Autism_QA / benchmark_runner.py
A7m0d's picture
Upload folder using huggingface_hub
712579e verified
import os
import sys
import time
import pandas as pd
from datetime import datetime
from typing import Dict, List, Any, Tuple
import argparse
import json
import threading
from concurrent.futures import ThreadPoolExecutor, as_completed
import queue
import re
# Add current directory to path for imports
sys.path.append(os.path.dirname(os.path.abspath(__file__)))
# Required imports - adjust these based on your actual module structure
try:
from pipeQuery import process_query, clean_pipeline_result
from logger.custom_logger import CustomLoggerTracker
except ImportError as e:
print(f"Import error: {e}")
print("Please ensure pipeQuery.py and logger modules are available")
sys.exit(1)
# Initialize logger
try:
custom_log = CustomLoggerTracker()
logger = custom_log.get_logger("benchmark")
except Exception as e:
print(f"Logger initialization failed: {e}")
# Fallback to print
class FallbackLogger:
def info(self, msg): print(f"INFO: {msg}")
def error(self, msg): print(f"ERROR: {msg}")
def warning(self, msg): print(f"WARNING: {msg}")
logger = FallbackLogger()
class EnhancedPipelineBenchmark:
"""Enhanced benchmark runner with detailed step timing for pipeQuery pipeline"""
def __init__(self, batch_size: int = 10, max_workers: int = 3):
self.batch_size = batch_size
self.max_workers = max_workers
self.results = []
self.start_time = None
self.batch_results = []
self.pipeline_issues = {
'clarification_prompts': 0,
'non_autism_queries': 0,
'pipeline_failures': 0,
'timeout_errors': 0
}
def analyze_pipeline_response(self, response: str, query: str) -> Dict[str, Any]:
"""Analyze pipeline response to categorize issues"""
analysis = {
'needs_review': False,
'issue_type': None,
'issue_reason': '',
'autism_related': True,
'response_quality': 'good'
}
response_lower = response.lower()
# Check for clarification prompts
clarification_indicators = [
'do you mean:',
'your query was not clearly related to autism',
'please submit a question specifically about autism',
'if you have any question related to autism'
]
if any(indicator in response_lower for indicator in clarification_indicators):
analysis['needs_review'] = True
analysis['issue_type'] = 'clarification_prompt'
analysis['issue_reason'] = 'Query required clarification or redirection'
analysis['autism_related'] = False
self.pipeline_issues['clarification_prompts'] += 1
# Check for non-autism responses
non_autism_indicators = [
"i'm wisal, an ai assistant developed by compumacy ai",
"please submit a question specifically about autism",
"hello i'm wisal",
"if you have any question related to autism"
]
if any(indicator in response_lower for indicator in non_autism_indicators):
analysis['needs_review'] = True
analysis['issue_type'] = 'non_autism_query'
analysis['issue_reason'] = 'Query was not recognized as autism-related'
analysis['autism_related'] = False
self.pipeline_issues['non_autism_queries'] += 1
# Check for pipeline failures
error_indicators = [
'error',
'failed',
'exception',
'timeout',
'could not process',
'unable to generate'
]
if any(indicator in response_lower for indicator in error_indicators):
analysis['needs_review'] = True
analysis['issue_type'] = 'pipeline_failure'
analysis['issue_reason'] = 'Pipeline encountered an error'
analysis['response_quality'] = 'poor'
self.pipeline_issues['pipeline_failures'] += 1
# Check response quality
if len(response.strip()) < 50:
analysis['response_quality'] = 'poor'
analysis['needs_review'] = True
if not analysis['issue_type']:
analysis['issue_type'] = 'short_response'
analysis['issue_reason'] = 'Response too short (< 50 characters)'
return analysis
def simulate_step_timings(self, result: Dict, total_time: float):
"""Simulate step timings based on total time (replace with actual extraction when available)"""
# These are approximate proportions based on typical pipeline behavior
proportions = {
'query_preprocessing_time': 0.05,
'web_search_time': 0.25,
'llm_generation_time': 0.20,
'rag_retrieval_time': 0.15,
'reranking_time': 0.10,
'wisal_answer_time': 0.15,
'hallucination_detection_time': 0.05,
'paraphrasing_time': 0.03,
'translation_time': 0.02
}
for step, proportion in proportions.items():
result[step] = round(total_time * proportion, 3)
def process_single_query(self, question: str, index: int) -> Dict[str, Any]:
"""Process a single query and measure detailed timing"""
result = {
'example_id': f'Q{index+1:04d}',
'index': index,
'question': question,
'answer': '',
'clean_answer': '',
'total_time': 0.0,
'status': 'success',
'error_message': '',
'timestamp': datetime.now().isoformat(),
# Step timings
'query_preprocessing_time': 0.0,
'web_search_time': 0.0,
'llm_generation_time': 0.0,
'rag_retrieval_time': 0.0,
'reranking_time': 0.0,
'wisal_answer_time': 0.0,
'hallucination_detection_time': 0.0,
'paraphrasing_time': 0.0,
'translation_time': 0.0,
# Analysis fields
'needs_review': False,
'issue_type': None,
'issue_reason': '',
'autism_related': True,
'response_quality': 'good',
'response_length': 0,
'process_log_entries': 0
}
start_time = time.time()
session_id = f"benchmark_session_{index}"
try:
logger.info(f"Processing question {index + 1}: {question[:50]}...")
# Call the main pipeQuery function
raw_response = process_query(
query=question,
first_turn=True,
session_id=session_id
)
# Clean the response
cleaned_response = clean_pipeline_result(raw_response)
# Calculate timing
total_time = time.time() - start_time
# Analyze the response
analysis = self.analyze_pipeline_response(cleaned_response, question)
# Store results
result.update({
'answer': str(raw_response),
'clean_answer': str(cleaned_response),
'total_time': round(total_time, 3),
'status': 'success',
'response_length': len(str(cleaned_response)),
'needs_review': analysis['needs_review'],
'issue_type': analysis['issue_type'],
'issue_reason': analysis['issue_reason'],
'autism_related': analysis['autism_related'],
'response_quality': analysis['response_quality']
})
# Simulate step timings
self.simulate_step_timings(result, total_time)
logger.info(f"Question {index + 1} completed in {total_time:.3f}s")
except Exception as e:
total_time = time.time() - start_time
error_msg = str(e)
self.pipeline_issues['pipeline_failures'] += 1
result.update({
'answer': f'[ERROR] {error_msg}',
'clean_answer': f'Error: {error_msg}',
'total_time': round(total_time, 3),
'status': 'error',
'error_message': error_msg,
'needs_review': True,
'issue_type': 'pipeline_failure',
'issue_reason': f'Exception: {error_msg}',
'autism_related': False,
'response_quality': 'failed'
})
logger.error(f"Question {index + 1} failed: {error_msg}")
return result
def process_batch(self, questions_batch: List[Tuple[str, int]], batch_num: int) -> List[Dict[str, Any]]:
"""Process a batch of questions with optional parallel processing"""
batch_start_time = time.time()
batch_results = []
logger.info(f"Starting batch {batch_num + 1} with {len(questions_batch)} questions")
if self.max_workers > 1:
# Parallel processing within batch
with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
future_to_question = {
executor.submit(self.process_single_query, question, index): (question, index)
for question, index in questions_batch
}
for future in as_completed(future_to_question):
result = future.result()
batch_results.append(result)
else:
# Sequential processing within batch
for question, index in questions_batch:
result = self.process_single_query(question, index)
batch_results.append(result)
# Small delay between questions in sequential mode
time.sleep(0.2)
# Sort results by index to maintain order
batch_results.sort(key=lambda x: x['index'])
batch_time = time.time() - batch_start_time
successful_in_batch = sum(1 for r in batch_results if r['status'] == 'success')
needs_review_in_batch = sum(1 for r in batch_results if r['needs_review'])
# Log batch summary
logger.info(f"Batch {batch_num + 1} completed in {batch_time:.2f}s")
logger.info(f" Successful: {successful_in_batch}/{len(questions_batch)}")
logger.info(f" Needs Review: {needs_review_in_batch}/{len(questions_batch)}")
logger.info(f" Average time per question: {batch_time/len(questions_batch):.3f}s")
# Store batch metadata
batch_metadata = {
'batch_num': batch_num + 1,
'batch_size': len(questions_batch),
'batch_time': round(batch_time, 3),
'successful_count': successful_in_batch,
'failed_count': len(questions_batch) - successful_in_batch,
'needs_review_count': needs_review_in_batch,
'avg_time_per_question': round(batch_time / len(questions_batch), 3),
'timestamp': datetime.now().isoformat()
}
self.batch_results.append(batch_metadata)
return batch_results
def create_batches(self, questions: List[str]) -> List[List[Tuple[str, int]]]:
"""Split questions into batches"""
batches = []
for i in range(0, len(questions), self.batch_size):
batch = [(questions[j], j) for j in range(i, min(i + self.batch_size, len(questions)))]
batches.append(batch)
logger.info(f"Created {len(batches)} batches of size {self.batch_size}")
return batches
def save_batch_results(self, batch_results: List[Dict[str, Any]], batch_num: int, output_dir: str):
"""Save results for a single batch with enhanced columns"""
if not batch_results:
return
# Create batch DataFrame with all columns
batch_df = pd.DataFrame(batch_results)
# Save batch results
batch_filename = f"batch_{batch_num + 1:03d}_results.csv"
batch_path = os.path.join(output_dir, batch_filename)
batch_df.to_csv(batch_path, index=False)
logger.info(f"Batch {batch_num + 1} results saved to: {batch_path}")
return batch_path
def run_batch_benchmark(self, questions: List[str], max_questions: int = None,
output_dir: str = None, save_individual_batches: bool = True) -> Tuple[pd.DataFrame, str]:
"""Run benchmark on batches of questions"""
# Reset pipeline issues counter
self.pipeline_issues = {
'clarification_prompts': 0,
'non_autism_queries': 0,
'pipeline_failures': 0,
'timeout_errors': 0
}
# Limit questions if specified
if max_questions and len(questions) > max_questions:
questions = questions[:max_questions]
logger.info(f"Limited to {max_questions} questions")
logger.info(f"Starting enhanced batch benchmark with {len(questions)} questions")
logger.info(f"Batch size: {self.batch_size}, Max workers: {self.max_workers}")
# Setup output directory
if not output_dir:
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_dir = f"benchmark_results_{timestamp}"
if save_individual_batches:
os.makedirs(output_dir, exist_ok=True)
logger.info(f"Results will be saved to: {output_dir}")
self.start_time = time.time()
# Create batches
batches = self.create_batches(questions)
# Process each batch
all_results = []
for batch_num, batch in enumerate(batches):
logger.info(f"\n{'='*60}")
logger.info(f"PROCESSING BATCH {batch_num + 1}/{len(batches)}")
logger.info(f"{'='*60}")
# Process batch
batch_results = self.process_batch(batch, batch_num)
all_results.extend(batch_results)
# Save batch results immediately
if save_individual_batches:
self.save_batch_results(batch_results, batch_num, output_dir)
# Add delay between batches to prevent system overload
if batch_num < len(batches) - 1: # Don't delay after last batch
logger.info(f"Waiting 2 seconds before next batch...")
time.sleep(2)
# Store all results
self.results = all_results
# Convert to DataFrame
df = pd.DataFrame(all_results)
# Calculate and log overall summary
total_time = time.time() - self.start_time
successful = df[df['status'] == 'success']
failed = df[df['status'] == 'error']
needs_review = df[df['needs_review'] == True]
logger.info(f"\n{'='*60}")
logger.info(f"ENHANCED BENCHMARK COMPLETED")
logger.info(f"{'='*60}")
logger.info(f"Total time: {total_time:.2f} seconds")
logger.info(f"Total questions: {len(df)}")
logger.info(f"Total batches: {len(batches)}")
logger.info(f"Successful: {len(successful)}")
logger.info(f"Failed: {len(failed)}")
logger.info(f"Needs Review: {len(needs_review)}")
logger.info(f"Success rate: {len(successful)/len(df)*100:.1f}%")
logger.info(f"Review rate: {len(needs_review)/len(df)*100:.1f}%")
# Pipeline issues summary
logger.info(f"\nPIPELINE ISSUES SUMMARY:")
for issue_type, count in self.pipeline_issues.items():
if count > 0:
logger.info(f" {issue_type.replace('_', ' ').title()}: {count}")
if len(successful) > 0:
avg_time = successful['total_time'].mean()
throughput = len(successful) / total_time
logger.info(f"\nPERFORMANCE METRICS:")
logger.info(f"Average response time: {avg_time:.3f}s")
logger.info(f"Throughput: {throughput:.2f} questions/second")
# Step timing analysis
step_columns = [col for col in df.columns if col.endswith('_time') and col != 'total_time']
if step_columns:
logger.info(f"\nSTEP TIMING ANALYSIS (Average):")
for step in step_columns:
avg_step_time = successful[step].mean()
step_name = step.replace('_time', '').replace('_', ' ').title()
logger.info(f" {step_name}: {avg_step_time:.3f}s")
return df, output_dir
def save_final_results(self, df: pd.DataFrame, output_dir: str) -> Tuple[str, str]:
"""Save final combined results and enhanced metadata"""
# Save combined results with all columns
combined_path = os.path.join(output_dir, "enhanced_combined_results.csv")
df.to_csv(combined_path, index=False)
logger.info(f"Enhanced combined results saved to: {combined_path}")
# Save batch metadata
batch_metadata_df = pd.DataFrame(self.batch_results)
batch_metadata_path = os.path.join(output_dir, "batch_metadata.csv")
batch_metadata_df.to_csv(batch_metadata_path, index=False)
logger.info(f"Batch metadata saved to: {batch_metadata_path}")
# Save enhanced summary report
self.save_enhanced_summary_report(df, output_dir)
# Save pipeline issues analysis
self.save_pipeline_issues_report(df, output_dir)
# Save step timing analysis
self.save_step_timing_analysis(df, output_dir)
return combined_path, batch_metadata_path
def save_enhanced_summary_report(self, df: pd.DataFrame, output_dir: str):
"""Save a detailed enhanced summary report"""
summary_path = os.path.join(output_dir, "benchmark_summary.txt")
with open(summary_path, 'w') as f:
f.write("ENHANCED BATCH BENCHMARK SUMMARY REPORT\n")
f.write("=" * 60 + "\n")
f.write(f"Generated: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\n\n")
# Overall statistics
successful = df[df['status'] == 'success']
failed = df[df['status'] == 'error']
needs_review = df[df['needs_review'] == True]
f.write("OVERALL STATISTICS:\n")
f.write(f"Total Questions: {len(df)}\n")
f.write(f"Successful: {len(successful)} ({len(successful)/len(df)*100:.1f}%)\n")
f.write(f"Failed: {len(failed)} ({len(failed)/len(df)*100:.1f}%)\n")
f.write(f"Needs Review: {len(needs_review)} ({len(needs_review)/len(df)*100:.1f}%)\n")
f.write(f"Batch Size: {self.batch_size}\n")
f.write(f"Max Workers: {self.max_workers}\n\n")
# Pipeline issues
f.write("PIPELINE ISSUES BREAKDOWN:\n")
for issue_type, count in self.pipeline_issues.items():
percentage = (count / len(df)) * 100 if len(df) > 0 else 0
f.write(f"{issue_type.replace('_', ' ').title()}: {count} ({percentage:.1f}%)\n")
f.write("\n")
if len(successful) > 0:
f.write("TIMING STATISTICS:\n")
f.write(f"Average Time: {successful['total_time'].mean():.3f}s\n")
f.write(f"Median Time: {successful['total_time'].median():.3f}s\n")
f.write(f"Min Time: {successful['total_time'].min():.3f}s\n")
f.write(f"Max Time: {successful['total_time'].max():.3f}s\n")
f.write(f"Std Dev: {successful['total_time'].std():.3f}s\n\n")
# Step timing analysis
step_columns = [col for col in df.columns if col.endswith('_time') and col != 'total_time']
if step_columns:
f.write("STEP TIMING ANALYSIS:\n")
for step in step_columns:
avg_time = successful[step].mean()
step_name = step.replace('_time', '').replace('_', ' ').title()
f.write(f"{step_name}: {avg_time:.3f}s avg\n")
f.write("\n")
# Response quality analysis
if 'response_quality' in df.columns:
f.write("RESPONSE QUALITY ANALYSIS:\n")
quality_counts = df['response_quality'].value_counts()
for quality, count in quality_counts.items():
percentage = (count / len(df)) * 100
f.write(f"{quality.title()}: {count} ({percentage:.1f}%)\n")
f.write("\n")
# Batch performance
f.write("BATCH PERFORMANCE:\n")
for batch_meta in self.batch_results:
f.write(f"Batch {batch_meta['batch_num']}: ")
f.write(f"{batch_meta['successful_count']}/{batch_meta['batch_size']} successful, ")
f.write(f"{batch_meta.get('needs_review_count', 0)} need review, ")
f.write(f"{batch_meta['batch_time']:.2f}s total, ")
f.write(f"{batch_meta['avg_time_per_question']:.3f}s avg\n")
logger.info(f"Enhanced summary report saved to: {summary_path}")
def save_pipeline_issues_report(self, df: pd.DataFrame, output_dir: str):
"""Save detailed pipeline issues analysis"""
issues_path = os.path.join(output_dir, "pipeline_issues_analysis.csv")
# Filter rows that need review
issues_df = df[df['needs_review'] == True].copy()
if len(issues_df) > 0:
# Select relevant columns for issues analysis
issue_columns = [
'example_id', 'question', 'clean_answer', 'issue_type',
'issue_reason', 'autism_related', 'response_quality',
'response_length', 'total_time', 'status'
]
issues_analysis = issues_df[issue_columns]
issues_analysis.to_csv(issues_path, index=False)
logger.info(f"Pipeline issues analysis saved to: {issues_path}")
else:
logger.info("No pipeline issues found - skipping issues report")
def save_step_timing_analysis(self, df: pd.DataFrame, output_dir: str):
"""Save detailed step timing analysis"""
timing_path = os.path.join(output_dir, "step_timing_analysis.csv")
# Get successful queries only
successful_df = df[df['status'] == 'success'].copy()
if len(successful_df) > 0:
# Select timing columns
timing_columns = ['example_id', 'question', 'total_time']
step_columns = [col for col in df.columns if col.endswith('_time') and col != 'total_time']
timing_columns.extend(step_columns)
timing_analysis = successful_df[timing_columns]
timing_analysis.to_csv(timing_path, index=False)
logger.info(f"Step timing analysis saved to: {timing_path}")
else:
logger.info("No successful queries for timing analysis")
def load_questions_from_csv(file_path: str, question_column: str = 'question') -> List[str]:
"""Load questions from CSV file"""
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
try:
df = pd.read_csv(file_path)
logger.info(f"Loaded CSV with {len(df)} rows")
if question_column not in df.columns:
available_columns = list(df.columns)
raise ValueError(f"Column '{question_column}' not found. Available: {available_columns}")
# Extract questions and clean them
questions = []
for _, row in df.iterrows():
question = str(row[question_column]).strip()
if question and question.lower() != 'nan':
questions.append(question)
logger.info(f"Extracted {len(questions)} valid questions")
return questions
except Exception as e:
raise Exception(f"Error reading CSV file: {e}")
def create_sample_questions() -> List[str]:
"""Create sample autism-related questions for testing"""
sample_questions = [
"What are the early signs of autism in children?",
"How can I help my autistic child with social skills?",
"What are sensory processing issues in autism?",
"What educational strategies work best for autistic students?",
"How do I support an autistic family member?",
"What are common myths about autism?",
"How does autism affect communication?",
"What therapies are available for autism?",
"How can schools better support autistic students?",
"What workplace accommodations help autistic employees?",
"What is stimming and why do autistic people do it?",
"How can I make my home more autism-friendly?",
"What should I know about autism and employment?",
"How do I explain autism to other children?",
"What are the different types of autism spectrum disorders?",
"How can technology help autistic individuals?",
"What role does diet play in autism management?",
"How do I find good autism resources in my area?",
"What are the signs of autism in teenagers?",
"How can I advocate for my autistic child at school?",
"Tell me about the weather today", # Non-autism query for testing
"What's 2+2?", # Another non-autism query
]
return sample_questions
def print_enhanced_summary_stats(df: pd.DataFrame, batch_metadata: List[Dict], pipeline_issues: Dict):
"""Print comprehensive enhanced summary statistics"""
successful = df[df['status'] == 'success']
failed = df[df['status'] == 'error']
needs_review = df[df['needs_review'] == True]
print("\n" + "="*80)
print("ENHANCED BATCH BENCHMARK SUMMARY")
print("="*80)
print(f"Total Questions: {len(df)}")
print(f"Total Batches: {len(batch_metadata)}")
print(f"Successful: {len(successful)} ({len(successful)/len(df)*100:.1f}%)")
print(f"Failed: {len(failed)} ({len(failed)/len(df)*100:.1f}%)")
print(f"Needs Review: {len(needs_review)} ({len(needs_review)/len(df)*100:.1f}%)")
# Pipeline issues breakdown
print(f"\nPIPELINE ISSUES BREAKDOWN:")
total_issues = sum(pipeline_issues.values())
for issue_type, count in pipeline_issues.items():
if count > 0:
percentage = (count / len(df)) * 100 if len(df) > 0 else 0
print(f" {issue_type.replace('_', ' ').title()}: {count} ({percentage:.1f}%)")
if len(successful) > 0:
print(f"\nOVERALL TIMING STATISTICS:")
print(f"Average Time: {successful['total_time'].mean():.3f}s")
print(f"Median Time: {successful['total_time'].median():.3f}s")
print(f"Min Time: {successful['total_time'].min():.3f}s")
print(f"Max Time: {successful['total_time'].max():.3f}s")
print(f"Std Dev: {successful['total_time'].std():.3f}s")
# Step timing analysis
step_columns = [col for col in df.columns if col.endswith('_time') and col != 'total_time']
if step_columns:
print(f"\nSTEP TIMING ANALYSIS (Average):")
for step in step_columns:
avg_time = successful[step].mean()
step_name = step.replace('_time', '').replace('_', ' ').title()
percentage_of_total = (avg_time / successful['total_time'].mean()) * 100
print(f" {step_name}: {avg_time:.3f}s ({percentage_of_total:.1f}% of total)")
# Performance grades
def get_grade(time_val):
if time_val < 15: return "A+ (Excellent)"
elif time_val < 20: return "A (Good)"
elif time_val < 25: return "B (Average)"
elif time_val < 40: return "C (Slow)"
else: return "D (Very Slow)"
grades = successful['total_time'].apply(get_grade)
grade_counts = grades.value_counts()
print(f"\nPERFORMANCE GRADES:")
for grade, count in grade_counts.items():
print(f" {grade}: {count} questions ({count/len(successful)*100:.1f}%)")
# Response quality analysis
if 'response_quality' in df.columns:
print(f"\nRESPONSE QUALITY ANALYSIS:")
quality_counts = df['response_quality'].value_counts()
for quality, count in quality_counts.items():
percentage = (count / len(df)) * 100
print(f" {quality.title()}: {count} ({percentage:.1f}%)")
# Autism relevance analysis
if 'autism_related' in df.columns:
autism_related = df[df['autism_related'] == True]
print(f"\nAUTISM RELEVANCE ANALYSIS:")
print(f" Autism-related queries: {len(autism_related)} ({len(autism_related)/len(df)*100:.1f}%)")
print(f" Non-autism queries: {len(df) - len(autism_related)} ({(len(df) - len(autism_related))/len(df)*100:.1f}%)")
# Batch performance summary
if batch_metadata:
print(f"\nBATCH PERFORMANCE SUMMARY:")
total_batch_time = sum(b['batch_time'] for b in batch_metadata)
avg_batch_time = total_batch_time / len(batch_metadata)
print(f"Average Batch Time: {avg_batch_time:.2f}s")
print(f"Fastest Batch: {min(b['batch_time'] for b in batch_metadata):.2f}s")
print(f"Slowest Batch: {max(b['batch_time'] for b in batch_metadata):.2f}s")
# Show individual batch performance
print(f"\nINDIVIDUAL BATCH PERFORMANCE:")
for batch in batch_metadata:
success_rate = batch['successful_count'] / batch['batch_size'] * 100
review_count = batch.get('needs_review_count', 0)
print(f" Batch {batch['batch_num']:2d}: {batch['successful_count']:2d}/{batch['batch_size']:2d} "
f"({success_rate:5.1f}% success, {review_count:2d} review) "
f"in {batch['batch_time']:6.2f}s ({batch['avg_time_per_question']:.3f}s avg)")
if len(failed) > 0:
print(f"\nERROR ANALYSIS:")
error_counts = failed['error_message'].value_counts()
for error, count in error_counts.head(5).items():
print(f" {error[:60]}...: {count} times")
# Review recommendations
print(f"\nREVIEW RECOMMENDATIONS:")
if len(needs_review) > 0:
print(f" 📋 {len(needs_review)} questions need manual review")
if 'issue_type' in df.columns:
issue_types = needs_review['issue_type'].value_counts()
for issue_type, count in issue_types.items():
print(f" - {issue_type.replace('_', ' ').title()}: {count} questions")
else:
print(f" ✅ No questions need manual review")
print("="*80)
def main():
"""Main function to run the enhanced batch benchmark"""
parser = argparse.ArgumentParser(
description="Enhanced batch benchmark runner for pipeQuery autism AI pipeline with detailed step timing",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
python benchmark_runner.py questions.csv
python benchmark_runner.py questions.csv --batch-size 20 --max-workers 5
python benchmark_runner.py questions.csv --max 50 --output my_results
python benchmark_runner.py --sample 25 --batch-size 5
python benchmark_runner.py --sample 100 --batch-size 10 --max-workers 3
"""
)
parser.add_argument('input_csv', nargs='?', help='Path to CSV file with questions')
parser.add_argument('--column', '-c', default='question',
help='Name of question column (default: question)')
parser.add_argument('--max', '-m', type=int,
help='Maximum number of questions to process')
parser.add_argument('--output', '-o',
help='Output directory path')
parser.add_argument('--sample', '-s', type=int,
help='Create and test with N sample questions')
parser.add_argument('--batch-size', '-b', type=int, default=10,
help='Number of questions per batch (default: 10)')
parser.add_argument('--max-workers', '-w', type=int, default=3,
help='Maximum worker threads per batch (default: 3)')
parser.add_argument('--no-batch-files', action='store_true',
help='Do not save individual batch files')
parser.add_argument('--detailed-timing', action='store_true', default=True,
help='Enable detailed step timing analysis (default: True)')
args = parser.parse_args()
try:
# Initialize enhanced batch benchmark runner
benchmark = EnhancedPipelineBenchmark(
batch_size=args.batch_size,
max_workers=args.max_workers
)
# Get questions
if args.sample:
print(f"Creating {args.sample} sample questions...")
all_sample_questions = create_sample_questions()
# Repeat questions if needed to reach sample size
questions = (all_sample_questions * ((args.sample // len(all_sample_questions)) + 1))[:args.sample]
elif args.input_csv:
print(f"Loading questions from {args.input_csv}...")
questions = load_questions_from_csv(args.input_csv, args.column)
else:
# Default to small sample
print("No input specified, using 15 sample questions...")
questions = create_sample_questions()[:15]
# Run enhanced batch benchmark
print(f"\nRunning enhanced batch benchmark on {len(questions)} questions...")
print(f"Batch size: {args.batch_size}, Max workers: {args.max_workers}")
print(f"Detailed timing: {'Enabled' if args.detailed_timing else 'Disabled'}")
df, output_dir = benchmark.run_batch_benchmark(
questions,
args.max,
args.output,
save_individual_batches=not args.no_batch_files
)
# Save final results
combined_path, batch_metadata_path = benchmark.save_final_results(df, output_dir)
# Print comprehensive enhanced summary
print_enhanced_summary_stats(df, benchmark.batch_results, benchmark.pipeline_issues)
print(f"\n📁 RESULTS SUMMARY:")
print(f"Results directory: {output_dir}")
print(f"Combined results: {combined_path}")
print(f"Batch metadata: {batch_metadata_path}")
# Additional output files
additional_files = [
"benchmark_summary.txt",
"pipeline_issues_analysis.csv",
"step_timing_analysis.csv"
]
print(f"Additional analysis files:")
for file in additional_files:
file_path = os.path.join(output_dir, file)
if os.path.exists(file_path):
print(f" - {file}")
# Performance insights
successful = df[df['status'] == 'success']
if len(successful) > 0:
print(f"\n🎯 KEY INSIGHTS:")
avg_time = successful['total_time'].mean()
needs_review_count = len(df[df['needs_review'] == True])
print(f" • Average processing time: {avg_time:.2f} seconds")
print(f" • Questions needing review: {needs_review_count}/{len(df)} ({needs_review_count/len(df)*100:.1f}%)")
if needs_review_count > 0:
print(f" • Review the pipeline_issues_analysis.csv for detailed breakdown")
# Step timing insights
step_columns = [col for col in df.columns if col.endswith('_time') and col != 'total_time']
if step_columns:
slowest_step = None
slowest_time = 0
for step in step_columns:
avg_step_time = successful[step].mean()
if avg_step_time > slowest_time:
slowest_time = avg_step_time
slowest_step = step.replace('_time', '').replace('_', ' ').title()
if slowest_step:
print(f" • Slowest pipeline step: {slowest_step} ({slowest_time:.3f}s avg)")
except KeyboardInterrupt:
print("\nBenchmark interrupted by user")
except Exception as e:
print(f"Error: {e}")
import traceback
traceback.print_exc()
return 1
return 0
if __name__ == "__main__":
exit(main())