from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.ensemble import RandomForestClassifier from sklearn.pipeline import Pipeline from sklearn.model_selection import train_test_split, cross_val_score from sklearn.metrics import classification_report, confusion_matrix import numpy as np import pandas as pd import joblib from datetime import datetime class MisconceptionClassifier: def __init__(self): self.pipeline = Pipeline([ ('tfidf', TfidfVectorizer( max_features=1000, ngram_range=(1, 2), # Include bigrams stop_words='english' )), ('classifier', RandomForestClassifier( n_estimators=100, max_depth=10, random_state=42 )) ]) self.cv_scores = None self.classification_report = None self.confusion_mat = None self.training_history = [] def train(self, X: pd.Series, y: pd.Series, is_retraining: bool = False): """Train the model with cross-validation""" # Train the model self.pipeline.fit(X, y) # Get class distribution class_counts = y.value_counts() min_class_size = class_counts.min() n_classes = len(class_counts) # Initialize empty metrics self.cv_scores = np.array([]) # Only perform cross-validation if we have enough samples if min_class_size >= 3 and len(X) >= n_classes * 3: try: # Use number of splits based on minimum class size n_splits = min(3, min_class_size - 1) self.cv_scores = cross_val_score( self.pipeline, X, y, cv=n_splits, scoring='accuracy' ) except Exception as e: print(f"Cross-validation failed: {str(e)}") # Continue without CV scores # Generate classification metrics y_pred = self.pipeline.predict(X) self.classification_report = classification_report(y, y_pred, output_dict=True) self.confusion_mat = confusion_matrix(y, y_pred) # Record training history history_entry = { 'timestamp': datetime.now().isoformat(), 'samples': len(X), 'class_distribution': class_counts.to_dict(), 'metrics': self.get_model_metrics(), 'is_retraining': is_retraining } self.training_history.append(history_entry) def predict(self, X: pd.Series) -> np.ndarray: """Make predictions""" return self.pipeline.predict(X) def predict_proba(self, X: pd.Series) -> np.ndarray: """Get prediction probabilities""" return self.pipeline.predict_proba(X) def get_feature_importance(self) -> dict: """Get feature importance scores""" tfidf = self.pipeline.named_steps['tfidf'] clf = self.pipeline.named_steps['classifier'] feature_importance = dict(zip( tfidf.get_feature_names_out(), clf.feature_importances_ )) return feature_importance def get_model_metrics(self) -> dict: """Get model performance metrics""" metrics = { 'classification_report': self.classification_report, 'confusion_matrix': self.confusion_mat.tolist() if self.confusion_mat is not None else None } if self.cv_scores is not None and len(self.cv_scores) > 0: metrics.update({ 'cv_scores': self.cv_scores.tolist(), 'cv_mean': self.cv_scores.mean(), 'cv_std': self.cv_scores.std() }) return metrics def save_model(self, path: str = 'models/current_model.joblib'): """Save the current model to disk""" # Create models directory if it doesn't exist import os os.makedirs(os.path.dirname(path), exist_ok=True) joblib.dump(self, path) @classmethod def load_model(cls, path: str = 'models/current_model.joblib') -> 'MisconceptionClassifier': """Load a saved model from disk""" try: return joblib.load(path) except: return cls() def get_training_history(self) -> list: """Get the model's training history""" return self.training_history