import streamlit as st import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.tree import plot_tree, export_text import seaborn as sns from sklearn.preprocessing import LabelEncoder from sklearn.ensemble import RandomForestClassifier from sklearn.tree import DecisionTreeClassifier from sklearn.ensemble import GradientBoostingClassifier from sklearn.linear_model import LogisticRegression from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, roc_curve import shap from xgboost import XGBClassifier from lightgbm import LGBMClassifier def load_data(): data = pd.read_csv('exported_named_train_good.csv') data_test = pd.read_csv('exported_named_test_good.csv') X_train = data.drop("Target", axis=1) y_train = data['Target'] X_test = data_test.drop('Target', axis=1) y_test = data_test['Target'] return X_train, y_train, X_test, y_test, X_train.columns def train_models(X_train, y_train, X_test, y_test): models = { "Logistic Regression": LogisticRegression(random_state=42), "Decision Tree": DecisionTreeClassifier(random_state=42), "Random Forest": RandomForestClassifier(random_state=42), "Gradient Boost": GradientBoostingClassifier(random_state=42), "Extreme Gradient Boosting": XGBClassifier(random_state=42), "Light Gradient Boosting": LGBMClassifier(random_state=42) } results = {} for name, model in models.items(): model.fit(X_train, y_train) # Predictions y_train_pred = model.predict(X_train) y_test_pred = model.predict(X_test) # Metrics results[name] = { 'model': model, 'train_metrics': { 'accuracy': accuracy_score(y_train, y_train_pred), 'f1': f1_score(y_train, y_train_pred, average='weighted'), 'precision': precision_score(y_train, y_train_pred), 'recall': recall_score(y_train, y_train_pred), 'roc_auc': roc_auc_score(y_train, y_train_pred) }, 'test_metrics': { 'accuracy': accuracy_score(y_test, y_test_pred), 'f1': f1_score(y_test, y_test_pred, average='weighted'), 'precision': precision_score(y_test, y_test_pred), 'recall': recall_score(y_test, y_test_pred), 'roc_auc': roc_auc_score(y_test, y_test_pred) } } return results def plot_model_performance(results): metrics = ['accuracy', 'f1', 'precision', 'recall', 'roc_auc'] fig, axes = plt.subplots(1, 2, figsize=(15, 6)) # Training metrics train_data = {model: [results[model]['train_metrics'][metric] for metric in metrics] for model in results.keys()} train_df = pd.DataFrame(train_data, index=metrics) train_df.plot(kind='bar', ax=axes[0], title='Training Performance') axes[0].set_ylim(0, 1) # Test metrics test_data = {model: [results[model]['test_metrics'][metric] for metric in metrics] for model in results.keys()} test_df = pd.DataFrame(test_data, index=metrics) test_df.plot(kind='bar', ax=axes[1], title='Test Performance') axes[1].set_ylim(0, 1) plt.tight_layout() return fig def plot_feature_importance(model, feature_names, model_type): plt.figure(figsize=(10, 6)) if model_type in ["Decision Tree", "Random Forest", "Gradient Boost"]: importance = model.feature_importances_ elif model_type == "Logistic Regression": importance = np.abs(model.coef_[0]) importance_df = pd.DataFrame({ 'feature': feature_names, 'importance': importance }).sort_values('importance', ascending=True) plt.barh(importance_df['feature'], importance_df['importance']) plt.title(f"Feature Importance - {model_type}") return plt.gcf() def app(): st.title("Interpréteur de Modèles ML") # Load data X_train, y_train, X_test, y_test, feature_names = load_data() # Train models if not in session state if 'model_results' not in st.session_state: with st.spinner("Entraînement des modèles en cours..."): st.session_state.model_results = train_models(X_train, y_train, X_test, y_test) # Sidebar st.sidebar.title("Navigation") selected_model = st.sidebar.selectbox( "Sélectionnez un modèle", list(st.session_state.model_results.keys()) ) page = st.sidebar.radio( "Sélectionnez une section", ["Performance des modèles", "Interprétation du modèle", "Analyse des caractéristiques", "Simulateur de prédictions"] ) current_model = st.session_state.model_results[selected_model]['model'] # Performance des modèles if page == "Performance des modèles": st.header("Performance des modèles") # Plot global performance comparison st.subheader("Comparaison des performances") performance_fig = plot_model_performance(st.session_state.model_results) st.pyplot(performance_fig) # Detailed metrics for selected model st.subheader(f"Métriques détaillées - {selected_model}") col1, col2 = st.columns(2) with col1: st.write("Métriques d'entraînement:") for metric, value in st.session_state.model_results[selected_model]['train_metrics'].items(): st.write(f"{metric}: {value:.4f}") with col2: st.write("Métriques de test:") for metric, value in st.session_state.model_results[selected_model]['test_metrics'].items(): st.write(f"{metric}: {value:.4f}") # Interprétation du modèle elif page == "Interprétation du modèle": st.header(f"Interprétation du modèle - {selected_model}") if selected_model in ["Decision Tree", "Random Forest"]: if selected_model == "Decision Tree": st.subheader("Visualisation de l'arbre") max_depth = st.slider("Profondeur maximale à afficher", 1, 5, 3) fig, ax = plt.subplots(figsize=(20, 10)) plot_tree(current_model, feature_names=list(feature_names), max_depth=max_depth, filled=True, rounded=True) st.pyplot(fig) st.subheader("Règles de décision importantes") if selected_model == "Decision Tree": st.text(export_text(current_model, feature_names=list(feature_names))) # SHAP values for all models st.subheader("SHAP Values") with st.spinner("Calcul des valeurs SHAP en cours..."): explainer = shap.TreeExplainer(current_model) if selected_model != "Logistic Regression" \ else shap.LinearExplainer(current_model, X_train) shap_values = explainer.shap_values(X_train[:100]) # Using first 100 samples for speed fig, ax = plt.subplots(figsize=(10, 6)) shap.summary_plot(shap_values, X_train[:100], feature_names=list(feature_names), show=False) st.pyplot(fig) # Analyse des caractéristiques elif page == "Analyse des caractéristiques": st.header("Analyse des caractéristiques") # Feature importance st.subheader("Importance des caractéristiques") importance_fig = plot_feature_importance(current_model, feature_names, selected_model) st.pyplot(importance_fig) # Feature correlation st.subheader("Matrice de corrélation") corr_matrix = X_train.corr() fig, ax = plt.subplots(figsize=(10, 8)) sns.heatmap(corr_matrix, annot=True, cmap='coolwarm', center=0) st.pyplot(fig) # Simulateur de prédictions else: st.header("Simulateur de prédictions") input_values = {} for feature in feature_names: if X_train[feature].dtype == 'object': input_values[feature] = st.selectbox( f"Sélectionnez {feature}", options=X_train[feature].unique() ) else: input_values[feature] = st.slider( f"Valeur pour {feature}", float(X_train[feature].min()), float(X_train[feature].max()), float(X_train[feature].mean()) ) if st.button("Prédire"): input_df = pd.DataFrame([input_values]) prediction = current_model.predict_proba(input_df) st.write("Probabilités prédites:") st.write({f"Classe {i}": f"{prob:.2%}" for i, prob in enumerate(prediction[0])}) if selected_model == "Decision Tree": st.subheader("Chemin de décision") node_indicator = current_model.decision_path(input_df) leaf_id = current_model.apply(input_df) node_index = node_indicator.indices[node_indicator.indptr[0]:node_indicator.indptr[1]] rules = [] for node_id in node_index: if node_id != leaf_id[0]: threshold = current_model.tree_.threshold[node_id] feature = feature_names[current_model.tree_.feature[node_id]] if input_df.iloc[0][feature] <= threshold: rules.append(f"{feature} ≤ {threshold:.2f}") else: rules.append(f"{feature} > {threshold:.2f}") for rule in rules: st.write(rule) if __name__ == "__main__": app()