Spaces:
Sleeping
Sleeping
File size: 11,783 Bytes
f9b1ad5 |
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 |
#!/usr/bin/env python3
"""
Post-Process Benchmark Data
============================
Strategy:
1. Load raw benchmark results
2. Stratify by difficulty tier (low/medium/high success)
3. Select balanced sample for vector DB:
- 30% LOW success (0-30%): Hard questions - model limitations
- 40% MEDIUM success (30-70%): Capability boundary - most interesting
- 30% HIGH success (70-100%): Within capability - baseline
4. Export stratified sample for vector DB indexing
This ensures we have good coverage across the capability spectrum.
"""
import json
import logging
from pathlib import Path
from typing import Dict, List, Any
from collections import defaultdict
import random
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
class BenchmarkDataPostProcessor:
"""Post-process raw benchmark data for vector DB"""
def __init__(self, input_file: Path = Path("./data/benchmark_results/raw_benchmark_results.json")):
self.input_file = input_file
self.questions = {}
self.stratified_sample = {}
def load_raw_data(self):
"""Load raw benchmark results"""
logger.info(f"Loading raw data from {self.input_file}...")
with open(self.input_file, 'r') as f:
data = json.load(f)
self.questions = data['questions']
logger.info(f"Loaded {len(self.questions)} questions")
return self.questions
def analyze_difficulty_distribution(self) -> Dict[str, Any]:
"""Analyze distribution across difficulty tiers"""
logger.info("Analyzing difficulty distribution...")
distribution = {
"low": [], # 0-30% success
"medium": [], # 30-70% success
"high": [] # 70-100% success
}
for qid, q in self.questions.items():
tier = q.get('difficulty_tier')
if tier and tier in distribution:
distribution[tier].append(qid)
stats = {
"total_questions": len(self.questions),
"low_success_count": len(distribution["low"]),
"medium_success_count": len(distribution["medium"]),
"high_success_count": len(distribution["high"]),
"low_success_pct": len(distribution["low"]) / len(self.questions) * 100,
"medium_success_pct": len(distribution["medium"]) / len(self.questions) * 100,
"high_success_pct": len(distribution["high"]) / len(self.questions) * 100
}
logger.info(f" LOW success (0-30%): {stats['low_success_count']} ({stats['low_success_pct']:.1f}%)")
logger.info(f" MEDIUM success (30-70%): {stats['medium_success_count']} ({stats['medium_success_pct']:.1f}%)")
logger.info(f" HIGH success (70-100%): {stats['high_success_count']} ({stats['high_success_pct']:.1f}%)")
return distribution, stats
def stratified_sampling(
self,
target_size: int = 1000,
low_pct: float = 0.30,
medium_pct: float = 0.40,
high_pct: float = 0.30
) -> Dict[str, Any]:
"""
Create stratified sample with balanced difficulty distribution.
Args:
target_size: Total number of questions to sample
low_pct: Percentage of LOW success questions (0-30% success)
medium_pct: Percentage of MEDIUM success questions (30-70%)
high_pct: Percentage of HIGH success questions (70-100%)
"""
logger.info(f"Creating stratified sample (target: {target_size} questions)...")
logger.info(f" Target distribution: {low_pct*100:.0f}% low, {medium_pct*100:.0f}% medium, {high_pct*100:.0f}% high")
distribution, _ = self.analyze_difficulty_distribution()
# Calculate target counts per tier
target_counts = {
"low": int(target_size * low_pct),
"medium": int(target_size * medium_pct),
"high": int(target_size * high_pct)
}
sampled = {}
random.seed(42) # Reproducibility
for tier, target_count in target_counts.items():
available = distribution[tier]
if len(available) >= target_count:
# Sample from available
selected = random.sample(available, target_count)
else:
# Take all available
selected = available
logger.warning(f" Only {len(available)} {tier} questions available (target: {target_count})")
for qid in selected:
sampled[qid] = self.questions[qid]
logger.info(f" Sampled {len(selected)} {tier} success questions")
self.stratified_sample = sampled
logger.info(f"Total sampled: {len(sampled)} questions")
return sampled
def export_for_vector_db(self, output_file: Path = Path("./data/benchmark_results/stratified_sample.json")):
"""Export stratified sample in format ready for vector DB"""
logger.info(f"Exporting stratified sample to {output_file}...")
# Create output format
export_data = {
"metadata": {
"total_questions": len(self.stratified_sample),
"sampling_strategy": "stratified_by_difficulty",
"tiers": {
"low": "0-30% success rate",
"medium": "30-70% success rate",
"high": "70-100% success rate"
}
},
"questions": []
}
# Group by tier for summary
tier_counts = defaultdict(int)
benchmark_counts = defaultdict(int)
for qid, q in self.stratified_sample.items():
tier_counts[q.get('difficulty_tier', 'unknown')] += 1
benchmark_counts[q.get('source_benchmark', 'unknown')] += 1
# Simplify for export
export_q = {
"question_id": qid,
"source_benchmark": q['source_benchmark'],
"domain": q['domain'],
"question_text": q['question_text'],
"correct_answer": q['correct_answer'],
"choices": q.get('choices'),
"success_rate": q.get('success_rate'),
"difficulty_tier": q.get('difficulty_tier'),
"difficulty_label": q.get('difficulty_label'),
"num_models_tested": q.get('num_models', 0)
}
export_data["questions"].append(export_q)
export_data["metadata"]["distribution"] = {
"by_tier": dict(tier_counts),
"by_benchmark": dict(benchmark_counts)
}
# Save
output_file.parent.mkdir(parents=True, exist_ok=True)
with open(output_file, 'w') as f:
json.dump(export_data, f, indent=2)
logger.info(f"✓ Exported {len(export_data['questions'])} questions")
logger.info(f" By tier: {dict(tier_counts)}")
logger.info(f" By benchmark: {dict(benchmark_counts)}")
return output_file
def generate_summary_report(self) -> str:
"""Generate markdown summary report"""
report = ["# Benchmark Data Post-Processing Report\n"]
# Overall stats
report.append("## Overall Statistics\n")
report.append(f"- **Total questions collected**: {len(self.questions)}")
report.append(f"- **Stratified sample size**: {len(self.stratified_sample)}\n")
# Difficulty distribution
report.append("## Difficulty Distribution\n")
tier_counts = defaultdict(int)
for q in self.stratified_sample.values():
tier_counts[q.get('difficulty_tier', 'unknown')] += 1
report.append("| Tier | Count | Percentage | Description |")
report.append("|------|-------|------------|-------------|")
total = len(self.stratified_sample)
for tier in ['low', 'medium', 'high']:
count = tier_counts[tier]
pct = count / total * 100 if total > 0 else 0
desc = {
'low': 'Hard - model limitations (0-30% success)',
'medium': 'Capability boundary (30-70% success)',
'high': 'Within capability (70-100% success)'
}[tier]
report.append(f"| {tier.upper()} | {count} | {pct:.1f}% | {desc} |")
report.append("\n")
# Benchmark distribution
report.append("## Source Benchmark Distribution\n")
benchmark_counts = defaultdict(int)
for q in self.stratified_sample.values():
benchmark_counts[q.get('source_benchmark', 'unknown')] += 1
report.append("| Benchmark | Count | Percentage |")
report.append("|-----------|-------|------------|")
for benchmark, count in sorted(benchmark_counts.items()):
pct = count / total * 100 if total > 0 else 0
report.append(f"| {benchmark} | {count} | {pct:.1f}% |")
report.append("\n")
# Success rate stats
report.append("## Success Rate Statistics\n")
success_rates = [q.get('success_rate', 0) for q in self.stratified_sample.values() if q.get('success_rate') is not None]
if success_rates:
import numpy as np
report.append(f"- **Min**: {np.min(success_rates):.1%}")
report.append(f"- **Max**: {np.max(success_rates):.1%}")
report.append(f"- **Mean**: {np.mean(success_rates):.1%}")
report.append(f"- **Median**: {np.median(success_rates):.1%}\n")
# Next steps
report.append("## Next Steps\n")
report.append("1. Load stratified sample into vector database")
report.append("2. Generate embeddings for all questions")
report.append("3. Test difficulty assessment on real prompts")
report.append("4. Validate accuracy against known hard/easy questions\n")
return "\n".join(report)
def save_summary_report(self, output_file: Path = Path("./data/benchmark_results/PROCESSING_REPORT.md")):
"""Save summary report"""
report = self.generate_summary_report()
with open(output_file, 'w') as f:
f.write(report)
logger.info(f"Saved summary report to {output_file}")
return output_file
def main():
"""Main execution"""
logger.info("="*80)
logger.info("Post-Processing Benchmark Data")
logger.info("="*80)
# Initialize
processor = BenchmarkDataPostProcessor()
# Load raw data
processor.load_raw_data()
# Analyze distribution
processor.analyze_difficulty_distribution()
# Create stratified sample
# Target: 1000 questions with 30% low, 40% medium, 30% high
processor.stratified_sampling(
target_size=1000,
low_pct=0.30,
medium_pct=0.40,
high_pct=0.30
)
# Export for vector DB
export_path = processor.export_for_vector_db()
# Generate summary report
report_path = processor.save_summary_report()
# Print summary
print("\n" + processor.generate_summary_report())
print("="*80)
print("✓ Post-processing complete!")
print("="*80)
print(f"\nOutput files:")
print(f" - Stratified sample: {export_path}")
print(f" - Summary report: {report_path}")
print(f"\nNext: Run vector DB builder with stratified sample")
if __name__ == "__main__":
main()
|