running
File size: 11,829 Bytes
65dfa4b | 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 | """CLI: Pooled retrieval annotation tool (TREC-style).
Runs BM25, vector, and hybrid retrieval on each question, pools unique
chunks, and presents them for human relevance judgment.
Usage:
python scripts/annotate.py
python scripts/annotate.py --question "What is LoRA?"
python scripts/annotate.py --top-k 10
"""
import argparse
import json
import logging
import sys
from datetime import datetime, timezone
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent))
from src.config import PROJECT_ROOT, get_config
from src.ingestion.embeddings import EmbeddingGenerator
from src.retrieval.bm25_retriever import BM25Retriever
from src.retrieval.hybrid_retriever import HybridRetriever
from src.retrieval.vector_retriever import VectorRetriever
from src.storage.chroma_store import ChromaStore
from src.storage.sqlite_db import SQLiteDB
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
)
logger = logging.getLogger(__name__)
QUESTIONS_PATH = PROJECT_ROOT / "data" / "questions.json"
EVAL_SET_PATH = PROJECT_ROOT / "data" / "eval_set.json"
# ββ Retrieval helpers ββββββββββββββββββββββββββββββββββββββββββββββββ
def init_retrievers(config):
"""Initialize DB, BM25, vector, and hybrid retrievers."""
db = SQLiteDB(config.sqlite_db_path)
chroma = ChromaStore(config.chroma_db_path)
embed_gen = EmbeddingGenerator(config.embedding_model)
bm25 = BM25Retriever(db)
bm25.build_index()
vector = VectorRetriever(chroma, embed_gen)
hybrid = HybridRetriever(bm25, vector)
return db, bm25, vector, hybrid
def run_pooled_retrieval(
question: str,
bm25: BM25Retriever,
vector: VectorRetriever,
hybrid: HybridRetriever,
top_k: int = 10,
) -> tuple[list[dict], dict[str, list]]:
"""Run all three methods and pool unique chunks.
Returns:
(pooled_chunks, method_pools) where pooled_chunks has unique chunk_ids
and method_pools maps method name β list of chunk_ids retrieved.
"""
bm25_results = bm25.search(question, top_k=top_k)
vector_results = vector.search(question, top_k=top_k)
hybrid_results = hybrid.search(question, top_k=top_k)
bm25_ids = [r["chunk_id"] for r in bm25_results]
vector_ids = [r["chunk_id"] for r in vector_results]
hybrid_ids = [r["chunk_id"] for r in hybrid_results]
method_pools = {
"bm25_top10": bm25_ids,
"vector_top10": vector_ids,
"hybrid_top10": hybrid_ids,
}
# Deduplicate, preserving first-seen order
seen = set()
pooled_ids = []
for cid in bm25_ids + vector_ids + hybrid_ids:
if cid not in seen:
seen.add(cid)
pooled_ids.append(cid)
return pooled_ids, method_pools
def resolve_chunks(db: SQLiteDB, chunk_ids: list) -> dict:
"""Look up chunk + paper metadata for each chunk ID."""
all_chunks = db.get_all_chunks()
chunk_map = {c["id"]: c for c in all_chunks}
resolved = {}
for cid in chunk_ids:
chunk = chunk_map.get(cid)
if chunk is not None:
resolved[cid] = chunk
return resolved
def chunk_methods(chunk_id, method_pools: dict) -> list[str]:
"""Which methods retrieved this chunk."""
methods = []
for method, ids in method_pools.items():
if chunk_id in ids:
methods.append(method.split("_")[0]) # "bm25", "vector", "hybrid"
return methods
# ββ Annotation state βββββββββββββββββββββββββββββββββββββββββββββββββ
def load_eval_set() -> list[dict]:
if EVAL_SET_PATH.exists():
with open(EVAL_SET_PATH, encoding="utf-8") as f:
return json.load(f)
return []
def save_eval_set(eval_set: list[dict]) -> None:
EVAL_SET_PATH.parent.mkdir(parents=True, exist_ok=True)
with open(EVAL_SET_PATH, "w", encoding="utf-8") as f:
json.dump(eval_set, f, indent=2, ensure_ascii=False)
def find_entry(eval_set: list[dict], question_id: str) -> dict | None:
for entry in eval_set:
if entry.get("id") == question_id:
return entry
return None
# ββ Display helpers ββββββββββββββββββββββββββββββββββββββββββββββββββ
def truncate(text: str, max_len: int = 500) -> str:
if len(text) <= max_len:
return text
return text[:max_len] + "..."
def display_chunk(
idx: int,
total: int,
chunk: dict,
methods: list[str],
show_full: bool = False,
):
"""Print a single chunk for annotation."""
print(f"\n{'='*60}")
print(f" [{idx}/{total}] Methods: {', '.join(methods)}")
print(f" Title: {chunk.get('title', '?')}")
print(f" Venue: {chunk.get('venue', '?')} | Year: {chunk.get('year', '?')}")
print(f" Paper ID: {chunk.get('paper_id', '?')}")
print(f" Chunk ID: {chunk.get('id', '?')} | Type: {chunk.get('chunk_type', '?')}")
print(f"{'β'*60}")
text = chunk.get("chunk_text", "")
if show_full or len(text) <= 500:
print(text)
else:
print(truncate(text, 500))
print(f" [{len(text)} chars total β press 'f' to show full]")
print(f"{'β'*60}")
print(" (y) relevant (n) not relevant (s) skip (f) full text (q) quit")
# ββ Annotation loop ββββββββββββββββββββββββββββββββββββββββββββββββββ
def annotate_question(
question: dict,
db: SQLiteDB,
bm25: BM25Retriever,
vector: VectorRetriever,
hybrid: HybridRetriever,
eval_set: list[dict],
top_k: int = 10,
) -> bool:
"""Annotate one question. Returns False if user quit."""
qid = question["id"]
qtext = question["question"]
print(f"\n{'#'*60}")
print(f" Question [{qid}]: {qtext}")
print(f" Type: {question.get('type', '?')}")
kw = question.get("expected_keywords", [])
if kw:
print(f" Expected keywords: {', '.join(kw)}")
print(f"{'#'*60}")
print("\nRunning retrieval (BM25 + vector + hybrid)...")
pooled_ids, method_pools = run_pooled_retrieval(
qtext, bm25, vector, hybrid, top_k=top_k,
)
resolved = resolve_chunks(db, pooled_ids)
ordered_ids = [cid for cid in pooled_ids if cid in resolved]
total = len(ordered_ids)
if total == 0:
print(" No chunks retrieved. Skipping.")
return True
print(f"\nPooled {total} unique chunks from {top_k * 3} candidates.\n")
relevant_chunk_ids = []
irrelevant_chunk_ids = []
skipped_chunk_ids = []
i = 0
while i < total:
cid = ordered_ids[i]
chunk = resolved[cid]
methods = chunk_methods(cid, method_pools)
display_chunk(i + 1, total, chunk, methods, show_full=False)
action = input(" > ").strip().lower()
if action == "y":
relevant_chunk_ids.append(cid)
i += 1
elif action == "n":
irrelevant_chunk_ids.append(cid)
i += 1
elif action == "s":
skipped_chunk_ids.append(cid)
i += 1
elif action == "f":
display_chunk(i + 1, total, chunk, methods, show_full=True)
# Don't advance β let user judge after seeing full text
elif action == "q":
# Save partial progress before quitting
_save_annotation(
eval_set, question, method_pools,
relevant_chunk_ids, irrelevant_chunk_ids, skipped_chunk_ids,
resolved,
)
return False
else:
print(" Invalid. Use y/n/s/f/q.")
# Save completed annotation
_save_annotation(
eval_set, question, method_pools,
relevant_chunk_ids, irrelevant_chunk_ids, skipped_chunk_ids,
resolved,
)
return True
def _save_annotation(
eval_set: list[dict],
question: dict,
method_pools: dict,
relevant_chunk_ids: list,
irrelevant_chunk_ids: list,
skipped_chunk_ids: list,
resolved: dict,
):
"""Build and save the annotation entry."""
# Derive relevant paper IDs from relevant chunks
relevant_paper_ids = list(dict.fromkeys(
resolved[cid]["paper_id"]
for cid in relevant_chunk_ids
if cid in resolved
))
# Stringify chunk IDs for JSON
def to_str_ids(ids):
return [str(x) for x in ids]
entry = {
"id": question["id"],
"question": question["question"],
"type": question.get("type", ""),
"expected_keywords": question.get("expected_keywords", []),
"relevant_chunk_ids": to_str_ids(relevant_chunk_ids),
"irrelevant_chunk_ids": to_str_ids(irrelevant_chunk_ids),
"skipped_chunk_ids": to_str_ids(skipped_chunk_ids),
"relevant_paper_ids": relevant_paper_ids,
"pooled_from": {
k: to_str_ids(v) for k, v in method_pools.items()
},
"annotated_at": datetime.now(timezone.utc).isoformat(),
}
# Replace or append
existing = find_entry(eval_set, question["id"])
if existing is not None:
idx = eval_set.index(existing)
eval_set[idx] = entry
else:
eval_set.append(entry)
save_eval_set(eval_set)
n_rel = len(relevant_chunk_ids)
n_irr = len(irrelevant_chunk_ids)
n_skip = len(skipped_chunk_ids)
print(f"\n Saved: {n_rel} relevant, {n_irr} irrelevant, {n_skip} skipped")
# ββ Main βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def main():
parser = argparse.ArgumentParser(
description="Pooled retrieval annotation (TREC-style)"
)
parser.add_argument(
"--question", type=str, default=None,
help="Annotate a single ad-hoc question (bypasses questions.json)",
)
parser.add_argument("--top-k", type=int, default=10, help="Results per method")
parser.add_argument(
"--force", action="store_true",
help="Re-annotate questions that already have judgments",
)
args = parser.parse_args()
config = get_config()
db, bm25, vector, hybrid = init_retrievers(config)
eval_set = load_eval_set()
if args.question:
# Ad-hoc single question
q = {
"id": f"adhoc_{datetime.now(timezone.utc).strftime('%Y%m%d%H%M%S')}",
"question": args.question,
"type": "factual",
"expected_keywords": [],
}
annotate_question(q, db, bm25, vector, hybrid, eval_set, top_k=args.top_k)
return
# Load questions from file
if not QUESTIONS_PATH.exists():
print(f"No questions file found at {QUESTIONS_PATH}")
print("Run: python scripts/write_questions.py")
sys.exit(1)
with open(QUESTIONS_PATH, encoding="utf-8") as f:
questions = json.load(f)
if not questions:
print("Questions file is empty. Add questions first.")
sys.exit(1)
annotated_ids = {e["id"] for e in eval_set}
for q in questions:
qid = q["id"]
if qid in annotated_ids and not args.force:
print(f"\n [{qid}] already annotated β skipping (use --force to redo)")
continue
if not annotate_question(q, db, bm25, vector, hybrid, eval_set, top_k=args.top_k):
print("\nAnnotation paused. Progress saved.")
break
print(f"\nDone. {len(eval_set)} entries in {EVAL_SET_PATH}")
if __name__ == "__main__":
main()
|