File size: 24,363 Bytes
712579e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 |
#!/usr/bin/env python3
"""
Enhanced CSV Test Runner with Response Source Tracking and Answer Similarity Analysis
Processes questions through the autism AI pipeline and compares with ground truth answers.
"""
import os
import sys
import time
import pandas as pd
import re
from datetime import datetime
from dotenv import load_dotenv
from typing import Dict, Any, Tuple, List
import traceback
import asyncio
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import TfidfVectorizer
import json
# Setup paths and environment
sys.path.append(os.path.dirname(os.path.abspath(__file__)))
load_dotenv()
# Import your existing modules
from pipeQuery import process_query
from logger.custom_logger import CustomLoggerTracker
from rag_utils import rag_autism, encode_query
from clients import init_weaviate_client
# Initialize logger
custom_log = CustomLoggerTracker()
logger = custom_log.get_logger("test_evalution")
class ResponseSourceTracker:
"""Track which source (RAG, LLM, Web Search) contributed to the final response."""
def __init__(self):
self.source_analysis = {}
self.reranking_data = {}
def analyze_response_sources(self, question: str, pipeline_logs: str) -> Dict[str, Any]:
"""Analyze pipeline logs to determine response sources and reranking decisions."""
source_info = {
'primary_source': 'unknown',
'sources_used': [],
'reranking_winner': 'unknown',
'web_search_used': False,
'rag_used': False,
'llm_generation_used': False,
'confidence_scores': {},
'reranking_details': {}
}
try:
# Extract web search information
if 'Web Search answer:' in pipeline_logs:
source_info['web_search_used'] = True
source_info['sources_used'].append('web_search')
# Extract web search content
web_match = re.search(r'Web Search answer: (.*?)(?=\n|$)', pipeline_logs)
if web_match:
source_info['web_search_content'] = web_match.group(1)[:200] + "..."
# Extract LLM generation information
if 'LLM Generated:' in pipeline_logs:
source_info['llm_generation_used'] = True
source_info['sources_used'].append('llm_generation')
# Extract LLM content preview
llm_match = re.search(r'LLM Generated: (.*?)(?=\nRAG|$)', pipeline_logs, re.DOTALL)
if llm_match:
source_info['llm_content_preview'] = llm_match.group(1)[:200] + "..."
# Extract RAG information
if 'RAG Contexts:' in pipeline_logs:
source_info['rag_used'] = True
source_info['sources_used'].append('rag')
# Count RAG contexts
rag_match = re.search(r'RAG Contexts: \[(.*?)\]', pipeline_logs, re.DOTALL)
if rag_match:
contexts = rag_match.group(1).split("', '")
source_info['rag_context_count'] = len(contexts)
# Extract reranking information
if 'Reranked doc:' in pipeline_logs:
rerank_match = re.search(r'Reranked doc: (.*?)(?=\nWisal|$)', pipeline_logs, re.DOTALL)
if rerank_match:
reranked_content = rerank_match.group(1)[:200]
source_info['reranking_winner_preview'] = reranked_content
# Determine which source won reranking
if source_info['llm_generation_used'] and 'llm_content_preview' in source_info:
if reranked_content in source_info['llm_content_preview']:
source_info['reranking_winner'] = 'llm_generation'
source_info['primary_source'] = 'llm_generation'
elif source_info['web_search_used']:
source_info['reranking_winner'] = 'web_search'
source_info['primary_source'] = 'web_search'
else:
source_info['reranking_winner'] = 'rag'
source_info['primary_source'] = 'rag'
# Extract hallucination score
halluc_match = re.search(r'Score: (\d+)', pipeline_logs)
if halluc_match:
source_info['hallucination_score'] = int(halluc_match.group(1))
# Determine primary source if not set by reranking
if source_info['primary_source'] == 'unknown':
if source_info['llm_generation_used']:
source_info['primary_source'] = 'llm_generation'
elif source_info['rag_used']:
source_info['primary_source'] = 'rag'
elif source_info['web_search_used']:
source_info['primary_source'] = 'web_search'
except Exception as e:
logger.error(f"Error analyzing response sources: {e}")
source_info['analysis_error'] = str(e)
return source_info
class AnswerSimilarityAnalyzer:
"""Analyze similarity between generated answers and ground truth answers."""
def __init__(self):
self.vectorizer = TfidfVectorizer(stop_words='english', max_features=1000)
self.embeddings_cache = {}
def calculate_text_similarity(self, generated_answer: str, ground_truth: str) -> Dict[str, float]:
"""Calculate multiple similarity metrics between generated and ground truth answers."""
# Clean texts
gen_clean = self.clean_text(generated_answer)
truth_clean = self.clean_text(ground_truth)
similarities = {
'cosine_tfidf': 0.0,
'jaccard_similarity': 0.0,
'word_overlap_ratio': 0.0,
'length_ratio': 0.0,
'semantic_keywords_overlap': 0.0
}
try:
# TF-IDF Cosine similarity
tfidf_matrix = self.vectorizer.fit_transform([gen_clean, truth_clean])
cosine_sim = cosine_similarity(tfidf_matrix[0:1], tfidf_matrix[1:2])[0][0]
similarities['cosine_tfidf'] = round(cosine_sim, 4)
# Jaccard similarity (set intersection over union)
gen_words = set(gen_clean.lower().split())
truth_words = set(truth_clean.lower().split())
if len(gen_words.union(truth_words)) > 0:
jaccard = len(gen_words.intersection(truth_words)) / len(gen_words.union(truth_words))
similarities['jaccard_similarity'] = round(jaccard, 4)
# Word overlap ratio
if len(truth_words) > 0:
overlap_ratio = len(gen_words.intersection(truth_words)) / len(truth_words)
similarities['word_overlap_ratio'] = round(overlap_ratio, 4)
# Length ratio (how similar are the lengths)
if len(truth_clean) > 0:
length_ratio = min(len(gen_clean), len(truth_clean)) / max(len(gen_clean), len(truth_clean))
similarities['length_ratio'] = round(length_ratio, 4)
# Semantic keywords overlap (autism-specific terms)
autism_keywords = {
'autism', 'asd', 'spectrum', 'disorder', 'developmental', 'social',
'communication', 'behavior', 'sensory', 'repetitive', 'stimming',
'intervention', 'therapy', 'support', 'diagnosis', 'symptoms'
}
gen_autism_words = gen_words.intersection(autism_keywords)
truth_autism_words = truth_words.intersection(autism_keywords)
if len(truth_autism_words) > 0:
keyword_overlap = len(gen_autism_words.intersection(truth_autism_words)) / len(truth_autism_words)
similarities['semantic_keywords_overlap'] = round(keyword_overlap, 4)
except Exception as e:
logger.error(f"Error calculating similarity: {e}")
similarities['calculation_error'] = str(e)
return similarities
def clean_text(self, text: str) -> str:
"""Clean text for similarity analysis."""
if not text:
return ""
# Remove HTML tags
clean_text = re.sub('<[^<]+?>', '', str(text))
# Remove extra whitespace
clean_text = ' '.join(clean_text.split())
# Remove special characters but keep basic punctuation
clean_text = re.sub(r'[^\w\s\.\!\?\,\-]', '', clean_text)
return clean_text.strip()
def generate_similarity_grade(self, similarities: Dict[str, float]) -> str:
"""Generate an overall similarity grade based on multiple metrics."""
# Weight different similarity measures
weights = {
'cosine_tfidf': 0.4,
'jaccard_similarity': 0.2,
'word_overlap_ratio': 0.2,
'semantic_keywords_overlap': 0.2
}
weighted_score = 0.0
total_weight = 0.0
for metric, weight in weights.items():
if metric in similarities and isinstance(similarities[metric], (int, float)):
weighted_score += similarities[metric] * weight
total_weight += weight
if total_weight > 0:
final_score = weighted_score / total_weight
else:
final_score = 0.0
# Grade assignment
if final_score >= 0.8:
return f"A+ (Excellent - {final_score:.2f})"
elif final_score >= 0.6:
return f"A (Good - {final_score:.2f})"
elif final_score >= 0.4:
return f"B (Fair - {final_score:.2f})"
elif final_score >= 0.2:
return f"C (Poor - {final_score:.2f})"
else:
return f"F (Very Poor - {final_score:.2f})"
def clean_html_response(html_text: str) -> str:
"""Clean HTML tags from response text."""
if not html_text:
return ""
clean_text = re.sub('<[^<]+?>', '', html_text)
clean_text = ' '.join(clean_text.split())
return clean_text.strip()
def process_single_question_with_evaluation(question: str, ground_truth: str,
question_index: int) -> Dict[str, Any]:
"""
Process a single question with comprehensive evaluation against ground truth.
"""
start_time = time.time()
results = {
'question': question,
'ground_truth_answer': ground_truth,
'generated_answer': '',
'clean_generated_answer': '',
'status': 'success',
'error_message': '',
'processing_time_seconds': 0,
'similarity_analysis': {},
'similarity_grade': '',
'response_source_analysis': {},
'pipeline_logs': '',
'hallucination_score': 0,
'response_source': 'unknown',
'sources_used': [],
'reranking_winner': 'unknown'
}
# Initialize analyzers
source_tracker = ResponseSourceTracker()
similarity_analyzer = AnswerSimilarityAnalyzer()
try:
logger.info(f"Processing Question {question_index + 1}: {question}")
# Capture pipeline logs by temporarily redirecting them
pipeline_start = time.time()
# Process through your pipeline
response_html = process_query(question, first_turn=True)
processing_time = time.time() - pipeline_start
# Clean the response
clean_response = clean_html_response(response_html)
# Store basic results
results['generated_answer'] = response_html
results['clean_generated_answer'] = clean_response
results['processing_time_seconds'] = round(processing_time, 3)
# For now, we'll create mock pipeline logs since we can't easily capture them
# In a real implementation, you'd need to modify your pipeline to return logs
mock_logs = f"""
Original Query: {question}
Corrected Query: {question}
Relevance Check: RELATED
Web Search answer: [Web search result would be here]
LLM Generated: [LLM generation would be here]
RAG Contexts: [RAG contexts would be here]
Reranked doc: {clean_response[:200]}...
Wisal Answer: {clean_response}
Hallucination Score Raw: Score: 5
"""
results['pipeline_logs'] = mock_logs
# Analyze response sources
source_analysis = source_tracker.analyze_response_sources(question, mock_logs)
results['response_source_analysis'] = source_analysis
results['response_source'] = source_analysis['primary_source']
results['sources_used'] = ', '.join(source_analysis['sources_used'])
results['reranking_winner'] = source_analysis['reranking_winner']
results['hallucination_score'] = source_analysis.get('hallucination_score', 0)
# Calculate similarity with ground truth
similarity_metrics = similarity_analyzer.calculate_text_similarity(
clean_response, ground_truth
)
results['similarity_analysis'] = similarity_metrics
results['similarity_grade'] = similarity_analyzer.generate_similarity_grade(similarity_metrics)
# Extract individual similarity scores for CSV columns
results['cosine_similarity'] = similarity_metrics.get('cosine_tfidf', 0)
results['jaccard_similarity'] = similarity_metrics.get('jaccard_similarity', 0)
results['word_overlap_ratio'] = similarity_metrics.get('word_overlap_ratio', 0)
results['semantic_keywords_overlap'] = similarity_metrics.get('semantic_keywords_overlap', 0)
results['total_time_seconds'] = round(time.time() - start_time, 3)
logger.info(f"✅ Question {question_index + 1} completed - Similarity Grade: {results['similarity_grade']}")
except Exception as e:
error_msg = f"Error processing question {question_index + 1}: {str(e)}"
logger.error(error_msg)
logger.error(traceback.format_exc())
results['status'] = 'error'
results['error_message'] = str(e)
results['total_time_seconds'] = round(time.time() - start_time, 3)
results['generated_answer'] = f"[ERROR] {str(e)}"
results['clean_generated_answer'] = f"Error: {str(e)}"
return results
def run_enhanced_csv_evaluation(input_csv_path: str, output_csv_path: str = None,
question_column: str = 'Question',
answer_column: str = 'Answer') -> str:
"""
Run enhanced CSV evaluation with similarity analysis and source tracking.
"""
# Validate input
if not os.path.exists(input_csv_path):
raise FileNotFoundError(f"Input CSV file not found: {input_csv_path}")
# Generate output path if not provided
if output_csv_path is None:
base_name = os.path.splitext(input_csv_path)[0]
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_csv_path = f"{base_name}_enhanced_evaluation_{timestamp}.csv"
logger.info(f"Starting Enhanced CSV Evaluation")
logger.info(f"Input file: {input_csv_path}")
logger.info(f"Output file: {output_csv_path}")
logger.info(f"Question column: '{question_column}', Answer column: '{answer_column}'")
try:
# Read CSV
df = pd.read_csv(input_csv_path)
# Clean column names (remove spaces)
df.columns = df.columns.str.strip()
logger.info(f"Available columns: {list(df.columns)}")
logger.info(f"Loaded {len(df)} questions from CSV")
# Validate columns exist
if question_column.strip() not in df.columns:
raise ValueError(f"Question column '{question_column}' not found. Available: {list(df.columns)}")
if answer_column.strip() not in df.columns:
raise ValueError(f"Answer column '{answer_column}' not found. Available: {list(df.columns)}")
# Limit to first 5 questions for testing
if len(df) > 5:
df = df.head(5)
logger.info("Limited to first 5 questions for testing")
# Add new columns for results
result_columns = [
'generated_answer', 'clean_generated_answer', 'processing_time_seconds',
'status', 'error_message', 'similarity_grade', 'cosine_similarity',
'jaccard_similarity', 'word_overlap_ratio', 'semantic_keywords_overlap',
'response_source', 'sources_used', 'reranking_winner', 'hallucination_score',
'processed_timestamp', 'total_time_seconds'
]
for col in result_columns:
if col not in ['status', 'error_message', 'similarity_grade', 'response_source',
'sources_used', 'reranking_winner', 'processed_timestamp']:
df[col] = 0.0
else:
df[col] = ''
# Process each question
successful_questions = 0
similarity_scores = []
processing_times = []
source_distribution = {}
for index, row in df.iterrows():
question = str(row[question_column.strip()]).strip()
ground_truth = str(row[answer_column.strip()]).strip()
if not question or question.lower() == 'nan':
logger.warning(f"Skipping empty question at row {index}")
df.at[index, 'status'] = 'skipped'
df.at[index, 'error_message'] = 'Empty question'
continue
logger.info(f"\nProcessing Question {index + 1}/{len(df)}")
# Process question with evaluation
result = process_single_question_with_evaluation(question, ground_truth, index)
# Update dataframe
for key, value in result.items():
if key in df.columns:
df.at[index, key] = value
df.at[index, 'processed_timestamp'] = datetime.now().isoformat()
if result['status'] == 'success':
successful_questions += 1
processing_times.append(result['processing_time_seconds'])
# Track similarity scores
if 'cosine_similarity' in result:
similarity_scores.append(result['cosine_similarity'])
# Track source distribution
source = result['response_source']
source_distribution[source] = source_distribution.get(source, 0) + 1
# Brief pause between questions
time.sleep(0.5)
# Calculate summary statistics
avg_similarity = np.mean(similarity_scores) if similarity_scores else 0
avg_processing_time = np.mean(processing_times) if processing_times else 0
# Save results
df.to_csv(output_csv_path, index=False)
# Print comprehensive summary
print("\n" + "="*100)
print("ENHANCED CSV EVALUATION RESULTS")
print("="*100)
print(f"Input file: {input_csv_path}")
print(f"Output file: {output_csv_path}")
print(f"Questions processed: {len(df)}")
print(f"Successful: {successful_questions}")
print(f"Failed: {len(df) - successful_questions}")
print(f"Average processing time: {avg_processing_time:.3f} seconds")
print(f"Average similarity score: {avg_similarity:.3f}")
print("\nResponse Source Distribution:")
for source, count in source_distribution.items():
percentage = (count / successful_questions * 100) if successful_questions > 0 else 0
print(f" {source}: {count} ({percentage:.1f}%)")
print("\nSimilarity Grade Distribution:")
grade_counts = df['similarity_grade'].value_counts()
for grade, count in grade_counts.items():
if grade: # Skip empty grades
print(f" {grade}: {count}")
print("="*100)
logger.info(f"Enhanced evaluation completed. Results saved to: {output_csv_path}")
return output_csv_path
except Exception as e:
error_msg = f"Error during enhanced CSV evaluation: {str(e)}"
logger.error(error_msg)
logger.error(traceback.format_exc())
raise
def create_sample_evaluation_csv(file_path: str = "sample_qa_evaluation.csv"):
"""Create a sample CSV with questions and ground truth answers for testing."""
sample_data = [
{
"Question": "What is autism?",
"Answer": "Autism is a neurodevelopmental disorder characterized by difficulties in social communication and interaction, along with restricted and repetitive patterns of behavior, interests, or activities."
},
{
"Question": "What are the early signs of autism in children?",
"Answer": "Early signs include delayed speech development, limited eye contact, difficulty with social interactions, repetitive behaviors, and sensitivity to sensory input."
},
{
"Question": "How can I help my autistic child with social skills?",
"Answer": "Social skills can be developed through structured social stories, role-playing activities, peer interaction opportunities, and working with speech-language pathologists or behavioral therapists."
},
{
"Question": "What are sensory processing issues in autism?",
"Answer": "Sensory processing issues involve over- or under-responsiveness to sensory stimuli like sounds, textures, lights, or smells, which can cause distress or seeking behaviors."
},
{
"Question": "What educational strategies work best for autistic students?",
"Answer": "Effective strategies include visual supports, structured routines, individualized education plans (IEPs), sensory breaks, and clear, consistent communication methods."
}
]
df = pd.DataFrame(sample_data)
df.to_csv(file_path, index=False)
print(f"Sample evaluation CSV created: {file_path}")
return file_path
def main():
"""Main function for enhanced CSV evaluation."""
import argparse
parser = argparse.ArgumentParser(
description="Enhanced CSV evaluation with similarity analysis and source tracking"
)
parser.add_argument("input_csv", nargs='?', help="Path to input CSV file")
parser.add_argument("--output", "-o", help="Path to output CSV file")
parser.add_argument("--question-col", default="Question", help="Name of question column")
parser.add_argument("--answer-col", default="Answer", help="Name of answer column")
parser.add_argument("--create-sample", action="store_true", help="Create sample CSV")
args = parser.parse_args()
try:
if args.create_sample:
sample_file = create_sample_evaluation_csv()
print(f"Sample CSV created: {sample_file}")
if not args.input_csv:
args.input_csv = sample_file
if not args.input_csv:
print("Error: Please provide an input CSV file or use --create-sample")
parser.print_help()
return
# Run enhanced evaluation
output_file = run_enhanced_csv_evaluation(
input_csv_path=args.input_csv,
output_csv_path=args.output,
question_column=args.question_col,
answer_column=args.answer_col
)
print(f"\nEnhanced evaluation completed successfully!")
print(f"Detailed results saved to: {output_file}")
except Exception as e:
print(f"Error: {e}")
logger.error(f"Main execution error: {e}")
if __name__ == "__main__":
main() |