isa-model 0.3.91__py3-none-any.whl → 0.4.3__py3-none-any.whl

isa_model/inference/legacy_services/model_training.py

@@ -0,0 +1,561 @@
+"""
+Model Training Service - Step 1 of Model Pipeline
+Handles ML model training using existing model processors
+"""
+
+import pandas as pd
+from typing import Dict, List, Any, Optional
+import logging
+from dataclasses import dataclass, field
+from datetime import datetime
+
+# Lazy import to avoid mutex locks from ML libraries
+def _get_ml_processor():
+    from tools.services.data_analytics_service.processors.data_processors.model import MLProcessor
+    return MLProcessor
+
+def _get_time_series_processor():
+    from tools.services.data_analytics_service.processors.data_processors.model import TimeSeriesProcessor
+    return TimeSeriesProcessor
+
+logger = logging.getLogger(__name__)
+
+@dataclass
+class TrainingConfig:
+    """Configuration for model training"""
+    algorithm: str
+    problem_type: Optional[str] = None  # auto-detect if None
+    hyperparameters: Dict[str, Any] = field(default_factory=dict)
+    cross_validation: bool = True
+    cv_folds: int = 5
+    test_size: float = 0.2
+    preprocessing_options: Dict[str, Any] = field(default_factory=dict)
+    feature_selection: bool = False
+    feature_selection_method: str = "auto"
+
+@dataclass
+class TrainingResult:
+    """Result of model training step"""
+    success: bool
+    model_info: Optional[Dict[str, Any]] = None
+    training_metrics: Dict[str, Any] = field(default_factory=dict)
+    model_metadata: Dict[str, Any] = field(default_factory=dict)
+    performance_metrics: Dict[str, Any] = field(default_factory=dict)
+    warnings: List[str] = field(default_factory=list)
+    errors: List[str] = field(default_factory=list)
+
+class ModelTrainingService:
+    """
+    Model Training Service - Step 1 of Model Pipeline
+    
+    Handles:
+    - ML model training using various algorithms
+    - Problem type detection and algorithm recommendation
+    - Hyperparameter tuning
+    - Cross-validation and model selection
+    """
+    
+    def __init__(self):
+        self.execution_stats = {
+            'total_training_operations': 0,
+            'successful_training_operations': 0,
+            'failed_training_operations': 0,
+            'models_trained': 0,
+            'average_training_time': 0.0
+        }
+        
+        # Track trained models
+        self.trained_models = {}
+        
+        logger.info("Model Training Service initialized")
+    
+    def train_model(self,
+                   data: pd.DataFrame,
+                   target_column: str,
+                   training_config: TrainingConfig) -> TrainingResult:
+        """
+        Train a machine learning model
+        
+        Args:
+            data: Training dataset
+            target_column: Target variable column name
+            training_config: Training configuration
+            
+        Returns:
+            TrainingResult with model information and metrics
+        """
+        start_time = datetime.now()
+        
+        try:
+            logger.info(f"Starting model training: {training_config.algorithm}")
+            
+            # Initialize result
+            result = TrainingResult(
+                success=False,
+                model_metadata={
+                    'start_time': start_time,
+                    'target_column': target_column,
+                    'algorithm': training_config.algorithm,
+                    'data_shape': data.shape
+                }
+            )
+            
+            # Validate inputs
+            validation_result = self._validate_training_inputs(data, target_column, training_config)
+            if not validation_result['valid']:
+                result.errors.extend(validation_result['errors'])
+                return self._finalize_training_result(result, start_time)
+            
+            # Select appropriate processor based on algorithm/problem type
+            processor_result = self._get_model_processor(data, target_column, training_config)
+            if not processor_result['success']:
+                result.errors.extend(processor_result['errors'])
+                return self._finalize_training_result(result, start_time)
+            
+            processor = processor_result['processor']
+            detected_problem_type = processor_result['problem_type']
+            
+            # Execute training
+            training_result = self._execute_training(
+                processor, data, target_column, training_config, detected_problem_type
+            )
+            
+            if not training_result['success']:
+                result.errors.extend(training_result['errors'])
+                return self._finalize_training_result(result, start_time)
+            
+            # Store model information
+            model_id = training_result['model_id']
+            self.trained_models[model_id] = {
+                'processor': processor,
+                'model_instance': training_result.get('model_instance'),
+                'training_config': training_config,
+                'target_column': target_column,
+                'problem_type': detected_problem_type,
+                'created_at': start_time
+            }
+            
+            # Success
+            result.success = True
+            result.model_info = {
+                'model_id': model_id,
+                'algorithm': training_config.algorithm,
+                'problem_type': detected_problem_type
+            }
+            result.training_metrics = training_result.get('training_metrics', {})
+            result.model_metadata.update(training_result.get('metadata', {}))
+            
+            return self._finalize_training_result(result, start_time)
+            
+        except Exception as e:
+            logger.error(f"Model training failed: {e}")
+            result.errors.append(f"Training error: {str(e)}")
+            return self._finalize_training_result(result, start_time)
+    
+    def train_multiple_models(self,
+                             data: pd.DataFrame,
+                             target_column: str,
+                             algorithm_configs: List[TrainingConfig]) -> Dict[str, TrainingResult]:
+        """Train multiple models for comparison"""
+        results = {}
+        
+        for i, config in enumerate(algorithm_configs):
+            algorithm_name = config.algorithm
+            logger.info(f"Training model {i+1}/{len(algorithm_configs)}: {algorithm_name}")
+            
+            try:
+                result = self.train_model(data, target_column, config)
+                results[algorithm_name] = result
+            except Exception as e:
+                logger.error(f"Failed to train {algorithm_name}: {e}")
+                results[algorithm_name] = TrainingResult(
+                    success=False,
+                    errors=[str(e)]
+                )
+        
+        return results
+    
+    def get_algorithm_recommendations(self,
+                                    data: pd.DataFrame,
+                                    target_column: str) -> Dict[str, Any]:
+        """Get ML algorithm recommendations for the dataset"""
+        try:
+            # Use MLProcessor for comprehensive analysis
+            MLProcessor = _get_ml_processor()
+            MLProcessor = _get_ml_processor()
+            processor = MLProcessor(csv_processor=None, file_path=None)
+            processor.df = data
+            
+            # Get ML analysis and recommendations
+            analysis = processor.get_ml_analysis(target_column)
+            
+            if 'error' in analysis:
+                return {'success': False, 'error': analysis['error']}
+            
+            return {
+                'success': True,
+                'problem_analysis': analysis.get('problem_analysis', {}),
+                'algorithm_recommendations': analysis.get('algorithm_recommendations', {}),
+                'preprocessing_recommendations': analysis.get('preprocessing_recommendations', {}),
+                'data_preparation_needs': analysis.get('data_preparation', {})
+            }
+            
+        except Exception as e:
+            logger.error(f"Algorithm recommendation failed: {e}")
+            return {'success': False, 'error': str(e)}
+    
+    def tune_hyperparameters(self,
+                           data: pd.DataFrame,
+                           target_column: str,
+                           algorithm: str,
+                           param_grid: Optional[Dict[str, Any]] = None,
+                           search_type: str = "grid",
+                           cv_folds: int = 5) -> TrainingResult:
+        """Perform hyperparameter tuning for a specific algorithm"""
+        start_time = datetime.now()
+        
+        try:
+            # Create processor
+            MLProcessor = _get_ml_processor()
+            processor = MLProcessor(csv_processor=None, file_path=None)
+            processor.df = data
+            
+            # Perform hyperparameter tuning
+            tuning_result = processor.hyperparameter_tuning(
+                target_column=target_column,
+                algorithm=algorithm,
+                param_grid=param_grid,
+                search_type=search_type,
+                cv_folds=cv_folds
+            )
+            
+            if 'error' in tuning_result:
+                return TrainingResult(
+                    success=False,
+                    errors=[tuning_result['error']]
+                )
+            
+            # Store tuned model
+            model_id = tuning_result['model_id']
+            self.trained_models[model_id] = {
+                'processor': processor,
+                'model_instance': processor.models.get(model_id),
+                'training_config': TrainingConfig(
+                    algorithm=algorithm,
+                    hyperparameters=tuning_result['best_parameters']
+                ),
+                'target_column': target_column,
+                'problem_type': 'auto_detected',
+                'created_at': start_time,
+                'is_tuned': True
+            }
+            
+            result = TrainingResult(
+                success=True,
+                model_info={
+                    'model_id': model_id,
+                    'algorithm': algorithm,
+                    'is_hyperparameter_tuned': True
+                },
+                training_metrics=tuning_result,
+                model_metadata={
+                    'tuning_method': search_type,
+                    'cv_folds': cv_folds,
+                    'best_parameters': tuning_result['best_parameters'],
+                    'best_score': tuning_result['best_score']
+                }
+            )
+            
+            return self._finalize_training_result(result, start_time)
+            
+        except Exception as e:
+            logger.error(f"Hyperparameter tuning failed: {e}")
+            return TrainingResult(
+                success=False,
+                errors=[str(e)]
+            )
+    
+    def _validate_training_inputs(self,
+                                 data: pd.DataFrame,
+                                 target_column: str,
+                                 config: TrainingConfig) -> Dict[str, Any]:
+        """Validate training inputs"""
+        errors = []
+        
+        # Check data
+        if data.empty:
+            errors.append("Training data is empty")
+        
+        # Check target column
+        if target_column not in data.columns:
+            errors.append(f"Target column '{target_column}' not found in data")
+        
+        # Check algorithm
+        supported_algorithms = [
+            'logistic_regression', 'linear_regression', 'ridge', 'lasso',
+            'random_forest_classifier', 'random_forest_regressor',
+            'decision_tree_classifier', 'decision_tree_regressor',
+            'svm_classifier', 'svm_regressor', 'knn_classifier', 'knn_regressor',
+            'naive_bayes', 'xgboost_classifier', 'xgboost_regressor', 'prophet'
+        ]
+        
+        if config.algorithm.lower() not in supported_algorithms:
+            errors.append(f"Unsupported algorithm: {config.algorithm}")
+        
+        # Check test size
+        if not 0 < config.test_size < 1:
+            errors.append("Test size must be between 0 and 1")
+        
+        return {
+            'valid': len(errors) == 0,
+            'errors': errors
+        }
+    
+    def _get_model_processor(self,
+                           data: pd.DataFrame,
+                           target_column: str,
+                           config: TrainingConfig) -> Dict[str, Any]:
+        """Get appropriate model processor based on algorithm and problem type"""
+        try:
+            algorithm = config.algorithm.lower()
+            
+            # Check if it's a time series problem
+            if algorithm == 'prophet' or 'ds' in data.columns:
+                try:
+                    TimeSeriesProcessor = _get_time_series_processor()
+                    processor = TimeSeriesProcessor(csv_processor=None, file_path=None)
+                except Exception:
+                    # Use MLProcessor for time series if TimeSeriesProcessor not available
+                    MLProcessor = _get_ml_processor()
+                    processor = MLProcessor(csv_processor=None, file_path=None)
+                    processor.df = data
+                    return {
+                        'success': True,
+                        'processor': processor,
+                        'problem_type': 'time_series'
+                    }
+                processor.df = data
+                return {
+                    'success': True,
+                    'processor': processor,
+                    'problem_type': 'time_series'
+                }
+            
+            # MLProcessor causes mutex locks due to complex dependency chains
+            # Use lightweight DataFrameMLProcessor instead
+            class DataFrameMLProcessor:
+                """Lightweight ML processor that works directly with DataFrames"""
+                def __init__(self, dataframe):
+                    self.df = dataframe
+                    
+                def train_model(self, target_column, algorithm, test_size=0.2, **hyperparameters):
+                    """Train model using sklearn directly"""
+                    try:
+                        from sklearn.model_selection import train_test_split
+                        from sklearn.metrics import accuracy_score, r2_score, mean_squared_error
+                        from sklearn.linear_model import LogisticRegression, LinearRegression
+                        from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
+                        from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
+                        import time
+                        
+                        start_time = time.time()
+                        
+                        # Prepare data
+                        X = self.df.drop(columns=[target_column])
+                        y = self.df[target_column]
+                        
+                        # Select numeric features only
+                        numeric_features = X.select_dtypes(include=['int64', 'float64']).columns
+                        X = X[numeric_features]
+                        
+                        if len(X.columns) == 0:
+                            return {'error': 'No numeric features available for training'}
+                        
+                        # Split data
+                        X_train, X_test, y_train, y_test = train_test_split(
+                            X, y, test_size=test_size, random_state=42
+                        )
+                        
+                        # Detect problem type
+                        if y.dtype in ['int64', 'float64'] and y.nunique() > 10:
+                            problem_type = 'regression'
+                        else:
+                            problem_type = 'classification'
+                        
+                        # Select and train model
+                        if problem_type == 'classification':
+                            if algorithm.lower() in ['logistic_regression', 'logistic']:
+                                model = LogisticRegression(random_state=42, max_iter=1000, **hyperparameters)
+                            elif algorithm.lower() in ['random_forest', 'rf']:
+                                model = RandomForestClassifier(random_state=42, n_estimators=100, **hyperparameters)
+                            elif algorithm.lower() in ['decision_tree', 'tree']:
+                                model = DecisionTreeClassifier(random_state=42, **hyperparameters)
+                            else:
+                                model = LogisticRegression(random_state=42, max_iter=1000)
+                        else:  # regression
+                            if algorithm.lower() in ['linear_regression', 'linear']:
+                                model = LinearRegression(**hyperparameters)
+                            elif algorithm.lower() in ['random_forest', 'rf']:
+                                model = RandomForestRegressor(random_state=42, n_estimators=100, **hyperparameters)
+                            elif algorithm.lower() in ['decision_tree', 'tree']:
+                                model = DecisionTreeRegressor(random_state=42, **hyperparameters)
+                            else:
+                                model = LinearRegression()
+                        
+                        # Train
+                        model.fit(X_train, y_train)
+                        
+                        # Evaluate
+                        y_pred = model.predict(X_test)
+                        if problem_type == 'classification':
+                            score = accuracy_score(y_test, y_pred)
+                            metrics = {
+                                'accuracy': float(score),
+                                'test_samples': len(y_test)
+                            }
+                        else:
+                            r2 = r2_score(y_test, y_pred)
+                            mse = mean_squared_error(y_test, y_pred)
+                            metrics = {
+                                'r2_score': float(r2),
+                                'mse': float(mse),
+                                'test_samples': len(y_test)
+                            }
+                        
+                        training_time = time.time() - start_time
+                        
+                        # Generate model ID
+                        from datetime import datetime
+                        model_id = f'{algorithm}_{int(datetime.now().timestamp())}'
+                        
+                        return {
+                            'model_id': model_id,
+                            'model_instance': model,
+                            'problem_type': problem_type,
+                            'metrics': metrics,
+                            'features_used': list(X.columns),
+                            'n_samples': len(X),
+                            'train_size': len(X_train),
+                            'test_size': len(X_test),
+                            'training_time_seconds': training_time,
+                            'algorithm': algorithm
+                        }
+                        
+                    except Exception as e:
+                        return {'error': f'Model training failed: {str(e)}'}
+            
+            processor = DataFrameMLProcessor(data)
+            
+            return {
+                'success': True,
+                'processor': processor,
+                'problem_type': 'classification'  # Will be detected during training
+            }
+            
+        except Exception as e:
+            return {
+                'success': False,
+                'errors': [f"Processor initialization failed: {str(e)}"]
+            }
+    
+    def _execute_training(self,
+                         processor,
+                         data: pd.DataFrame,
+                         target_column: str,
+                         config: TrainingConfig,
+                         problem_type: str) -> Dict[str, Any]:
+        """Execute the actual model training"""
+        try:
+            if hasattr(processor, 'train_model'):
+                # Use processor's train_model method
+                training_result = processor.train_model(
+                    target_column=target_column,
+                    algorithm=config.algorithm,
+                    test_size=config.test_size,
+                    **config.hyperparameters
+                )
+                
+                if 'error' in training_result:
+                    return {
+                        'success': False,
+                        'errors': [training_result['error']]
+                    }
+                
+                return {
+                    'success': True,
+                    'model_id': training_result['model_id'],
+                    'training_metrics': training_result,
+                    'metadata': {
+                        'algorithm': config.algorithm,
+                        'problem_type': problem_type,
+                        'training_time': training_result.get('training_time_seconds', 0)
+                    }
+                }
+            else:
+                return {
+                    'success': False,
+                    'errors': [f"Processor does not support training method"]
+                }
+                
+        except Exception as e:
+            return {
+                'success': False,
+                'errors': [f"Training execution failed: {str(e)}"]
+            }
+    
+    def _finalize_training_result(self,
+                                 result: TrainingResult,
+                                 start_time: datetime) -> TrainingResult:
+        """Finalize training result with timing and stats"""
+        end_time = datetime.now()
+        duration = (end_time - start_time).total_seconds()
+        
+        # Update performance metrics
+        result.performance_metrics['training_duration_seconds'] = duration
+        result.performance_metrics['end_time'] = end_time
+        result.model_metadata['end_time'] = end_time
+        result.model_metadata['duration_seconds'] = duration
+        
+        # Update execution stats
+        self.execution_stats['total_training_operations'] += 1
+        if result.success:
+            self.execution_stats['successful_training_operations'] += 1
+            self.execution_stats['models_trained'] += 1
+        else:
+            self.execution_stats['failed_training_operations'] += 1
+        
+        # Update average duration
+        total = self.execution_stats['total_training_operations']
+        old_avg = self.execution_stats['average_training_time']
+        self.execution_stats['average_training_time'] = (old_avg * (total - 1) + duration) / total
+        
+        logger.info(f"Training completed: success={result.success}, duration={duration:.2f}s")
+        return result
+    
+    def get_trained_model(self, model_id: str) -> Optional[Dict[str, Any]]:
+        """Get information about a trained model"""
+        return self.trained_models.get(model_id)
+    
+    def list_trained_models(self) -> List[Dict[str, Any]]:
+        """List all trained models"""
+        models_info = []
+        for model_id, model_data in self.trained_models.items():
+            models_info.append({
+                'model_id': model_id,
+                'algorithm': model_data.get('training_config').algorithm,
+                'problem_type': model_data.get('problem_type'),
+                'target_column': model_data.get('target_column'),
+                'created_at': model_data.get('created_at'),
+                'is_tuned': model_data.get('is_tuned', False)
+            })
+        return models_info
+    
+    def get_execution_stats(self) -> Dict[str, Any]:
+        """Get service execution statistics"""
+        return {
+            **self.execution_stats,
+            'success_rate': (
+                self.execution_stats['successful_training_operations'] / 
+                max(1, self.execution_stats['total_training_operations'])
+            )
+        }

isa_model/inference/models/__init__.py

@@ -0,0 +1,21 @@
+"""
+Inference Models
+
+Data models for inference operations following the ISA Model architecture pattern.
+"""
+
+from .inference_record import InferenceRequest, UsageStatistics, ModelUsageSnapshot
+from .inference_config import InferenceConfig, ProviderConfig, ModelConfig
+from .performance_models import PerformanceMetrics, LatencyProfile, ThroughputProfile
+
+__all__ = [
+    "InferenceRequest",
+    "UsageStatistics", 
+    "ModelUsageSnapshot",
+    "InferenceConfig",
+    "ProviderConfig",
+    "ModelConfig",
+    "PerformanceMetrics",
+    "LatencyProfile",
+    "ThroughputProfile"
+]