tests/comprehensive_accuracy_test.py

#!/usr/bin/env python3
"""
Comprehensive Accuracy Test - Full GAIA Benchmark Evaluation
Runs all 20 questions through the async batch processor for complete accuracy assessment
"""

import asyncio
import sys
from pathlib import Path
from datetime import datetime
import json

# Add parent directory to path for imports
sys.path.append(str(Path(__file__).parent.parent))

from tests.async_batch_processor import BatchQuestionProcessor
from gaia_web_loader import GAIAQuestionLoaderWeb


async def run_comprehensive_accuracy_test():
    """Run comprehensive accuracy test on all available GAIA questions"""
    
    print("🎯 COMPREHENSIVE GAIA ACCURACY TEST")
    print("=" * 80)
    print(f"🕐 Start Time: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
    print(f"🎯 Goal: Establish baseline accuracy and identify improvement areas")
    print()
    
    try:
        # Load all questions
        print("📋 Loading all GAIA questions...")
        loader = GAIAQuestionLoaderWeb()
        all_questions = loader.questions
        
        print(f"✅ Loaded {len(all_questions)} questions from GAIA benchmark")
        
        # Show question distribution by level
        level_counts = {}
        classification_preview = {}
        
        for q in all_questions:
            level = q.get('Level', 'Unknown')
            level_counts[level] = level_counts.get(level, 0) + 1
            
            # Quick classification preview (first 5 questions)
            if len(classification_preview) < 5:
                task_id = q.get('task_id', 'unknown')
                question_preview = q.get('question', '')[:60] + "..."
                has_file = "Yes" if q.get('file_name') else "No"
                classification_preview[task_id[:8]] = {
                    'question': question_preview,
                    'level': level,
                    'has_file': has_file
                }
        
        print(f"\n📊 Question Distribution:")
        for level, count in sorted(level_counts.items()):
            print(f"  Level {level}: {count} questions")
        
        print(f"\n📋 Sample Questions:")
        for task_id, info in classification_preview.items():
            print(f"  {task_id}... | L{info['level']} | File: {info['has_file']} | {info['question']}")
        
        # Initialize batch processor with production settings
        print(f"\n🚀 Initializing production-grade batch processor...")
        processor = BatchQuestionProcessor(
            max_concurrent=3,  # Balanced concurrency for stability
            question_timeout=900,  # 15 minutes per question for complex cases
            progress_interval=15   # Progress updates every 15 seconds
        )
        
        print(f"⚙️  Configuration:")
        print(f"   - Max Concurrent: {processor.max_concurrent}")
        print(f"   - Question Timeout: {processor.question_timeout}s (15 minutes)")
        print(f"   - Progress Interval: {processor.progress_interval}s")
        print(f"   - Expected Duration: ~{len(all_questions) * 3 // processor.max_concurrent // 60} minutes")
        
        # Confirm before starting
        print(f"\n⚠️  This will process ALL {len(all_questions)} questions concurrently.")
        print(f"📊 Estimated time: {len(all_questions) * 3 // processor.max_concurrent} minutes")
        print(f"🔄 Starting comprehensive accuracy test...")
        print()
        
        # Process all questions
        start_time = datetime.now()
        results = await processor.process_questions_batch(
            all_questions, 
            solver_kwargs={
                "use_kluster": True, 
                "kluster_model": "qwen3-235b"
            }
        )
        end_time = datetime.now()
        
        # Comprehensive results analysis
        print(f"\n" + "=" * 80)
        print(f"🏁 COMPREHENSIVE TEST RESULTS")
        print(f"=" * 80)
        
        duration = (end_time - start_time).total_seconds()
        accuracy = results["accuracy_metrics"]["accuracy_rate"]
        success = results["accuracy_metrics"]["success_rate"]
        
        print(f"⏱️  Total Duration: {int(duration // 60)}m {int(duration % 60)}s")
        print(f"✅ Overall Accuracy: {accuracy:.1%} ({results['accuracy_metrics']['correct_answers']}/{results['completed_questions']})")
        print(f"🎯 Success Rate: {success:.1%} (including partial matches)")
        print(f"⚡ Average per Question: {results['performance_metrics']['average_duration']:.1f}s")
        
        # Detailed breakdown
        print(f"\n📊 DETAILED BREAKDOWN:")
        print(f"  ✅ CORRECT: {results['accuracy_metrics']['correct_answers']}")
        print(f"  🟡 PARTIAL: {results['accuracy_metrics']['partial_answers']}")
        print(f"  ❌ INCORRECT: {results['accuracy_metrics']['incorrect_answers']}")
        print(f"  ⏱️  TIMEOUT: {results['accuracy_metrics']['timeouts']}")
        print(f"  💥 ERROR: {results['accuracy_metrics']['errors']}")
        
        # Classification performance analysis
        print(f"\n🎯 CLASSIFICATION PERFORMANCE:")
        classification_performance = {}
        
        for result in results["detailed_results"]:
            classification = result.classification
            if classification not in classification_performance:
                classification_performance[classification] = {
                    'total': 0, 'correct': 0, 'partial': 0, 'incorrect': 0
                }
            
            classification_performance[classification]['total'] += 1
            if result.status == 'CORRECT':
                classification_performance[classification]['correct'] += 1
            elif result.status == 'PARTIAL':
                classification_performance[classification]['partial'] += 1
            elif result.status == 'INCORRECT':
                classification_performance[classification]['incorrect'] += 1
        
        # Sort by accuracy for prioritization
        sorted_classifications = sorted(
            classification_performance.items(), 
            key=lambda x: (x[1]['correct'] + x[1]['partial'] * 0.5) / x[1]['total'] if x[1]['total'] > 0 else 0
        )
        
        for classification, perf in sorted_classifications:
            total = perf['total']
            if total > 0:
                accuracy_rate = perf['correct'] / total
                success_rate = (perf['correct'] + perf['partial']) / total
                print(f"  {classification:15} | {accuracy_rate:.1%} acc | {success_rate:.1%} success | {total:2d} questions")
        
        # Identify improvement priorities
        print(f"\n🔧 IMPROVEMENT PRIORITIES:")
        improvement_priorities = []
        
        for classification, perf in sorted_classifications:
            total = perf['total']
            if total > 0:
                accuracy_rate = perf['correct'] / total
                impact_score = total * (1 - accuracy_rate)  # Questions * failure rate
                
                if accuracy_rate < 0.7:  # Less than 70% accuracy
                    priority = "HIGH" if impact_score > 2 else "MEDIUM"
                    improvement_priorities.append({
                        'classification': classification,
                        'accuracy': accuracy_rate,
                        'total_questions': total,
                        'impact_score': impact_score,
                        'priority': priority
                    })
        
        for priority_item in sorted(improvement_priorities, key=lambda x: x['impact_score'], reverse=True):
            classification = priority_item['classification']
            accuracy = priority_item['accuracy']
            total = priority_item['total_questions']
            priority = priority_item['priority']
            impact = priority_item['impact_score']
            
            print(f"  🔥 {priority:6} | {classification:15} | {accuracy:.1%} accuracy | {total} questions | Impact: {impact:.1f}")
        
        # Save detailed results
        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
        results_file = f"logs/comprehensive_accuracy_test_{timestamp}.json"
        
        with open(results_file, 'w') as f:
            json.dump({
                'test_metadata': {
                    'timestamp': timestamp,
                    'total_questions': len(all_questions),
                    'duration_seconds': duration,
                    'configuration': {
                        'max_concurrent': processor.max_concurrent,
                        'question_timeout': processor.question_timeout,
                        'model': 'qwen3-235b'
                    }
                },
                'overall_metrics': results['accuracy_metrics'],
                'classification_performance': classification_performance,
                'improvement_priorities': improvement_priorities,
                'detailed_results': [
                    {
                        'task_id': r.task_id,
                        'classification': r.classification,
                        'status': r.status,
                        'accuracy_score': r.accuracy_score,
                        'our_answer': r.our_answer,
                        'expected_answer': r.expected_answer,
                        'duration': r.total_duration,
                        'error_type': r.error_type
                    } for r in results['detailed_results']
                ]
            }, f, indent=2)
        
        print(f"\n📁 Detailed results saved to: {results_file}")
        
        # Summary and next steps
        print(f"\n🎯 NEXT STEPS RECOMMENDATION:")
        if accuracy >= 0.9:
            print(f"  🏆 EXCELLENT: {accuracy:.1%} accuracy achieved! Focus on edge cases.")
        elif accuracy >= 0.7:
            print(f"  ✅ GOOD: {accuracy:.1%} accuracy. Target specific classifications for 90%+.")
        elif accuracy >= 0.5:
            print(f"  🔧 MODERATE: {accuracy:.1%} accuracy. Implement targeted improvements.")
        else:
            print(f"  🚨 NEEDS WORK: {accuracy:.1%} accuracy. Focus on high-impact areas.")
        
        if improvement_priorities:
            top_priority = improvement_priorities[0]
            print(f"  🎯 TOP PRIORITY: {top_priority['classification']} ({top_priority['accuracy']:.1%} accuracy, {top_priority['total_questions']} questions)")
        
        return results
        
    except Exception as e:
        print(f"❌ Comprehensive test failed: {e}")
        import traceback
        traceback.print_exc()
        return None


async def main():
    """Run the comprehensive accuracy test"""
    results = await run_comprehensive_accuracy_test()
    
    if results:
        accuracy = results["accuracy_metrics"]["accuracy_rate"]
        print(f"\n🎉 Comprehensive accuracy test completed!")
        print(f"📊 Final Accuracy: {accuracy:.1%}")
        
        if accuracy >= 0.7:
            print(f"🎯 TARGET ACHIEVED: 70%+ accuracy reached!")
        else:
            gap = 0.7 - accuracy
            print(f"🔧 GAP TO TARGET: {gap:.1%} improvement needed for 70%")


if __name__ == "__main__":
    asyncio.run(main())