File size: 12,148 Bytes
dcc24f8 |
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 |
"""
Model Evaluation Module for Phase 1 Completion.
Evaluates the fine-tuned model on the held-out test set,
computing per-bank accuracy and generating a completion report.
Author: Ranjit Behera
"""
import json
import time
from pathlib import Path
from typing import Dict, List, Optional, Tuple
from dataclasses import dataclass, field
from collections import defaultdict
from datetime import datetime
@dataclass
class EvaluationResult:
"""Result for a single evaluation sample."""
bank: str
expected: Dict
predicted: Dict
raw_response: str
correct_fields: Dict[str, bool] = field(default_factory=dict)
overall_correct: bool = False
inference_time_ms: float = 0.0
def compute_accuracy(self):
"""Compute per-field accuracy."""
for key in ["amount", "type", "date", "reference", "merchant", "category"]:
expected_val = str(self.expected.get(key, "")).lower().strip()
predicted_val = str(self.predicted.get(key, "")).lower().strip()
# Normalize amount (remove commas, trailing zeros)
if key == "amount":
try:
expected_val = str(float(expected_val.replace(",", "")))
predicted_val = str(float(predicted_val.replace(",", "")))
except:
pass
self.correct_fields[key] = expected_val == predicted_val
# Core fields that must match
core_fields = ["amount", "type"]
self.overall_correct = all(
self.correct_fields.get(f, False) for f in core_fields
)
@dataclass
class BankMetrics:
"""Metrics for a single bank."""
bank: str
total: int = 0
correct: int = 0
field_correct: Dict[str, int] = field(default_factory=lambda: defaultdict(int))
field_total: Dict[str, int] = field(default_factory=lambda: defaultdict(int))
avg_inference_time_ms: float = 0.0
@property
def accuracy(self) -> float:
return self.correct / self.total if self.total > 0 else 0.0
def field_accuracy(self, field: str) -> float:
total = self.field_total[field]
return self.field_correct[field] / total if total > 0 else 0.0
class ModelEvaluator:
"""
Evaluates the fine-tuned model on test data.
Generates per-bank accuracy metrics and a phase completion report.
"""
def __init__(
self,
model_path: str = "models/base/phi3-mini",
adapter_path: str = "models/adapters/finance-lora-v3",
test_file: str = "data/synthetic/test_emails.json",
project_root: Optional[Path] = None
):
self.project_root = project_root or Path.cwd()
self.model_path = self.project_root / model_path
self.adapter_path = self.project_root / adapter_path
self.test_file = self.project_root / test_file
self.predictor = None
self.results: List[EvaluationResult] = []
self.bank_metrics: Dict[str, BankMetrics] = {}
def load_model(self):
"""Load the model with adapter."""
from src.inference.predict import Predictor
print(f"π Loading model from {self.model_path}")
print(f" With adapter: {self.adapter_path}")
self.predictor = Predictor(
model_path=self.model_path,
adapter_path=self.adapter_path,
max_tokens=250
)
self.predictor.load()
def load_test_data(self) -> List[Dict]:
"""Load test samples."""
with open(self.test_file) as f:
data = json.load(f)
print(f"π Loaded {len(data)} test samples")
return data
def evaluate(self, limit: Optional[int] = None) -> Dict:
"""
Run evaluation on the test set.
Args:
limit: Optional limit on number of samples to evaluate
Returns:
Summary metrics dictionary
"""
if self.predictor is None:
self.load_model()
test_data = self.load_test_data()
if limit:
test_data = test_data[:limit]
print(f"\nπ§ͺ Running evaluation on {len(test_data)} samples...")
print("=" * 60)
for i, sample in enumerate(test_data):
bank = sample.get("bank", "unknown")
subject = sample.get("subject", "")
body = sample.get("body", sample.get("raw_text", ""))
expected = sample.get("entities", {})
# Run inference
start_time = time.time()
result = self.predictor.predict(subject=subject, body=body)
inference_time = (time.time() - start_time) * 1000
# Create evaluation result
eval_result = EvaluationResult(
bank=bank,
expected=expected,
predicted=result.entities,
raw_response=result.raw_response,
inference_time_ms=inference_time
)
eval_result.compute_accuracy()
self.results.append(eval_result)
# Progress indicator
status = "β" if eval_result.overall_correct else "β"
if (i + 1) % 10 == 0 or i == len(test_data) - 1:
print(f" Processed {i + 1}/{len(test_data)} samples")
# Compute bank-level metrics
self._compute_bank_metrics()
return self.get_summary()
def _compute_bank_metrics(self):
"""Compute per-bank metrics."""
bank_results = defaultdict(list)
for result in self.results:
bank_results[result.bank].append(result)
for bank, results in bank_results.items():
metrics = BankMetrics(bank=bank)
metrics.total = len(results)
metrics.correct = sum(1 for r in results if r.overall_correct)
total_time = 0
for result in results:
total_time += result.inference_time_ms
for field, correct in result.correct_fields.items():
metrics.field_total[field] += 1
if correct:
metrics.field_correct[field] += 1
metrics.avg_inference_time_ms = total_time / len(results)
self.bank_metrics[bank] = metrics
def get_summary(self) -> Dict:
"""Get evaluation summary."""
total = len(self.results)
correct = sum(1 for r in self.results if r.overall_correct)
# Field-level accuracy
field_accuracy = {}
for field in ["amount", "type", "date", "reference", "merchant", "category"]:
field_correct = sum(
1 for r in self.results
if r.correct_fields.get(field, False)
)
field_accuracy[field] = field_correct / total if total > 0 else 0.0
return {
"total_samples": total,
"correct_samples": correct,
"overall_accuracy": correct / total if total > 0 else 0.0,
"field_accuracy": field_accuracy,
"per_bank": {
bank: {
"total": m.total,
"correct": m.correct,
"accuracy": m.accuracy,
"avg_inference_ms": m.avg_inference_time_ms
}
for bank, m in self.bank_metrics.items()
}
}
def print_report(self):
"""Print a formatted evaluation report."""
summary = self.get_summary()
print("\n")
print("β" + "β" * 62 + "β")
print("β" + " π PHASE 1 EVALUATION REPORT".center(62) + "β")
print("β " + "β" * 62 + "β£")
# Overall metrics
acc_pct = summary["overall_accuracy"] * 100
print(f"β Overall Accuracy: {acc_pct:>6.1f}% ({summary['correct_samples']}/{summary['total_samples']})".ljust(63) + "β")
print("β " + "β" * 62 + "β£")
# Per-bank accuracy
print("β" + " π¦ Per-Bank Accuracy".center(62) + "β")
print("β " + "β" * 62 + "β£")
print("β " + f"{'Bank':<12} {'Samples':>8} {'Correct':>8} {'Accuracy':>10} {'Avg Time':>12}" + " β")
print("β " + "β" * 62 + "β£")
for bank in sorted(summary["per_bank"].keys()):
data = summary["per_bank"][bank]
acc = data["accuracy"] * 100
time_ms = data["avg_inference_ms"]
emoji = "β
" if acc >= 80 else "β οΈ" if acc >= 60 else "β"
print(f"β {emoji} {bank.upper():<10} {data['total']:>6} {data['correct']:>8} {acc:>9.1f}% {time_ms:>10.0f}ms" + " β")
print("β " + "β" * 62 + "β£")
# Per-field accuracy
print("β" + " π― Per-Field Accuracy".center(62) + "β")
print("β " + "β" * 62 + "β£")
for field, acc in sorted(summary["field_accuracy"].items(), key=lambda x: -x[1]):
acc_pct = acc * 100
bar_len = int(acc * 25)
bar = "β" * bar_len + "β" * (25 - bar_len)
emoji = "β
" if acc >= 0.8 else "β οΈ" if acc >= 0.6 else "β"
print(f"β {emoji} {field:<12} {bar} {acc_pct:>5.1f}%" + " β")
print("β" + "β" * 62 + "β")
# Phase 1 completion status
print("\n")
overall_acc = summary["overall_accuracy"]
if overall_acc >= 0.85:
print("π PHASE 1 COMPLETE! Model achieves >85% accuracy on all banks.")
elif overall_acc >= 0.75:
print("β
Phase 1 nearly complete. Consider more training for edge cases.")
else:
print("β οΈ Phase 1 needs more work. Review failed samples and retrain.")
def save_report(self, output_path: Optional[Path] = None):
"""Save evaluation report to JSON."""
if output_path is None:
output_path = self.project_root / "data/evaluation_report.json"
report = {
"timestamp": datetime.now().isoformat(),
"model_path": str(self.model_path),
"adapter_path": str(self.adapter_path),
"test_file": str(self.test_file),
"summary": self.get_summary(),
"detailed_results": [
{
"bank": r.bank,
"expected": r.expected,
"predicted": r.predicted,
"correct_fields": r.correct_fields,
"overall_correct": r.overall_correct,
"inference_time_ms": r.inference_time_ms
}
for r in self.results
]
}
with open(output_path, "w") as f:
json.dump(report, f, indent=2)
print(f"\nπΎ Report saved to {output_path}")
def get_failed_samples(self) -> List[EvaluationResult]:
"""Get samples where prediction was incorrect."""
return [r for r in self.results if not r.overall_correct]
def main():
"""Run evaluation and generate report."""
import argparse
parser = argparse.ArgumentParser(description="Evaluate fine-tuned model")
parser.add_argument("--model", default="models/base/phi3-mini")
parser.add_argument("--adapter", default="models/adapters/finance-lora-v3")
parser.add_argument("--test-file", default="data/synthetic/test_emails.json")
parser.add_argument("--limit", type=int, default=None, help="Limit samples")
parser.add_argument("--save", action="store_true", help="Save report to JSON")
args = parser.parse_args()
evaluator = ModelEvaluator(
model_path=args.model,
adapter_path=args.adapter,
test_file=args.test_file
)
evaluator.evaluate(limit=args.limit)
evaluator.print_report()
if args.save:
evaluator.save_report()
if __name__ == "__main__":
main()
|