File size: 11,716 Bytes
cacd4d0 |
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 |
"""
Log Parser for Extracting Candidates and Feedback
Parses optimization logs to extract candidate prompts, feedback, and scores.
"""
import re
from typing import List, Dict, Optional, Tuple
from pathlib import Path
class OptimizationLogParser:
"""Parse optimization logs to extract candidates and feedback"""
def __init__(self, log_file: str):
"""
Initialize parser with log file path.
Args:
log_file: Path to log file
"""
self.log_file = Path(log_file)
self.content = ""
if self.log_file.exists():
with open(self.log_file, 'r', encoding='utf-8') as f:
self.content = f.read()
def extract_iterations(self) -> List[Dict]:
"""Extract iteration information from logs"""
iterations = []
# Pattern to find iteration markers
iteration_pattern = r'Iteration\s+(\d+)|Starting GEPA optimization|🚀 Starting GEPA optimization'
# Find all iteration starts
for match in re.finditer(iteration_pattern, self.content):
# Try to extract iteration number
iter_num = 1
if match.group(1):
iter_num = int(match.group(1))
# Find the section for this iteration
start_pos = match.start()
next_match = list(re.finditer(iteration_pattern, self.content))
next_idx = next((i for i, m in enumerate(next_match) if m.start() > start_pos), None)
if next_idx is not None:
end_pos = next_match[next_idx].start()
iter_content = self.content[start_pos:end_pos]
else:
iter_content = self.content[start_pos:]
iterations.append({
'iteration': iter_num,
'content': iter_content,
'start_pos': start_pos
})
return iterations
def extract_candidates(self, iteration_content: str) -> List[Dict]:
"""
Extract candidate prompts from iteration content.
Args:
iteration_content: Content for a single iteration
Returns:
List of candidate dictionaries
"""
candidates = []
# Pattern 1: GEPA Reflection candidates
# Look for "PROPOSED PROMPT" or "📝 PROPOSED PROMPT"
gepa_patterns = [
r'📝 PROPOSED PROMPT.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
r'PROPOSED PROMPT.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
r'GEPA REFLECTION.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
]
for pattern in gepa_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
prompt = match.group(1).strip()
if prompt and len(prompt) > 20: # Valid prompt
candidates.append({
'source': 'GEPA_Reflection',
'prompt': prompt,
'position': match.start()
})
# Pattern 2: LLEGO Crossover candidates
crossover_patterns = [
r'🧬 Crossover.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
r'Crossover.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
]
for pattern in crossover_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
prompt = match.group(1).strip()
if prompt and len(prompt) > 20:
candidates.append({
'source': 'LLEGO_Crossover',
'prompt': prompt,
'position': match.start()
})
# Pattern 3: LLEGO Mutation candidates
mutation_patterns = [
r'🎲 Mutation.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
r'Mutation.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
]
for pattern in mutation_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
prompt = match.group(1).strip()
if prompt and len(prompt) > 20:
candidates.append({
'source': 'LLEGO_Mutation',
'prompt': prompt,
'position': match.start()
})
# Pattern 4: Generic candidate markers
# Look for prompts in quotes or code blocks
generic_patterns = [
r'"([^"]{50,})"', # Quoted prompts
r'```\s*(.*?)\s*```', # Code blocks
]
for pattern in generic_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
prompt = match.group(1).strip()
# Check if it looks like a prompt (contains task instructions)
if (len(prompt) > 50 and
any(keyword in prompt.lower() for keyword in
['you are', 'task', 'instruction', 'element', 'identify', 'select'])):
# Check if we haven't already captured this
if not any(c['prompt'] == prompt for c in candidates):
candidates.append({
'source': 'Unknown',
'prompt': prompt,
'position': match.start()
})
# Sort by position
candidates.sort(key=lambda x: x['position'])
return candidates
def extract_feedback(self, iteration_content: str) -> List[Dict]:
"""
Extract feedback from iteration content.
Args:
iteration_content: Content for a single iteration
Returns:
List of feedback dictionaries
"""
feedback_list = []
# Pattern 1: Explicit feedback markers
feedback_patterns = [
r'💬 FEEDBACK:\s*(.*?)(?=\n\n|\n📊|\n🚀|\n💡|$)',
r'FEEDBACK:\s*(.*?)(?=\n\n|\n📊|\n🚀|\n💡|$)',
r'Feedback:\s*(.*?)(?=\n\n|\n📊|\n🚀|\n💡|$)',
]
for pattern in feedback_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
feedback_text = match.group(1).strip()
if feedback_text and len(feedback_text) > 10:
feedback_list.append({
'feedback': feedback_text,
'position': match.start()
})
# Pattern 2: LLM-as-Judge feedback
judge_patterns = [
r'LLM-as-Judge.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
r'Judge Feedback.*?----------------------------------------\s*(.*?)(?=----------------------------------------|📊|🚀|$)',
]
for pattern in judge_patterns:
for match in re.finditer(pattern, iteration_content, re.DOTALL):
feedback_text = match.group(1).strip()
if feedback_text and len(feedback_text) > 10:
feedback_list.append({
'feedback': feedback_text,
'position': match.start(),
'source': 'LLM-as-Judge'
})
# Sort by position
feedback_list.sort(key=lambda x: x['position'])
return feedback_list
def extract_scores(self, iteration_content: str) -> List[Dict]:
"""
Extract scores from iteration content.
Args:
iteration_content: Content for a single iteration
Returns:
List of score dictionaries
"""
scores = []
# Pattern for scores
score_patterns = [
r'Score:\s*([\d.]+)',
r'Average score:\s*([\d.]+)',
r'🎯 SCORE:\s*([\d.]+)',
r'📊 Score:\s*([\d.]+)',
]
for pattern in score_patterns:
for match in re.finditer(pattern, iteration_content):
score_value = float(match.group(1))
scores.append({
'score': score_value,
'position': match.start()
})
# Sort by position
scores.sort(key=lambda x: x['position'])
return scores
def parse_all(self) -> Dict:
"""
Parse entire log file and extract all information.
Returns:
Dictionary with all extracted information
"""
iterations = self.extract_iterations()
result = {
'iterations': [],
'total_iterations': len(iterations),
'all_candidates': [],
'all_feedback': []
}
for iter_info in iterations:
iter_num = iter_info['iteration']
iter_content = iter_info['content']
candidates = self.extract_candidates(iter_content)
feedback = self.extract_feedback(iter_content)
scores = self.extract_scores(iter_content)
# Try to associate scores with candidates
for i, candidate in enumerate(candidates):
# Find nearest score after this candidate
candidate_pos = candidate['position']
nearest_score = None
min_distance = float('inf')
for score_info in scores:
if score_info['position'] > candidate_pos:
distance = score_info['position'] - candidate_pos
if distance < min_distance:
min_distance = distance
nearest_score = score_info['score']
if nearest_score is not None:
candidate['score'] = nearest_score
# Try to associate feedback
nearest_feedback = None
min_distance = float('inf')
for feedback_info in feedback:
if feedback_info['position'] > candidate_pos:
distance = feedback_info['position'] - candidate_pos
if distance < min_distance and distance < 5000: # Within reasonable distance
min_distance = distance
nearest_feedback = feedback_info['feedback']
if nearest_feedback:
candidate['feedback'] = nearest_feedback
result['iterations'].append({
'iteration': iter_num,
'candidates': candidates,
'feedback': feedback,
'scores': scores
})
result['all_candidates'].extend(candidates)
result['all_feedback'].extend(feedback)
return result
|