File size: 7,947 Bytes
6c5f29f | 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 | """Compare two natural OracleMem coverage packages.
The comparison is intentionally conservative. Unit identifiers can differ
across annotators, so the report compares normalized required-unit text and
candidate-coverage text pairs in addition to exact ids.
"""
from __future__ import annotations
import argparse
import json
import re
from pathlib import Path
from typing import Any, Iterable, Mapping
TOKEN_RE = re.compile(r"[a-z0-9]+")
def read_jsonl(path: Path) -> list[dict[str, Any]]:
if not path.exists():
return []
rows: list[dict[str, Any]] = []
with path.open("r", encoding="utf-8") as handle:
for line in handle:
line = line.strip()
if line:
rows.append(json.loads(line))
return rows
def norm_text(value: Any) -> str:
return " ".join(TOKEN_RE.findall(str(value).lower()))
def package_rows(path: Path) -> dict[str, list[dict[str, Any]]]:
return {
"queries": read_jsonl(path / "queries.jsonl"),
"evidence_units": read_jsonl(path / "evidence_units.jsonl"),
"candidate_memories": read_jsonl(path / "candidate_memories.jsonl"),
"coverage_matrix": read_jsonl(path / "coverage_matrix.jsonl"),
}
def unit_text_map(rows: Iterable[Mapping[str, Any]]) -> dict[str, str]:
return {
str(row.get("unit_id")): norm_text(row.get("canonical_text") or row.get("unit_id"))
for row in rows
if row.get("unit_id")
}
def candidate_text_map(rows: Iterable[Mapping[str, Any]]) -> dict[str, str]:
return {
str(row.get("candidate_id")): norm_text(row.get("text") or row.get("serialized") or row.get("candidate_id"))
for row in rows
if row.get("candidate_id")
}
def jaccard(left: set[str], right: set[str]) -> float:
if not left and not right:
return 1.0
union = left | right
if not union:
return 0.0
return len(left & right) / len(union)
def required_texts(query: Mapping[str, Any], unit_text: Mapping[str, str]) -> set[str]:
return {
unit_text.get(str(unit_id), norm_text(unit_id))
for unit_id in query.get("required_unit_ids", []) or []
if unit_text.get(str(unit_id), norm_text(unit_id))
}
def coverage_text_edges(
coverage_rows: Iterable[Mapping[str, Any]],
unit_text: Mapping[str, str],
candidate_text: Mapping[str, str],
) -> set[tuple[str, str]]:
edges: set[tuple[str, str]] = set()
for row in coverage_rows:
cov = float(row.get("coverage", 0.0) or 0.0)
if cov <= 0:
continue
ctext = candidate_text.get(str(row.get("candidate_id")), "")
utext = unit_text.get(str(row.get("unit_id")), "")
if ctext and utext:
edges.add((ctext, utext))
return edges
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--primary", type=Path, required=True)
parser.add_argument("--secondary", type=Path, required=True)
parser.add_argument("--out-dir", type=Path, required=True)
args = parser.parse_args()
primary = package_rows(args.primary)
secondary = package_rows(args.secondary)
args.out_dir.mkdir(parents=True, exist_ok=True)
p_unit_text = unit_text_map(primary["evidence_units"])
s_unit_text = unit_text_map(secondary["evidence_units"])
p_candidate_text = candidate_text_map(primary["candidate_memories"])
s_candidate_text = candidate_text_map(secondary["candidate_memories"])
p_queries = {str(row.get("query_id")): row for row in primary["queries"] if row.get("query_id")}
s_queries = {str(row.get("query_id")): row for row in secondary["queries"] if row.get("query_id")}
common_query_ids = sorted(set(p_queries) & set(s_queries))
agreement_rows: list[dict[str, Any]] = []
exact_required_agree = 0
both_resolved = 0
primary_resolved = 0
secondary_resolved = 0
for query_id in common_query_ids:
p_required = required_texts(p_queries[query_id], p_unit_text)
s_required = required_texts(s_queries[query_id], s_unit_text)
if p_required:
primary_resolved += 1
if s_required:
secondary_resolved += 1
if p_required and s_required:
both_resolved += 1
if p_required == s_required:
exact_required_agree += 1
agreement_rows.append(
{
"query_id": query_id,
"primary_required_texts": sorted(p_required),
"secondary_required_texts": sorted(s_required),
"required_text_jaccard": jaccard(p_required, s_required),
"agreement_class": (
"AGREE"
if p_required == s_required
else "UNRESOLVED"
if not p_required or not s_required
else "MINOR_DISAGREEMENT"
if jaccard(p_required, s_required) >= 0.5
else "MAJOR_DISAGREEMENT"
),
}
)
p_edges = coverage_text_edges(primary["coverage_matrix"], p_unit_text, p_candidate_text)
s_edges = coverage_text_edges(secondary["coverage_matrix"], s_unit_text, s_candidate_text)
summary = {
"schema_version": 1,
"primary": str(args.primary),
"secondary": str(args.secondary),
"common_queries": len(common_query_ids),
"primary_resolved": primary_resolved,
"secondary_resolved": secondary_resolved,
"both_resolved": both_resolved,
"exact_required_text_agreement_rate": (exact_required_agree / len(common_query_ids)) if common_query_ids else 0.0,
"mean_required_text_jaccard": (
sum(float(row["required_text_jaccard"]) for row in agreement_rows) / len(agreement_rows)
if agreement_rows
else 0.0
),
"major_disagreement_count": sum(1 for row in agreement_rows if row["agreement_class"] == "MAJOR_DISAGREEMENT"),
"minor_disagreement_count": sum(1 for row in agreement_rows if row["agreement_class"] == "MINOR_DISAGREEMENT"),
"unresolved_disagreement_count": sum(1 for row in agreement_rows if row["agreement_class"] == "UNRESOLVED"),
"coverage_edge_text_jaccard": jaccard(p_edges, s_edges),
"primary_coverage_edges": len(p_edges),
"secondary_coverage_edges": len(s_edges),
}
(args.out_dir / "summary.json").write_text(json.dumps(summary, indent=2, sort_keys=True) + "\n", encoding="utf-8")
with (args.out_dir / "agreement_rows.jsonl").open("w", encoding="utf-8") as handle:
for row in agreement_rows:
handle.write(json.dumps(row, sort_keys=True) + "\n")
report = [
"# Natural Coverage Annotation Agreement",
"",
f"- Primary: `{args.primary}`",
f"- Secondary: `{args.secondary}`",
f"- Common queries: {summary['common_queries']}",
f"- Primary resolved: {summary['primary_resolved']}",
f"- Secondary resolved: {summary['secondary_resolved']}",
f"- Both resolved: {summary['both_resolved']}",
f"- Exact required-text agreement: {summary['exact_required_text_agreement_rate']:.3f}",
f"- Mean required-text Jaccard: {summary['mean_required_text_jaccard']:.3f}",
f"- Coverage-edge text Jaccard: {summary['coverage_edge_text_jaccard']:.3f}",
f"- Major disagreements: {summary['major_disagreement_count']}",
f"- Minor disagreements: {summary['minor_disagreement_count']}",
f"- Unresolved disagreements: {summary['unresolved_disagreement_count']}",
"",
"This is a model-model agreement audit. It does not certify semantic truth; it identifies which examples need manual adjudication.",
]
(args.out_dir / "REPORT.md").write_text("\n".join(report) + "\n", encoding="utf-8")
print(json.dumps(summary, indent=2, sort_keys=True))
if __name__ == "__main__":
main()
|