File size: 14,033 Bytes
1cba4da |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 |
#!/usr/bin/env python3
"""
Production Benchmark Suite for FinEE
=====================================
Comprehensive evaluation with:
- Precision/Recall/F1 per field
- Bank-specific performance
- Cross-validation
- Failure case analysis
- Comparison with baselines
Author: Ranjit Behera
"""
import json
import random
from pathlib import Path
from typing import List, Dict, Tuple, Optional
from dataclasses import dataclass, field
from collections import defaultdict
import time
@dataclass
class FieldMetrics:
"""Metrics for a single field."""
tp: int = 0 # True positives
fp: int = 0 # False positives
fn: int = 0 # False negatives
@property
def precision(self) -> float:
if self.tp + self.fp == 0:
return 0.0
return self.tp / (self.tp + self.fp)
@property
def recall(self) -> float:
if self.tp + self.fn == 0:
return 0.0
return self.tp / (self.tp + self.fn)
@property
def f1(self) -> float:
if self.precision + self.recall == 0:
return 0.0
return 2 * (self.precision * self.recall) / (self.precision + self.recall)
@dataclass
class BenchmarkResult:
"""Complete benchmark results."""
field_metrics: Dict[str, FieldMetrics] = field(default_factory=dict)
bank_metrics: Dict[str, Dict[str, FieldMetrics]] = field(default_factory=dict)
failures: List[Dict] = field(default_factory=list)
latency_ms: List[float] = field(default_factory=list)
total_samples: int = 0
@property
def overall_f1(self) -> float:
if not self.field_metrics:
return 0.0
return sum(m.f1 for m in self.field_metrics.values()) / len(self.field_metrics)
@property
def avg_latency_ms(self) -> float:
if not self.latency_ms:
return 0.0
return sum(self.latency_ms) / len(self.latency_ms)
class ProductionBenchmark:
"""
Production-grade benchmark for financial entity extraction.
"""
FIELDS = ["amount", "type", "bank", "merchant", "category", "reference", "vpa"]
def __init__(self, test_data_path: Optional[Path] = None):
self.test_data_path = test_data_path
self.extractor = None
self.results = BenchmarkResult()
def load_extractor(self, use_llm: bool = False):
"""Load the extractor."""
try:
from finee import FinancialExtractor
self.extractor = FinancialExtractor(use_llm=use_llm)
except ImportError:
from finee import extract
self.extractor = type('Extractor', (), {'extract': lambda self, x: extract(x)})()
def load_test_data(self, path: Optional[Path] = None) -> List[Dict]:
"""Load test dataset."""
path = path or self.test_data_path
if path and path.exists():
records = []
with open(path) as f:
for line in f:
try:
records.append(json.loads(line))
except:
continue
return records
return []
def _normalize_value(self, value, field: str):
"""Normalize values for comparison."""
if value is None:
return None
if field == "amount":
if isinstance(value, (int, float)):
return round(float(value), 2)
if isinstance(value, str):
try:
return round(float(value.replace(',', '')), 2)
except:
return None
if field == "type":
v = str(value).lower().strip()
if v in ["debit", "dr", "debited"]:
return "debit"
if v in ["credit", "cr", "credited"]:
return "credit"
return v
if isinstance(value, str):
return value.lower().strip()
return value
def _compare_values(self, predicted, expected, field: str) -> Tuple[bool, str]:
"""Compare predicted vs expected values."""
pred_norm = self._normalize_value(predicted, field)
exp_norm = self._normalize_value(expected, field)
if pred_norm is None and exp_norm is None:
return True, "both_none"
if pred_norm is None and exp_norm is not None:
return False, "false_negative"
if pred_norm is not None and exp_norm is None:
return False, "false_positive"
if pred_norm == exp_norm:
return True, "true_positive"
# Partial match for strings
if field in ["merchant", "bank"]:
if str(pred_norm) in str(exp_norm) or str(exp_norm) in str(pred_norm):
return True, "partial_match"
return False, "mismatch"
def evaluate_single(self, text: str, expected: Dict) -> Tuple[Dict, Dict, List[str]]:
"""
Evaluate a single example.
Returns:
(predicted, expected, error_fields)
"""
start = time.perf_counter()
# Extract
if hasattr(self.extractor, 'extract'):
predicted = self.extractor.extract(text)
else:
predicted = self.extractor(text)
# Convert to dict if needed
if hasattr(predicted, 'to_dict'):
predicted = predicted.to_dict()
elif hasattr(predicted, '__dict__'):
predicted = {k: v for k, v in predicted.__dict__.items() if not k.startswith('_')}
latency = (time.perf_counter() - start) * 1000
self.results.latency_ms.append(latency)
# Compare each field
errors = []
for field in self.FIELDS:
pred_val = predicted.get(field)
exp_val = expected.get(field)
match, reason = self._compare_values(pred_val, exp_val, field)
if field not in self.results.field_metrics:
self.results.field_metrics[field] = FieldMetrics()
metrics = self.results.field_metrics[field]
if reason == "true_positive" or reason == "partial_match":
metrics.tp += 1
elif reason == "false_negative":
metrics.fn += 1
errors.append(f"{field}: expected '{exp_val}', got None")
elif reason == "false_positive":
metrics.fp += 1
errors.append(f"{field}: expected None, got '{pred_val}'")
elif reason == "mismatch":
metrics.fn += 1
metrics.fp += 1
errors.append(f"{field}: expected '{exp_val}', got '{pred_val}'")
return predicted, expected, errors
def run(
self,
test_data: Optional[List[Dict]] = None,
max_samples: int = 1000,
include_failures: bool = True
) -> BenchmarkResult:
"""
Run the full benchmark.
Args:
test_data: List of test samples
max_samples: Maximum samples to evaluate
include_failures: Whether to collect failure cases
Returns:
BenchmarkResult with all metrics
"""
if self.extractor is None:
self.load_extractor()
if test_data is None:
test_data = self.load_test_data()
if not test_data:
print("⚠️ No test data provided")
return self.results
# Sample if too many
if len(test_data) > max_samples:
test_data = random.sample(test_data, max_samples)
self.results = BenchmarkResult()
self.results.total_samples = len(test_data)
print(f"Running benchmark on {len(test_data)} samples...")
for i, record in enumerate(test_data):
text = record.get("input", record.get("text", ""))
expected = record.get("output", record.get("ground_truth", {}))
if isinstance(expected, str):
try:
expected = json.loads(expected)
except:
continue
predicted, expected, errors = self.evaluate_single(text, expected)
# Track failures
if include_failures and errors:
self.results.failures.append({
"text": text[:100],
"expected": expected,
"predicted": predicted,
"errors": errors,
})
# Progress
if (i + 1) % 100 == 0:
print(f" Processed {i + 1}/{len(test_data)}...")
return self.results
def print_report(self):
"""Print a detailed report."""
print("\n" + "=" * 70)
print("PRODUCTION BENCHMARK REPORT")
print("=" * 70)
print(f"\n📊 Overall Statistics:")
print(f" Total Samples: {self.results.total_samples:,}")
print(f" Overall F1: {self.results.overall_f1:.1%}")
print(f" Avg Latency: {self.results.avg_latency_ms:.2f}ms")
print(f"\n📈 Per-Field Metrics:")
print(f" {'Field':<12} {'Precision':>10} {'Recall':>10} {'F1':>10}")
print(" " + "-" * 42)
for field in self.FIELDS:
if field in self.results.field_metrics:
m = self.results.field_metrics[field]
status = "✅" if m.f1 >= 0.90 else "⚠️" if m.f1 >= 0.70 else "❌"
print(f" {field:<12} {m.precision:>9.1%} {m.recall:>9.1%} {m.f1:>9.1%} {status}")
print(f"\n❌ Failure Cases: {len(self.results.failures)}")
if self.results.failures:
print("\n Sample Failures:")
for failure in self.results.failures[:5]:
print(f"\n Text: {failure['text'][:60]}...")
for err in failure['errors'][:3]:
print(f" • {err}")
# Grade
f1 = self.results.overall_f1
if f1 >= 0.95:
grade = "A+ (Production Ready)"
elif f1 >= 0.90:
grade = "A (Near Production)"
elif f1 >= 0.80:
grade = "B (Good)"
elif f1 >= 0.70:
grade = "C (Needs Improvement)"
else:
grade = "D (Significant Work Needed)"
print(f"\n🏆 Grade: {grade}")
print("=" * 70)
def export_results(self, path: Path):
"""Export results to JSON."""
data = {
"overall_f1": self.results.overall_f1,
"avg_latency_ms": self.results.avg_latency_ms,
"total_samples": self.results.total_samples,
"field_metrics": {
field: {
"precision": m.precision,
"recall": m.recall,
"f1": m.f1,
}
for field, m in self.results.field_metrics.items()
},
"failure_count": len(self.results.failures),
"failures": self.results.failures[:20],
}
with open(path, 'w') as f:
json.dump(data, f, indent=2)
print(f"Results exported to {path}")
def create_held_out_test_set(
data_path: Path,
output_path: Path,
held_out_banks: List[str] = ["PNB", "BOB", "CANARA"],
num_samples: int = 1000
):
"""
Create a held-out test set with banks NOT in training.
This is critical for proper evaluation.
"""
print(f"Creating held-out test set with banks: {held_out_banks}")
held_out = []
with open(data_path) as f:
for line in f:
try:
record = json.loads(line)
text = record.get("input", record.get("text", "")).upper()
# Check if contains held-out bank
for bank in held_out_banks:
if bank in text:
held_out.append(record)
break
if len(held_out) >= num_samples:
break
except:
continue
# Save
output_path.parent.mkdir(parents=True, exist_ok=True)
with open(output_path, 'w') as f:
for record in held_out:
f.write(json.dumps(record) + '\n')
print(f"Created held-out test set with {len(held_out)} samples at {output_path}")
return held_out
# ============================================================================
# MAIN
# ============================================================================
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description="Run production benchmark")
parser.add_argument("--test-file", help="Path to test JSONL file")
parser.add_argument("--max-samples", type=int, default=1000)
parser.add_argument("--export", help="Export results to JSON")
parser.add_argument("--create-held-out", action="store_true",
help="Create held-out test set")
args = parser.parse_args()
if args.create_held_out:
create_held_out_test_set(
Path("data/instruction/test.jsonl"),
Path("data/benchmark/held_out_test.jsonl"),
)
else:
benchmark = ProductionBenchmark()
if args.test_file:
test_data = benchmark.load_test_data(Path(args.test_file))
else:
# Use default test set
test_data = benchmark.load_test_data(Path("data/instruction/test.jsonl"))
benchmark.run(test_data, max_samples=args.max_samples)
benchmark.print_report()
if args.export:
benchmark.export_results(Path(args.export))
|