File size: 8,938 Bytes
785dbb4 | 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 | #!/usr/bin/env python3
"""Evaluate Sys1Cal-v1 predictions against an exact-probability JSONL split."""
from __future__ import annotations
import argparse
import json
import statistics
import sys
from collections import defaultdict
from pathlib import Path
from typing import Any
def read_jsonl(path: Path) -> list[dict[str, Any]]:
rows = []
with path.open("r", encoding="utf-8") as handle:
for line_number, line in enumerate(handle, start=1):
if not line.strip():
continue
try:
rows.append(json.loads(line))
except json.JSONDecodeError as exc:
raise SystemExit(f"{path}:{line_number}: invalid JSON: {exc}") from exc
return rows
def normalize(distribution: dict[str, Any], outcomes: list[str]) -> dict[str, float]:
cleaned = {label: max(0.0, float(distribution.get(label, 0.0))) for label in outcomes}
total = sum(cleaned.values())
if total <= 0.0:
return {label: 1.0 / len(outcomes) for label in outcomes}
return {label: value / total for label, value in cleaned.items()}
def argmax(distribution: dict[str, float]) -> str:
return max(distribution.items(), key=lambda item: (item[1], item[0]))[0]
def total_variation(
gold: dict[str, float], prediction: dict[str, float], outcomes: list[str]
) -> float:
return 0.5 * sum(abs(gold[label] - prediction[label]) for label in outcomes)
def brier_regret(
gold: dict[str, float], prediction: dict[str, float], outcomes: list[str]
) -> float:
return sum((prediction[label] - gold[label]) ** 2 for label in outcomes)
def extract_distribution(row: dict[str, Any], item: dict[str, Any]) -> dict[str, float]:
outcomes = item["outcomes"]
for key in ("predicted_distribution", "distribution", "prediction", "probabilities"):
value = row.get(key)
if isinstance(value, dict):
return normalize(map_labels(value, outcomes), outcomes)
if "probability_true" in row:
p_true = float(row["probability_true"])
return normalize({"True": p_true, "False": 1.0 - p_true}, outcomes)
if "p_true" in row:
p_true = float(row["p_true"])
return normalize({"True": p_true, "False": 1.0 - p_true}, outcomes)
raise ValueError(
"prediction row must contain predicted_distribution, distribution, "
"prediction, probabilities, probability_true, or p_true"
)
def map_labels(distribution: dict[str, Any], outcomes: list[str]) -> dict[str, Any]:
if all(label in distribution for label in outcomes):
return distribution
lower_to_label = {str(label).lower(): label for label in outcomes}
mapped = {}
for raw_label, value in distribution.items():
label = lower_to_label.get(str(raw_label).lower())
if label is not None:
mapped[label] = value
return mapped
def prediction_group(row: dict[str, Any]) -> str:
model = row.get("model") or row.get("model_name") or "submission"
response_type = row.get("response_type")
api_metadata = row.get("api_metadata")
if response_type is None and isinstance(api_metadata, dict):
response_type = api_metadata.get("question_type")
return f"{model}:{response_type}" if response_type else str(model)
def summarize(values: list[float]) -> dict[str, float]:
if not values:
return {
"mean": 0.0,
"median": 0.0,
"std": 0.0,
"p90": 0.0,
"p95": 0.0,
"max": 0.0,
}
ordered = sorted(values)
return {
"mean": statistics.fmean(values),
"median": statistics.median(values),
"std": statistics.pstdev(values) if len(values) > 1 else 0.0,
"p90": ordered[min(len(ordered) - 1, int(0.90 * (len(ordered) - 1)))],
"p95": ordered[min(len(ordered) - 1, int(0.95 * (len(ordered) - 1)))],
"max": max(values),
}
def aggregate(rows: list[dict[str, Any]]) -> dict[str, Any]:
tv_values = [row["tv"] for row in rows]
brier_values = [row["brier_regret"] for row in rows]
accuracy_values = [1.0 if row["argmax_correct"] else 0.0 for row in rows]
return {
"n_predictions": len(rows),
"accuracy": statistics.fmean(accuracy_values) if accuracy_values else 0.0,
"pointwise_probability_fidelity": 1.0 - statistics.fmean(tv_values)
if tv_values
else 0.0,
"tv": summarize(tv_values),
"brier_regret": summarize(brier_values),
}
def evaluate(dataset_path: Path, predictions_path: Path, strict: bool) -> dict[str, Any]:
items = {row["instance_id"]: row for row in read_jsonl(dataset_path)}
prediction_rows = read_jsonl(predictions_path)
scored_rows = []
errors = []
for index, row in enumerate(prediction_rows, start=1):
instance_id = row.get("instance_id")
if instance_id not in items:
errors.append(f"prediction row {index}: unknown instance_id {instance_id!r}")
continue
item = items[instance_id]
try:
pred = extract_distribution(row, item)
except Exception as exc:
errors.append(f"prediction row {index}: {exc}")
continue
gold = {key: float(value) for key, value in item["gold_distribution"].items()}
outcomes = item["outcomes"]
tv = total_variation(gold, pred, outcomes)
scored_rows.append(
{
"group": prediction_group(row),
"instance_id": instance_id,
"latent_instance_id": item["latent_instance_id"],
"family": item["family"],
"representation": item["representation"],
"probability_band": item.get("probability_band"),
"tv": tv,
"pointwise_probability_fidelity": 1.0 - tv,
"brier_regret": brier_regret(gold, pred, outcomes),
"gold_argmax": item["gold_argmax"],
"predicted_argmax": argmax(pred),
"argmax_correct": argmax(gold) == argmax(pred),
}
)
covered = {row["instance_id"] for row in scored_rows}
missing = sorted(set(items) - covered)
if strict and (errors or missing):
for error in errors[:20]:
print(error, file=sys.stderr)
if len(errors) > 20:
print(f"... {len(errors) - 20} more prediction errors", file=sys.stderr)
if missing:
print(f"missing predictions for {len(missing)} dataset rows", file=sys.stderr)
raise SystemExit(1)
by_group = defaultdict(list)
by_family = defaultdict(list)
by_representation = defaultdict(list)
by_probability_band = defaultdict(list)
for row in scored_rows:
by_group[row["group"]].append(row)
by_family[(row["group"], row["family"])].append(row)
by_representation[(row["group"], row["representation"])].append(row)
by_probability_band[(row["group"], row["probability_band"])].append(row)
return {
"dataset_path": str(dataset_path),
"predictions_path": str(predictions_path),
"n_dataset_items": len(items),
"n_prediction_rows": len(prediction_rows),
"n_scored_rows": len(scored_rows),
"n_missing_items": len(missing),
"n_errors": len(errors),
"errors": errors,
"overall": aggregate(scored_rows),
"by_group": {str(key): aggregate(rows) for key, rows in sorted(by_group.items())},
"by_family": {
f"{key[0]} / {key[1]}": aggregate(rows)
for key, rows in sorted(by_family.items())
},
"by_representation": {
f"{key[0]} / {key[1]}": aggregate(rows)
for key, rows in sorted(by_representation.items())
},
"by_probability_band": {
f"{key[0]} / {key[1]}": aggregate(rows)
for key, rows in sorted(by_probability_band.items())
},
}
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--dataset",
type=Path,
default=Path("data/v0.1.0_tiny_cleanvars_parallel_primitives.jsonl"),
help="Sys1Cal-v1 JSONL file.",
)
parser.add_argument("--predictions", type=Path, required=True, help="Prediction JSONL.")
parser.add_argument("--output", type=Path, help="Optional path for summary JSON.")
parser.add_argument(
"--strict",
action="store_true",
help="Fail on unknown instance IDs, malformed predictions, or missing items.",
)
args = parser.parse_args()
summary = evaluate(args.dataset, args.predictions, args.strict)
text = json.dumps(summary, indent=2, sort_keys=True)
if args.output:
args.output.parent.mkdir(parents=True, exist_ok=True)
args.output.write_text(text + "\n", encoding="utf-8")
print(text)
if __name__ == "__main__":
main()
|