Update app.py

app.py

@@ -264,6 +264,11 @@ def train_and_save(df: pd.DataFrame, feature_cols, num_cols, cat_cols, n_bins: i
             "thresholds": [float(x) for x in roc_thresholds],
         },
     }
+    metrics.update({
+        "pr_curve": pr,
+        "calibration": cal,
+        "decision_curve": dca,
+    })
 
 
     joblib.dump(pipe, "model.joblib")
@@ -535,11 +540,7 @@ with tab_train:
                 c8.metric("Test N", m["n_test"])
 
 
-                metrics.update({
-                    "pr_curve": pr,
-                    "calibration": cal,
-                    "decision_curve": dca,
-                })
+         
 
                 
                 # Confusion matrix display
@@ -564,8 +565,11 @@ with tab_train:
 
                 st.divider()
                 st.subheader("Precision–Recall (PR) Curve")
-                
-                pr = meta["metrics"]["pr_curve"]
+                if "pr_curve" not in m:
+                    st.warning("PR curve not available in this model metadata. Retrain the model to generate it.")
+                    st.stop()
+                                
+                pr = m["pr_curve"]
                 c1, c2 = st.columns(2)
                 c1.metric("Average Precision (AP)", f"{pr['average_precision']:.3f}")
                 
@@ -579,7 +583,10 @@ with tab_train:
                 st.divider()
                 st.subheader("Calibration (Reliability Plot)")
                 
-                cal = meta["metrics"]["calibration"]
+                if "calibration" not in m:
+                    st.warning("calibration not available in this model metadata. Retrain the model to generate it.")
+                    st.stop()
+                cal = m["calibration"]
                 c1, c2 = st.columns(2)
                 c1.metric("Brier score", f"{cal['brier']:.4f}")
                 c2.write(f"Bins: {cal['n_bins']} | Strategy: {cal['strategy']}")
@@ -595,8 +602,12 @@ with tab_train:
 
                 st.divider()
                 st.subheader("Decision Curve Analysis (Clinical Usefulness)")
+
+                if "decision_curve" not in m:
+                    st.warning("decision_curve not available in this model metadata. Retrain the model to generate it.")
+                    st.stop()
                 
-                dca = meta["metrics"]["decision_curve"]
+                dca = m["decision_curve"]
                 
                 fig_dca = plt.figure()
                 plt.plot(dca["thresholds"], dca["net_benefit_model"])