Datasets:
Tasks:
Document Question Answering
Size:
n<1K
Tags:
benchmark
document-ai
information-extraction
structured-extraction
key-information-extraction
ocr
License:
File size: 11,518 Bytes
ca66b51 | 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 | """Command-line utilities for validating and scoring DocuBench."""
from __future__ import annotations
import argparse
import csv
import json
import sys
from pathlib import Path
from typing import Any
from scorer import score_standardization
DEFAULT_ENGINE_DISPLAY_NAMES = {
"claude": "Claude",
"docupipe_high": "DocuPipe high",
"docupipe_standard": "DocuPipe standard",
"extend": "Extend",
"gemini": "Gemini",
"gpt": "GPT",
}
def load_json(path: Path) -> Any:
with open(path, encoding="utf-8") as f:
return json.load(f)
def write_json(path: Path, data: Any) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
with open(path, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=2)
f.write("\n")
def benchmark_ids(root: Path) -> list[str]:
return sorted(p.stem for p in (root / "labels").glob("*.json"))
def discover_engines(root: Path) -> list[str]:
results_dir = root / "results"
if not results_dir.exists():
return []
return sorted(p.name for p in results_dir.iterdir() if p.is_dir())
def document_id_map(root: Path) -> dict[str, Path]:
docs: dict[str, Path] = {}
for path in (root / "documents").iterdir():
if path.is_file():
docs[path.stem] = path
return docs
def validate_benchmark(root: Path) -> tuple[list[str], list[str], dict[str, Any]]:
errors: list[str] = []
warnings: list[str] = []
required_dirs = ["documents", "labels", "schemas", "results"]
for dirname in required_dirs:
if not (root / dirname).is_dir():
errors.append(f"missing required directory: {dirname}/")
sources_path = root / "sources.json"
if not sources_path.is_file():
errors.append("missing sources.json")
sources = []
else:
try:
sources = load_json(sources_path)
except json.JSONDecodeError as exc:
errors.append(f"sources.json is invalid JSON: {exc}")
sources = []
if errors:
return errors, warnings, {}
doc_paths = document_id_map(root)
doc_ids = set(doc_paths)
label_ids = {p.stem for p in (root / "labels").glob("*.json")}
schema_ids = {p.stem for p in (root / "schemas").glob("*.json")}
source_ids = {row.get("doc_id") for row in sources if isinstance(row, dict)}
expected_ids = label_ids | schema_ids | source_ids
for collection_name, ids in [
("documents", doc_ids),
("labels", label_ids),
("schemas", schema_ids),
("sources", source_ids),
]:
missing = sorted(expected_ids - ids)
extra = sorted(ids - expected_ids)
if missing:
errors.append(f"{collection_name} missing ids: {', '.join(missing)}")
if extra:
errors.append(f"{collection_name} has unexpected ids: {', '.join(extra)}")
for directory in ["labels", "schemas"]:
for path in sorted((root / directory).glob("*.json")):
try:
load_json(path)
except json.JSONDecodeError as exc:
errors.append(f"{path.relative_to(root)} is invalid JSON: {exc}")
engines = discover_engines(root)
for engine in engines:
engine_dir = root / "results" / engine
result_ids = {p.stem for p in engine_dir.glob("*.json")}
missing = sorted(label_ids - result_ids)
extra = sorted(result_ids - label_ids)
if missing:
warnings.append(f"results/{engine} missing ids: {', '.join(missing)}")
if extra:
warnings.append(f"results/{engine} has extra ids: {', '.join(extra)}")
for path in sorted(engine_dir.glob("*.json")):
try:
payload = load_json(path)
except json.JSONDecodeError as exc:
errors.append(f"{path.relative_to(root)} is invalid JSON: {exc}")
continue
if not isinstance(payload, dict):
errors.append(f"{path.relative_to(root)} must contain a JSON object")
elif "data" not in payload:
warnings.append(f"{path.relative_to(root)} has no top-level data key")
summary = {
"documents": len(doc_ids),
"labels": len(label_ids),
"schemas": len(schema_ids),
"sources": len(source_ids),
"engines": engines,
}
return errors, warnings, summary
def score_engines(root: Path, engines: list[str] | None = None) -> dict[str, Any]:
doc_ids = benchmark_ids(root)
selected_engines = engines or discover_engines(root)
per_engine: dict[str, dict[str, float | None]] = {engine: {} for engine in selected_engines}
for doc_id in doc_ids:
schema = load_json(root / "schemas" / f"{doc_id}.json")
label = load_json(root / "labels" / f"{doc_id}.json")
for engine in selected_engines:
result_path = root / "results" / engine / f"{doc_id}.json"
if not result_path.exists():
per_engine[engine][doc_id] = None
continue
result_payload = load_json(result_path)
result = result_payload.get("data", {}) if isinstance(result_payload, dict) else {}
score = score_standardization(result=result, schema=schema, label=label)
per_engine[engine][doc_id] = score["final"]
per_doc = []
for doc_id in doc_ids:
row: dict[str, Any] = {"doc_id": doc_id}
for engine in selected_engines:
row[engine] = per_engine[engine][doc_id]
per_doc.append(row)
aggregates = {}
for engine, scores_by_doc in per_engine.items():
scores = [score for score in scores_by_doc.values() if score is not None]
aggregates[engine] = sum(scores) / len(scores) if scores else None
return {
"benchmark": {
"name": "DocuBench",
"version": "0.1.0",
"doc_count": len(doc_ids),
"metric": "macro_average_field_accuracy",
},
"engine_display_names": {
engine: DEFAULT_ENGINE_DISPLAY_NAMES.get(engine, engine)
for engine in selected_engines
},
"aggregates": aggregates,
"breakdowns": build_breakdowns(root, per_doc, selected_engines),
"per_doc": per_doc,
}
def build_breakdowns(root: Path, per_doc: list[dict[str, Any]], engines: list[str]) -> dict[str, Any]:
sources_path = root / "sources.json"
if not sources_path.exists():
return {}
sources = load_json(sources_path)
metadata = {row["doc_id"]: row for row in sources if isinstance(row, dict) and "doc_id" in row}
by_doc = {row["doc_id"]: row for row in per_doc}
breakdowns: dict[str, Any] = {}
for dimension in ["ftype", "lang"]:
groups: dict[str, list[str]] = {}
for doc_id, row in metadata.items():
value = str(row.get(dimension) or "unknown")
groups.setdefault(value, []).append(doc_id)
dimension_rows = []
for value in sorted(groups):
doc_ids = sorted(doc_id for doc_id in groups[value] if doc_id in by_doc)
out: dict[str, Any] = {"value": value, "doc_count": len(doc_ids)}
for engine in engines:
scores = [by_doc[doc_id][engine] for doc_id in doc_ids if by_doc[doc_id][engine] is not None]
out[engine] = sum(scores) / len(scores) if scores else None
dimension_rows.append(out)
breakdowns[dimension] = dimension_rows
return breakdowns
def print_score_table(scores: dict[str, Any]) -> None:
engines = list(scores["aggregates"].keys())
header = f"{'doc_id':<12}" + "".join(f"{engine:>20}" for engine in engines)
print(header)
print("-" * len(header))
for row in scores["per_doc"]:
line = f"{row['doc_id']:<12}"
for engine in engines:
score = row[engine]
line += f"{(f'{score:.4f}' if score is not None else 'n/a'):>20}"
print(line)
print("-" * len(header))
aggregate_line = f"{'AGGREGATE':<12}"
for engine in engines:
score = scores["aggregates"][engine]
aggregate_line += f"{(f'{score:.4f}' if score is not None else 'n/a'):>20}"
print(aggregate_line)
def write_summary_csv(path: Path, scores: dict[str, Any]) -> None:
engines = list(scores["aggregates"].keys())
path.parent.mkdir(parents=True, exist_ok=True)
with open(path, "w", encoding="utf-8", newline="") as f:
writer = csv.DictWriter(f, fieldnames=["doc_id", *engines], lineterminator="\n")
writer.writeheader()
for row in scores["per_doc"]:
writer.writerow({key: row.get(key) for key in ["doc_id", *engines]})
writer.writerow({"doc_id": "AGGREGATE", **scores["aggregates"]})
def cmd_validate(args: argparse.Namespace) -> int:
errors, warnings, summary = validate_benchmark(args.root)
for warning in warnings:
print(f"warning: {warning}", file=sys.stderr)
if errors:
for error in errors:
print(f"error: {error}", file=sys.stderr)
return 1
print(
"validated "
f"{summary['documents']} documents, "
f"{summary['labels']} labels, "
f"{summary['schemas']} schemas, "
f"{summary['sources']} source records, "
f"{len(summary['engines'])} result sets"
)
return 0
def cmd_score(args: argparse.Namespace) -> int:
scores = score_engines(args.root, args.engine)
if args.json:
print(json.dumps(scores, ensure_ascii=False, indent=2))
else:
print_score_table(scores)
return 0
def cmd_report(args: argparse.Namespace) -> int:
scores = score_engines(args.root, args.engine)
write_json(args.summary_json, scores)
write_summary_csv(args.summary_csv, scores)
print(f"wrote {args.summary_json}")
print(f"wrote {args.summary_csv}")
return 0
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(prog="docubench")
parser.add_argument("--root", type=Path, default=Path.cwd(), help="repository root")
subparsers = parser.add_subparsers(dest="command", required=True)
validate = subparsers.add_parser("validate", help="validate benchmark files")
validate.set_defaults(func=cmd_validate)
score = subparsers.add_parser("score", help="score committed result sets")
score.add_argument("--engine", action="append", help="result directory to score; repeatable")
score.add_argument("--json", action="store_true", help="emit JSON instead of a table")
score.set_defaults(func=cmd_score)
report = subparsers.add_parser("report", help="write summary JSON and CSV reports")
report.add_argument("--engine", action="append", help="result directory to score; repeatable")
report.add_argument("--summary-json", type=Path, default=Path("results/summary.json"))
report.add_argument("--summary-csv", type=Path, default=Path("results/summary.csv"))
report.set_defaults(func=cmd_report)
return parser
def main(argv: list[str] | None = None) -> int:
parser = build_parser()
args = parser.parse_args(argv)
args.root = args.root.resolve()
if hasattr(args, "summary_json") and not args.summary_json.is_absolute():
args.summary_json = args.root / args.summary_json
if hasattr(args, "summary_csv") and not args.summary_csv.is_absolute():
args.summary_csv = args.root / args.summary_csv
return args.func(args)
if __name__ == "__main__":
raise SystemExit(main())
|