Datasets:
File size: 4,208 Bytes
2faa5f6 | 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 | #!/usr/bin/env python3
"""Validate the public RAISER dataset tree."""
from __future__ import annotations
import argparse
import csv
import json
from pathlib import Path
ROLES = (
"source", "propagator", "missed_verifier", "contract_violation",
"symptom", "repair_leverage",
)
FORBIDDEN_RAW_FIELDS = {
"answer", "answer_key", "content", "conversation_history", "evidence",
"expected_answer", "ground_truth", "history", "model_text", "parsed_answer",
"prompt", "query", "question", "rationale", "raw_content", "raw_span",
"task_text", "thinking", "tool_output",
}
def jsonl(path: Path) -> list[dict]:
rows = []
for number, line in enumerate(path.read_text(encoding="utf-8").splitlines(), 1):
if not line.strip():
continue
row = json.loads(line)
leaked = FORBIDDEN_RAW_FIELDS.intersection(row)
assert not leaked, f"{path}:{number}: raw fields {sorted(leaked)}"
rows.append(row)
return rows
def csv_rows(path: Path) -> list[dict]:
with path.open(encoding="utf-8", newline="") as stream:
rows = list(csv.DictReader(stream))
if rows:
leaked = FORBIDDEN_RAW_FIELDS.intersection(rows[0])
assert not leaked, f"{path}: raw fields {sorted(leaked)}"
return rows
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("root", type=Path, nargs="?", default=Path("."))
root = parser.parse_args().root.resolve()
for filename, tier, expected in (
("consensus_gold.jsonl", "gold", 5516),
("model_silver.jsonl", "silver", 43077),
):
rows = jsonl(root / "data/model_benchmark" / filename)
assert len(rows) == expected
keys = set()
for row in rows:
assert row["annotation_tier"] == tier
assert row["split"] in {"train", "dev", "test"}
key = (row["dataset_source"], row["trace_id"], row["event_id"])
assert key not in keys
keys.add(key)
for role in ROLES:
assert row[role] in {0, 1}
confidence = row[f"{role}_confidence"]
assert confidence is None or 0 <= float(confidence) <= 1
human = jsonl(root / "data/human_gold/mv_cv_test.jsonl")
assert len(human) == 1600
assert len({row["candidate_id"] for row in human}) == 1600
assert {row["target_role"] for row in human} == {
"missed_verifier", "contract_violation"
}
assert all(row["label"] in {0, 1} for row in human)
source = csv_rows(root / "data/event_source/broad_cross_annotation.csv")
assert len(source) == 1267
assert len({row["source_gold_id"] for row in source}) == 1267
assert all(row["label_a"] in {"0", "1", "UNCERTAIN"} for row in source)
assert all(row[field] in {"0", "1"} for row in source for field in (
"label_b", "strict_positive", "positive_union"
))
untouched = csv_rows(root / "data/repair_evaluation/untouched_346.csv")
assert len(untouched) == 346
assert len({row["trace_id"] for row in untouched}) == 346
assert all(row[field] in {"0", "1"} for row in untouched for field in (
"raiser_repair_at_1", "query_only_repair_at_1"
))
component = csv_rows(root / "data/repair_evaluation/component_1200.csv")
assert len(component) == 4800
by_trace: dict[str, set[str]] = {}
for row in component:
by_trace.setdefault(row["trace_id"], set()).add(row["configuration"])
for field in ("oracle_at_k", "rank_at_1", "repair_at_1"):
assert row[field] in {"0", "1"}
assert len(by_trace) == 1200 and all(len(configs) == 4 for configs in by_trace.values())
native = csv_rows(root / "data/repair_evaluation/native_checkpoint_150.csv")
assert len(native) == 600
assert {row["framework"] for row in native} == {"langgraph", "autogen"}
assert {row["selector"] for row in native} == {"RAISER", "answer-majority"}
groups = {(row["framework"], row["trace_id"], row["selector"]) for row in native}
assert len(groups) == 600
assert all(row["success"] in {"0", "1"} for row in native)
print("RAISER dataset validation passed")
if __name__ == "__main__":
main()
|