File size: 14,815 Bytes
715cc5a | 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 | from __future__ import annotations
import argparse
from pathlib import Path
from typing import Any
from src.analysis.compare_predictions import compare_prediction_maps, load_prediction_map, row_id
from src.data.io_utils import read_jsonl, write_jsonl
BIOMEDICAL_RELATIONS = {"TREATS", "PREVENTS", "CAUSES", "INCREASES_RISK", "DECREASES_RISK"}
def compact(text: Any, limit: int = 220) -> str:
value = " ".join(str(text or "").split())
if len(value) <= limit:
return value
clipped = value[:limit].rsplit(" ", 1)[0].strip()
return f"{clipped}..."
def load_map(path: Path) -> dict[str, dict[str, Any]]:
return {row_id(row): row for row in read_jsonl(path)}
def load_claims_map(path: Path) -> dict[str, dict[str, Any]]:
rows = read_jsonl(path)
mapping: dict[str, dict[str, Any]] = {}
for row in rows:
for key in ("pair_id", "claim_id"):
value = row.get(key)
if value:
mapping[str(value)] = row
return mapping
def group_rows(path: Path, key_field: str = "claim_id") -> dict[str, list[dict[str, Any]]]:
grouped: dict[str, list[dict[str, Any]]] = {}
for row in read_jsonl(path):
grouped.setdefault(str(row.get(key_field, "")), []).append(row)
return grouped
def evidence_rows(topk_row: dict[str, Any] | None, limit: int = 3) -> list[dict[str, Any]]:
if not topk_row:
return []
rows: list[dict[str, Any]] = []
for item in (topk_row.get("candidates") or [])[:limit]:
rows.append(
{
"candidate_id": item.get("candidate_id") or item.get("doc_id"),
"rank": item.get("rank_wikikg") or item.get("final_rank") or item.get("rank_reranker"),
"wikikg_score": item.get("wikikg_final_score"),
"reranker_score": item.get("reranker_score"),
"text": compact(item.get("text", ""), limit=260),
}
)
return rows
def path_rows(subgraph: dict[str, Any] | None, topk_row: dict[str, Any] | None, limit: int = 3) -> list[dict[str, Any]]:
if not subgraph:
return []
preferred = {
str(item.get("candidate_id") or item.get("doc_id"))
for item in (topk_row or {}).get("candidates", [])[:5]
}
triples = list(subgraph.get("triples") or [])
triples.sort(
key=lambda row: (
0 if str(row.get("source_doc_id", "")) in preferred else 1,
str(row.get("triple_id", "")),
)
)
output: list[dict[str, Any]] = []
for triple in triples[:limit]:
output.append(
{
"path_text": f"{triple.get('head', '')} -- {triple.get('relation', '')} -- {triple.get('tail', '')}",
"source_doc_id": triple.get("source_doc_id", ""),
"source_text": compact(triple.get("source_text", ""), limit=260),
}
)
return output
def unsupported_rows(unsupported: list[dict[str, Any]], limit: int = 3) -> list[dict[str, Any]]:
rows = sorted(
unsupported,
key=lambda row: (
0 if "relation_demoted_to_associated" in set(row.get("removal_reasons") or []) | set(row.get("rule_notes") or []) else 1,
str(row.get("claim_id", "")),
),
)
output: list[dict[str, Any]] = []
for row in rows[:limit]:
output.append(
{
"path_text": f"{row.get('head', '')} -- {row.get('relation', '')} -- {row.get('tail', '')}",
"nli_label": row.get("nli_label", ""),
"removal_reasons": row.get("removal_reasons", []),
"source_text": compact(row.get("source_text", ""), limit=260),
}
)
return output
def feature_summary(feature_row: dict[str, Any] | None) -> dict[str, Any]:
if not feature_row:
return {}
candidates = list(feature_row.get("candidate_features") or [])
if not candidates:
return {}
max_final = max(float(item.get("final_score") or 0.0) for item in candidates)
max_path = max(float(item.get("kg_path_score") or 0.0) for item in candidates)
max_provenance = max(float(item.get("provenance_confidence") or 0.0) for item in candidates)
max_contradiction = max(float(item.get("contradiction_signal") or 0.0) for item in candidates)
return {
"max_final_score": round(max_final, 6),
"max_kg_path_score": round(max_path, 6),
"max_provenance_confidence": round(max_provenance, 6),
"max_contradiction_signal": round(max_contradiction, 6),
}
def build_prediction_case(
comparison: dict[str, Any],
category: str,
claims_map: dict[str, dict[str, Any]],
topk_map: dict[str, dict[str, Any]],
feature_map: dict[str, dict[str, Any]],
subgraph_map: dict[str, dict[str, Any]],
unsupported_map: dict[str, list[dict[str, Any]]],
) -> dict[str, Any]:
claim_id = comparison["id"]
claim_row = claims_map.get(claim_id, {})
topk_row = topk_map.get(claim_id)
subgraph = subgraph_map.get(claim_id)
unsupported = unsupported_map.get(claim_id, [])
return {
"dataset": comparison["dataset"],
"split": comparison["split"],
"category": category,
"id": claim_id,
"claim": comparison["claim"],
"gold": comparison["gold"],
"baseline_prediction": comparison["baseline_prediction"],
"wikikg_prediction": comparison["wikikg_prediction"],
"alternate_prediction": comparison["alternate_prediction"],
"baseline_correct": comparison["baseline_correct"],
"wikikg_correct": comparison["wikikg_correct"],
"alternate_correct": comparison["alternate_correct"],
"label": claim_row.get("label", ""),
"metadata": claim_row.get("metadata", {}),
"retrieval_metrics": {} if not topk_row else topk_row.get("metrics", {}),
"path_summary": feature_summary(feature_map.get(claim_id)),
"num_verified_facts": 0 if not subgraph else int(subgraph.get("num_facts") or 0),
"num_verified_triples": 0 if not subgraph else int(subgraph.get("num_triples") or 0),
"top_evidence": evidence_rows(topk_row),
"top_verified_paths": path_rows(subgraph, topk_row),
"top_unsupported_triples": unsupported_rows(unsupported),
}
def pick_prediction_cases(
comparisons: list[dict[str, Any]],
dataset: str,
claims_map: dict[str, dict[str, Any]],
topk_map: dict[str, dict[str, Any]],
feature_map: dict[str, dict[str, Any]],
subgraph_map: dict[str, dict[str, Any]],
unsupported_map: dict[str, list[dict[str, Any]]],
sample_per_category: int,
) -> list[dict[str, Any]]:
used: set[str] = set()
output: list[dict[str, Any]] = []
def take(category: str, predicate) -> None:
count = 0
for row in sorted(comparisons, key=lambda item: item["id"]):
if row["id"] in used or not predicate(row):
continue
output.append(
build_prediction_case(
row,
category,
claims_map,
topk_map,
feature_map,
subgraph_map,
unsupported_map,
)
)
used.add(row["id"])
count += 1
if count >= sample_per_category:
break
if dataset == "averitec":
take("baseline_wrong_wikikg_right", lambda row: row["baseline_correct"] is False and row["wikikg_correct"] is True)
take(
"verified_beats_unfiltered",
lambda row: row["wikikg_correct"] is True and row["alternate_correct"] is False,
)
take(
"nei_recovered_by_paths",
lambda row: row["gold"] == "NEI" and row["baseline_correct"] is False and row["wikikg_correct"] is True,
)
take("conflicting_failure", lambda row: row["gold"] == "CONFLICTING" and row["wikikg_correct"] is False)
take("wikikg_hurt_case", lambda row: row["baseline_correct"] is True and row["wikikg_correct"] is False)
else:
take("baseline_wrong_wikikg_right", lambda row: row["baseline_correct"] is False and row["wikikg_correct"] is True)
take("wikikg_hurt_case", lambda row: row["baseline_correct"] is True and row["wikikg_correct"] is False)
take(
"good_path_wrong_verdict",
lambda row: row["wikikg_correct"] is False and (subgraph_map.get(row["id"], {}).get("num_triples", 0) or subgraph_map.get(row["id"], {}).get("num_facts", 0)),
)
return output
def pick_healthver_relation_cases(
claims_map: dict[str, dict[str, Any]],
verified_map: dict[str, list[dict[str, Any]]],
unsupported_map: dict[str, list[dict[str, Any]]],
baseline_map: dict[str, dict[str, Any]] | None,
wikikg_map: dict[str, dict[str, Any]] | None,
sample_per_category: int,
) -> list[dict[str, Any]]:
used: set[tuple[str, str]] = set()
output: list[dict[str, Any]] = []
def add_case(category: str, row: dict[str, Any]) -> None:
key = (category, str(row.get("triple_id", "")))
if key in used:
return
claim_id = str(row.get("claim_id", ""))
claim_row = claims_map.get(claim_id, {})
baseline_pred = "" if not baseline_map or claim_id not in baseline_map else baseline_map[claim_id].get("prediction", "")
wikikg_pred = "" if not wikikg_map or claim_id not in wikikg_map else wikikg_map[claim_id].get("prediction", "")
output.append(
{
"dataset": "healthver",
"split": row.get("split", ""),
"category": category,
"id": claim_id,
"claim": row.get("claim", ""),
"gold": claim_row.get("label", ""),
"baseline_prediction": baseline_pred,
"wikikg_prediction": wikikg_pred,
"relation_original": row.get("relation_original", ""),
"relation": row.get("relation", ""),
"nli_label": row.get("nli_label", ""),
"entailment_score": row.get("entailment_score", ""),
"rule_notes": row.get("rule_notes", []),
"removal_reasons": row.get("removal_reasons", []),
"source_text": compact(row.get("source_text", ""), limit=320),
"verbalized_triple": row.get("verbalized_triple", ""),
"metadata": claim_row.get("metadata", {}),
}
)
used.add(key)
demoted_verified = [
row
for rows in verified_map.values()
for row in rows
if row.get("relation_original") and row.get("relation_original") != row.get("relation")
]
demoted_removed = [
row
for rows in unsupported_map.values()
for row in rows
if "relation_demoted_to_associated" in set(row.get("removal_reasons") or []) | set(row.get("rule_notes") or [])
]
strong_verified = [
row
for rows in verified_map.values()
for row in rows
if row.get("relation") in BIOMEDICAL_RELATIONS and not row.get("relation_original")
]
strong_removed = [
row
for rows in unsupported_map.values()
for row in rows
if row.get("relation") in BIOMEDICAL_RELATIONS or row.get("relation_original") in BIOMEDICAL_RELATIONS
]
for bucket_name, rows in (
("demoted_verified_relation", demoted_verified),
("removed_strong_relation", demoted_removed),
("verified_strong_relation", strong_verified),
("unsupported_biomedical_relation", strong_removed),
):
count = 0
for row in sorted(rows, key=lambda item: str(item.get("claim_id", ""))):
add_case(bucket_name, row)
count += 1
if count >= sample_per_category:
break
return output
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", required=True, choices=["averitec", "vifactcheck", "healthver"])
parser.add_argument("--split", required=True)
parser.add_argument("--claims", type=Path, required=True)
parser.add_argument("--retrieval-topk", type=Path)
parser.add_argument("--path-features", type=Path)
parser.add_argument("--verified-subgraphs", type=Path)
parser.add_argument("--verified-triples", type=Path, required=True)
parser.add_argument("--unsupported-triples", type=Path, required=True)
parser.add_argument("--baseline-pred", type=Path)
parser.add_argument("--wikikg-pred", type=Path)
parser.add_argument("--alternate-pred", type=Path)
parser.add_argument("--sample-per-category", type=int, default=3)
parser.add_argument("--output", type=Path, required=True)
args = parser.parse_args()
claims_map = load_claims_map(args.claims)
verified_map = group_rows(args.verified_triples)
unsupported_map = group_rows(args.unsupported_triples)
if args.dataset == "healthver":
baseline_map = load_prediction_map(args.baseline_pred) if args.baseline_pred else None
wikikg_map = load_prediction_map(args.wikikg_pred) if args.wikikg_pred else None
rows = pick_healthver_relation_cases(
claims_map=claims_map,
verified_map=verified_map,
unsupported_map=unsupported_map,
baseline_map=baseline_map,
wikikg_map=wikikg_map,
sample_per_category=args.sample_per_category,
)
write_jsonl(args.output, rows)
print(f"Wrote {len(rows)} healthver relation cases to {args.output}")
return
if not (args.baseline_pred and args.wikikg_pred and args.retrieval_topk and args.path_features and args.verified_subgraphs):
raise SystemExit("Prediction-comparison datasets require baseline/wikikg predictions and retrieval/subgraph inputs")
baseline_map = load_prediction_map(args.baseline_pred)
wikikg_map = load_prediction_map(args.wikikg_pred)
alternate_map = load_prediction_map(args.alternate_pred) if args.alternate_pred else None
comparisons = compare_prediction_maps(baseline_map, wikikg_map, alternate_map)
topk_map = load_map(args.retrieval_topk)
feature_map = load_map(args.path_features)
subgraph_map = load_map(args.verified_subgraphs)
rows = pick_prediction_cases(
comparisons=comparisons,
dataset=args.dataset,
claims_map=claims_map,
topk_map=topk_map,
feature_map=feature_map,
subgraph_map=subgraph_map,
unsupported_map=unsupported_map,
sample_per_category=args.sample_per_category,
)
write_jsonl(args.output, rows)
print(f"Wrote {len(rows)} case studies to {args.output}")
if __name__ == "__main__":
main()
|