evaluate_model.py

"""
Model Evaluation Script for Troviku-1.1

Comprehensive evaluation suite for testing the model's performance
on various coding benchmarks and tasks.
"""

import json
import time
from typing import List, Dict, Any, Optional, Tuple
from dataclasses import dataclass, asdict
from collections import defaultdict
import statistics


@dataclass
class EvaluationResult:
    """Result from a single evaluation."""
    task_id: str
    task_type: str
    language: str
    passed: bool
    score: float
    execution_time: float
    error_message: Optional[str] = None
    
    def to_dict(self) -> Dict[str, Any]:
        return asdict(self)


@dataclass
class BenchmarkResults:
    """Aggregated benchmark results."""
    benchmark_name: str
    total_tasks: int
    passed_tasks: int
    failed_tasks: int
    average_score: float
    pass_rate: float
    average_execution_time: float
    results_by_language: Dict[str, Dict[str, float]]
    
    def to_dict(self) -> Dict[str, Any]:
        return asdict(self)


class CodeEvaluator:
    """
    Evaluator for Troviku-1.1 model performance.
    
    Runs various benchmarks and coding tasks to assess model capabilities.
    """
    
    def __init__(self, api_key: str, model: str = "OpenTrouter/Troviku-1.1"):
        """
        Initialize the evaluator.
        
        Args:
            api_key: OpenTrouter API key
            model: Model identifier to evaluate
        """
        from troviku_client import TrovikuClient
        
        self.client = TrovikuClient(api_key=api_key, model=model)
        self.results: List[EvaluationResult] = []
    
    def evaluate_humaneval(self, problems: List[Dict[str, Any]]) -> BenchmarkResults:
        """
        Evaluate on HumanEval benchmark.
        
        Args:
            problems: List of HumanEval problems
            
        Returns:
            BenchmarkResults with aggregated scores
        """
        print("Evaluating HumanEval benchmark...")
        
        for problem in problems:
            task_id = problem['task_id']
            prompt = problem['prompt']
            test_cases = problem['test']
            
            try:
                start_time = time.time()
                response = self.client.generate(prompt, language="python")
                execution_time = time.time() - start_time
                
                # Execute test cases
                passed, error = self._execute_tests(response.code, test_cases)
                
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="code_generation",
                    language="python",
                    passed=passed,
                    score=1.0 if passed else 0.0,
                    execution_time=execution_time,
                    error_message=error
                )
                
                self.results.append(result)
                print(f"  {task_id}: {'PASS' if passed else 'FAIL'}")
                
            except Exception as e:
                print(f"  {task_id}: ERROR - {str(e)}")
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="code_generation",
                    language="python",
                    passed=False,
                    score=0.0,
                    execution_time=0.0,
                    error_message=str(e)
                )
                self.results.append(result)
        
        return self._aggregate_results("HumanEval")
    
    def evaluate_mbpp(self, problems: List[Dict[str, Any]]) -> BenchmarkResults:
        """
        Evaluate on MBPP (Mostly Basic Python Problems) benchmark.
        
        Args:
            problems: List of MBPP problems
            
        Returns:
            BenchmarkResults with aggregated scores
        """
        print("Evaluating MBPP benchmark...")
        
        for problem in problems:
            task_id = str(problem['task_id'])
            prompt = problem['text']
            test_cases = problem['test_list']
            
            try:
                start_time = time.time()
                response = self.client.generate(prompt, language="python")
                execution_time = time.time() - start_time
                
                passed, error = self._execute_tests(response.code, test_cases)
                
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="code_generation",
                    language="python",
                    passed=passed,
                    score=1.0 if passed else 0.0,
                    execution_time=execution_time,
                    error_message=error
                )
                
                self.results.append(result)
                print(f"  Task {task_id}: {'PASS' if passed else 'FAIL'}")
                
            except Exception as e:
                print(f"  Task {task_id}: ERROR - {str(e)}")
        
        return self._aggregate_results("MBPP")
    
    def evaluate_code_translation(
        self, 
        test_cases: List[Dict[str, Any]]
    ) -> BenchmarkResults:
        """
        Evaluate code translation between languages.
        
        Args:
            test_cases: List of translation test cases
            
        Returns:
            BenchmarkResults with translation accuracy
        """
        print("Evaluating code translation...")
        
        for test_case in test_cases:
            task_id = test_case['id']
            source_code = test_case['source_code']
            source_lang = test_case['source_language']
            target_lang = test_case['target_language']
            expected_behavior = test_case.get('expected_behavior')
            
            try:
                start_time = time.time()
                response = self.client.translate(
                    code=source_code,
                    source_language=source_lang,
                    target_language=target_lang
                )
                execution_time = time.time() - start_time
                
                # Validate translation (simplified - would need actual execution)
                score = self._validate_translation(
                    response.code, 
                    target_lang, 
                    expected_behavior
                )
                
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="code_translation",
                    language=f"{source_lang}_to_{target_lang}",
                    passed=score >= 0.8,
                    score=score,
                    execution_time=execution_time
                )
                
                self.results.append(result)
                print(f"  {task_id}: Score {score:.2f}")
                
            except Exception as e:
                print(f"  {task_id}: ERROR - {str(e)}")
        
        return self._aggregate_results("Code Translation")
    
    def evaluate_code_explanation(
        self, 
        test_cases: List[Dict[str, Any]]
    ) -> BenchmarkResults:
        """
        Evaluate code explanation quality.
        
        Args:
            test_cases: List of explanation test cases
            
        Returns:
            BenchmarkResults with explanation scores
        """
        print("Evaluating code explanation...")
        
        for test_case in test_cases:
            task_id = test_case['id']
            code = test_case['code']
            language = test_case['language']
            key_concepts = test_case.get('key_concepts', [])
            
            try:
                start_time = time.time()
                explanation = self.client.explain(code, language)
                execution_time = time.time() - start_time
                
                # Score explanation based on coverage of key concepts
                score = self._score_explanation(explanation, key_concepts)
                
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="code_explanation",
                    language=language,
                    passed=score >= 0.7,
                    score=score,
                    execution_time=execution_time
                )
                
                self.results.append(result)
                print(f"  {task_id}: Score {score:.2f}")
                
            except Exception as e:
                print(f"  {task_id}: ERROR - {str(e)}")
        
        return self._aggregate_results("Code Explanation")
    
    def evaluate_bug_detection(
        self, 
        test_cases: List[Dict[str, Any]]
    ) -> BenchmarkResults:
        """
        Evaluate bug detection and fixing capabilities.
        
        Args:
            test_cases: List of buggy code samples
            
        Returns:
            BenchmarkResults with bug fix success rate
        """
        print("Evaluating bug detection and fixing...")
        
        for test_case in test_cases:
            task_id = test_case['id']
            buggy_code = test_case['buggy_code']
            error_message = test_case['error_message']
            language = test_case['language']
            tests = test_case.get('tests', [])
            
            try:
                start_time = time.time()
                response = self.client.debug(buggy_code, error_message, language)
                execution_time = time.time() - start_time
                
                # Test if fixed code passes tests
                passed, error = self._execute_tests(response.code, tests)
                
                result = EvaluationResult(
                    task_id=task_id,
                    task_type="bug_fixing",
                    language=language,
                    passed=passed,
                    score=1.0 if passed else 0.0,
                    execution_time=execution_time,
                    error_message=error
                )
                
                self.results.append(result)
                print(f"  {task_id}: {'FIXED' if passed else 'FAILED'}")
                
            except Exception as e:
                print(f"  {task_id}: ERROR - {str(e)}")
        
        return self._aggregate_results("Bug Detection")
    
    def _execute_tests(
        self, 
        code: str, 
        test_cases: List[str]
    ) -> Tuple[bool, Optional[str]]:
        """
        Execute test cases against generated code.
        
        Args:
            code: Generated code to test
            test_cases: List of test case strings
            
        Returns:
            Tuple of (passed, error_message)
        """
        try:
            # Create execution environment
            namespace = {}
            exec(code, namespace)
            
            # Run test cases
            for test in test_cases:
                exec(test, namespace)
            
            return True, None
            
        except Exception as e:
            return False, str(e)
    
    def _validate_translation(
        self, 
        translated_code: str, 
        target_language: str,
        expected_behavior: Optional[Dict[str, Any]]
    ) -> float:
        """
        Validate translated code quality.
        
        Args:
            translated_code: Translated code
            target_language: Target language
            expected_behavior: Expected behavior specification
            
        Returns:
            Quality score (0.0 to 1.0)
        """
        # Simplified validation - in practice would need language-specific execution
        score = 0.0
        
        # Check for syntax validity (simplified)
        if len(translated_code.strip()) > 0:
            score += 0.3
        
        # Check for language-specific keywords
        if target_language.lower() in translated_code.lower():
            score += 0.2
        
        # If expected behavior is specified, score higher
        if expected_behavior:
            score += 0.5
        
        return min(score, 1.0)
    
    def _score_explanation(
        self, 
        explanation: str, 
        key_concepts: List[str]
    ) -> float:
        """
        Score explanation quality based on concept coverage.
        
        Args:
            explanation: Generated explanation
            key_concepts: List of key concepts that should be covered
            
        Returns:
            Quality score (0.0 to 1.0)
        """
        if not key_concepts:
            # Base score for reasonable length explanation
            return 0.8 if len(explanation) > 100 else 0.5
        
        explanation_lower = explanation.lower()
        covered = sum(1 for concept in key_concepts 
                     if concept.lower() in explanation_lower)
        
        coverage_score = covered / len(key_concepts)
        length_score = min(len(explanation) / 500, 1.0)
        
        return (coverage_score * 0.7 + length_score * 0.3)
    
    def _aggregate_results(self, benchmark_name: str) -> BenchmarkResults:
        """
        Aggregate evaluation results for a benchmark.
        
        Args:
            benchmark_name: Name of the benchmark
            
        Returns:
            BenchmarkResults with aggregated statistics
        """
        benchmark_results = [r for r in self.results 
                           if benchmark_name.lower() in r.task_id.lower() or
                              benchmark_name.lower() == r.task_type.lower()]
        
        if not benchmark_results:
            return BenchmarkResults(
                benchmark_name=benchmark_name,
                total_tasks=0,
                passed_tasks=0,
                failed_tasks=0,
                average_score=0.0,
                pass_rate=0.0,
                average_execution_time=0.0,
                results_by_language={}
            )
        
        total = len(benchmark_results)
        passed = sum(1 for r in benchmark_results if r.passed)
        failed = total - passed
        avg_score = statistics.mean(r.score for r in benchmark_results)
        pass_rate = passed / total if total > 0 else 0.0
        avg_time = statistics.mean(r.execution_time for r in benchmark_results)
        
        # Aggregate by language
        by_language = defaultdict(lambda: {"passed": 0, "total": 0, "score": []})
        for result in benchmark_results:
            lang = result.language
            by_language[lang]["total"] += 1
            if result.passed:
                by_language[lang]["passed"] += 1
            by_language[lang]["score"].append(result.score)
        
        results_by_language = {
            lang: {
                "pass_rate": stats["passed"] / stats["total"],
                "average_score": statistics.mean(stats["score"])
            }
            for lang, stats in by_language.items()
        }
        
        return BenchmarkResults(
            benchmark_name=benchmark_name,
            total_tasks=total,
            passed_tasks=passed,
            failed_tasks=failed,
            average_score=avg_score,
            pass_rate=pass_rate,
            average_execution_time=avg_time,
            results_by_language=results_by_language
        )
    
    def save_results(self, filepath: str):
        """
        Save evaluation results to JSON file.
        
        Args:
            filepath: Path to save results
        """
        results_data = {
            "individual_results": [r.to_dict() for r in self.results],
            "summary": self.get_summary()
        }
        
        with open(filepath, 'w') as f:
            json.dump(results_data, f, indent=2)
        
        print(f"\nResults saved to {filepath}")
    
    def get_summary(self) -> Dict[str, Any]:
        """
        Get summary of all evaluation results.
        
        Returns:
            Dictionary with summary statistics
        """
        if not self.results:
            return {"message": "No results available"}
        
        total = len(self.results)
        passed = sum(1 for r in self.results if r.passed)
        
        return {
            "total_tasks": total,
            "passed_tasks": passed,
            "failed_tasks": total - passed,
            "overall_pass_rate": passed / total,
            "average_score": statistics.mean(r.score for r in self.results),
            "average_execution_time": statistics.mean(r.execution_time for r in self.results),
            "by_task_type": self._group_by_field("task_type"),
            "by_language": self._group_by_field("language")
        }
    
    def _group_by_field(self, field: str) -> Dict[str, Dict[str, float]]:
        """Group results by a specific field."""
        grouped = defaultdict(lambda: {"passed": 0, "total": 0, "scores": []})
        
        for result in self.results:
            value = getattr(result, field)
            grouped[value]["total"] += 1
            if result.passed:
                grouped[value]["passed"] += 1
            grouped[value]["scores"].append(result.score)
        
        return {
            key: {
                "pass_rate": stats["passed"] / stats["total"],
                "average_score": statistics.mean(stats["scores"])
            }
            for key, stats in grouped.items()
        }
    
    def print_summary(self):
        """Print evaluation summary to console."""
        summary = self.get_summary()
        
        print("\n" + "="*60)
        print("EVALUATION SUMMARY")
        print("="*60)
        print(f"Total Tasks: {summary['total_tasks']}")
        print(f"Passed: {summary['passed_tasks']}")
        print(f"Failed: {summary['failed_tasks']}")
        print(f"Overall Pass Rate: {summary['overall_pass_rate']:.2%}")
        print(f"Average Score: {summary['average_score']:.2f}")
        print(f"Average Execution Time: {summary['average_execution_time']:.2f}s")
        
        print("\nBy Task Type:")
        for task_type, stats in summary['by_task_type'].items():
            print(f"  {task_type}:")
            print(f"    Pass Rate: {stats['pass_rate']:.2%}")
            print(f"    Avg Score: {stats['average_score']:.2f}")
        
        print("\nBy Language:")
        for language, stats in summary['by_language'].items():
            print(f"  {language}:")
            print(f"    Pass Rate: {stats['pass_rate']:.2%}")
            print(f"    Avg Score: {stats['average_score']:.2f}")
        
        print("="*60)


# Example usage
if __name__ == "__main__":
    # Initialize evaluator
    evaluator = CodeEvaluator(api_key="your_api_key_here")
    
    # Example HumanEval problems (simplified)
    humaneval_problems = [
        {
            "task_id": "HumanEval/0",
            "prompt": "Write a function that takes a list of numbers and returns True if the list contains a pair of numbers that sum to zero.",
            "test": "assert has_zero_sum([1, -1, 2]) == True\nassert has_zero_sum([1, 2, 3]) == False"
        }
    ]
    
    # Run evaluation
    results = evaluator.evaluate_humaneval(humaneval_problems)
    
    # Print and save results
    evaluator.print_summary()
    evaluator.save_results("evaluation_results.json")