Update src/streamlit_app.py

src/streamlit_app.py

@@ -463,8 +463,45 @@ if submitted:
         # If SELF donor case, prediction is forced to 0.0 and there is no meaningful SHAP
         if st.session_state.SELF:
             st.info("SHAP is not shown for SELF donor (prediction forced to 0).")
-        elif model_to_use.endswith("ensemble"):
-            st.info("SHAP for ensemble is available if your ensemble_shap() is configured. Skipping here for now.")
+        elif model_to_use.endswith("ensemble"):    
+            ensemble_data = load_model_ensemble(model_to_use)
+            models = ensemble_data["model"]
+        
+            # Call your ensemble SHAP helper
+            shap_values, shap_base, feature_names = ensemble_shap(
+                models=models,
+                X=X_model,
+                cat_features=cat_features,
+            )
+        
+            # Build SHAP explanation object (single patient)
+            one = shap.Explanation(
+                values=shap_values[0],
+                base_values=shap_base[0] if np.ndim(shap_base) > 0 else shap_base,
+                data=X_model.iloc[0].values,
+                feature_names=feature_names,
+            )
+        
+            vals = one.values
+            feats = np.array(one.feature_names)
+        
+            top_idx = np.argsort(np.abs(vals))[::-1][:20]
+        
+            shap_table = pd.DataFrame({
+                "Feature": feats[top_idx],
+                "Feature value": X_model.iloc[0][feats[top_idx]].values,
+                "SHAP value (pushes risk ↑ / ↓)": vals[top_idx],
+            })
+        
+            st.subheader("Top features driving this patient’s prediction")
+            st.dataframe(shap_table, use_container_width=True)
+        
+            st.subheader("Waterfall plot (ensemble)")
+            plt.figure(figsize=(10, 6))
+            shap.plots.waterfall(one, max_display=20, show=False)
+            st.pyplot(plt.gcf(), bbox_inches="tight")
+            plt.clf()
+            st.caption("SHAP values shown are averaged across ensemble models.")
         else:
             model_dict = load_model(model_to_use)
             model = model_dict["model"]  # XGBoost model