pipeline_classes/train_model.py

import pandas as pd
import numpy as np
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.model_selection import StratifiedGroupKFold
from skopt import BayesSearchCV
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC
from xgboost import XGBClassifier
import joblib
from skopt.space import Real, Integer, Categorical
from sklearn.metrics import classification_report, accuracy_score
import json
from sklearn.preprocessing import LabelEncoder
#from _config import config   

class TrainModel(BaseEstimator, TransformerMixin):
    def __init__(self, classifier, train_label, target):
        #self.config = config
        #self.target = config.get("target_label", None)  # User-defined target label in config
        self.classifier = classifier
        self.train_label = train_label
        self.target = target
        self.label_encoder = LabelEncoder()
        #self.selected_domains = self.config.get("selected_domains", "All domains")  # Default to all domains if None

        #if not self.target:
        #    raise ValueError("No target label specified in the config. Please set 'target_label'.")

    def get_default_param_space(self, classifier):
        """ Returns the default hyperparameter space for a given classifier. """
        if classifier == 'xgboost':
            return {
                'learning_rate': Real(0.01, 0.3, prior='log-uniform'),
                'n_estimators': Integer(100, 1000),
                'max_depth': Integer(3, 10),
                'min_child_weight': (1, 10),
                'subsample': (0.5, 1.0),
                'colsample_bytree': (0.5, 1.0),
                'gamma': (0, 10),
                'reg_alpha': (0, 10),
                'reg_lambda': (0, 10),
            }
        elif classifier == 'svm':
            return {
                'C': Real(0.1, 10, prior='log-uniform'),
                'kernel': Categorical(['linear', 'rbf'])
            }
        elif classifier == 'randomforest':
            return {
                'n_estimators': Integer(100, 1000),
                'max_depth': Integer(3, 10)
            }
        else:
            raise ValueError(f"Unsupported classifier type: {classifier}")

    def fit(self, X, y=None):
        # Ensure the target column exists in the dataset
        if self.target not in X.columns:
            raise ValueError(f"Target label '{self.target}' not found in the dataset.")
        
        # Fit the label encoder on the target column
        print(f"Encoding the target labels for '{self.target}'...")
        self.label_encoder.fit(X[self.target])

        # Print the mapping between original labels and encoded labels
        original_labels = list(self.label_encoder.classes_)
        encoded_labels = list(range(len(original_labels)))
        label_mapping = dict(zip(encoded_labels, original_labels))
        print(f"Label encoding complete. Mapping: {label_mapping}")

        # Transform the target column and add it as 'encoded_target'
        X['encoded_target'] = self.label_encoder.transform(X[self.target])

        # Value counts for the encoded target        
        value_counts = X['encoded_target'].value_counts().to_dict()
        print(f"Value counts for encoded target: {value_counts}")
        print(X.columns)
        # Pop unnecessary columns (groupid, emotion labels not being used, etc.)
        groups = X.pop('groupid')
        print(f"Group IDs popped from the dataset.")
        # Pop the label columns which aren't used

        self.train_label = self.train_label.split(",")
        for label in self.train_label:
            X.pop(label)

        print(f"Label columns popped from the dataset.")
        # Pop the encoded target as Y
        y = X.pop('encoded_target')
        print(f"Encoded target column popped from the dataset.")
        print(X.columns)

        # Store the feature names for later use
        feature_names = X.columns.tolist()
        print(f"hallo")
        # Choose classifier
        classifier = self.classifier
        if classifier == 'xgboost':
            model = XGBClassifier(objective='multi:softmax', random_state=42)
        elif classifier == 'svm':
            model = SVC(probability=True)
        elif classifier == 'randomforest':
            model = RandomForestClassifier(random_state=42)
        else:
            raise ValueError(f"Unsupported classifier type: {classifier}")

        print(f"Training the model using {classifier}...")

        # Use user-defined param_space if provided, otherwise use default
        print(f"Classifier: {classifier}")
        default_param_space = self.get_default_param_space(classifier)
        param_space = default_param_space     

        # Hyperparameter tuning using Bayesian optimization
        sgkf = StratifiedGroupKFold(n_splits=5)
        print(f"Parameter space being used: {param_space}")
        if param_space is None:
            raise ValueError("Parameter space cannot be None. Please check the classifier configuration.")

        opt = BayesSearchCV(
            estimator=model,
            search_spaces=param_space,
            cv=sgkf,
            n_iter=5,
            n_jobs=-1,
            n_points=1,
            verbose=1,
            scoring='accuracy'
        )

        print("Hyperparameter tuning in progress...")
        print(X.describe(),X.columns)
        print(f"stop")

        # Fit the model using the encoded target
        opt.fit(X, y, groups=groups)
        self.best_model = opt.best_estimator_
        print(f"Best parameters found: {opt.best_params_}")

        # Print classification metrics
        y_pred = self.best_model.predict(X)
        accuracy = accuracy_score(y, y_pred)
        report = classification_report(y, y_pred, target_names=self.label_encoder.classes_, output_dict=True)
        
        # Save classification report
        classification_report_json = report  
        with open(f'classification_report_{self.target}.json', 'w') as f:
            json.dump(classification_report_json, f, indent=4)

        print(f"Accuracy: {accuracy}")
        print(f"Classification Report:\n{report}")

        # Save the best model with the target label in the file name
        model_name = f"{classifier}_best_model_{self.target}.pkl"
        joblib.dump(self.best_model, model_name)
        print("Model saved successfully.")

        # Save model metadata
        model_metadata = {
            "best_params": opt.best_params_,
            "accuracy": accuracy,
            "classification_report": classification_report_json,
            "label_mapping": label_mapping,
            "model_name": model_name,
            "value_counts": value_counts,
            #"selected_domains": self.selected_domains,  
            #"include_magnitude": self.config.get("include_magnitude", True)      
        }

        if hasattr(self.best_model, "feature_importances_"):
            feature_importances = self.best_model.feature_importances_
            # Convert feature importances to native Python floats
            feature_importance_dict = {feature: float(importance) for feature, importance in zip(feature_names, feature_importances)}
            model_metadata["feature_importances"] = feature_importance_dict
            print("Feature Importances:")
            for feature, importance in feature_importance_dict.items():
                print(f"{feature}: {importance:.4f}")

        # Save metadata with the target name in the file name
        metadata_file = f"{classifier}_model_metadata_{self.target}.json"
        with open(metadata_file, "w") as f:
            json.dump(model_metadata, f, indent=4)
            print(f"Model metadata saved to {metadata_file}.")

        # Save file paths internally for later retrieval
        self.model_file = f"{classifier}_best_model_{self.target}.pkl"
        self.metadata_file = f"{classifier}_model_metadata_{self.target}.json"

        return self
    
    def get_output_files(self):
        return self.model_file, self.metadata_file

    def transform(self, X):
        return X  # Placeholder for transform step (not needed for training)