Update app.py

app.py

@@ -28,6 +28,45 @@ from sklearn.impute import SimpleImputer
 from sklearn.linear_model import LogisticRegression
 from sklearn.model_selection import train_test_split
 
+#Figures setting block
+import io
+
+def make_fig(figsize=(5.5, 3.6), dpi=120):
+    """
+    Small figure for laptop screens.
+    """
+    fig = plt.figure(figsize=figsize, dpi=dpi)
+    return fig
+
+def fig_to_png_bytes(fig, dpi=600):
+    """
+    Export current figure as PNG bytes at high DPI (>=600).
+    """
+    buf = io.BytesIO()
+    fig.savefig(buf, format="png", dpi=int(dpi), bbox_inches="tight")
+    buf.seek(0)
+    return buf.getvalue()
+
+def render_plot_with_download(
+    fig,
+    *,
+    title: str,
+    filename: str,
+    export_dpi: int = 600
+):
+    """
+    Show a compact plot in Streamlit + provide high-DPI PNG download.
+    """
+    st.pyplot(fig, clear_figure=True, use_container_width=True)
+
+    png_bytes = fig_to_png_bytes(fig, dpi=export_dpi)
+    st.download_button(
+        label=f"Download {title} (PNG {export_dpi} dpi)",
+        data=png_bytes,
+        file_name=filename,
+        mime="image/png",
+        key=f"dl_{filename}"
+    )
 
 
 # ============================================================
@@ -717,68 +756,85 @@ with tab_train:
                 )
                 st.markdown("**Confusion Matrix (threshold = 0.5)**")
                 st.dataframe(cm_df)
+
+                st.markdown("### Plot display settings")
+
+                plot_width = st.slider("Plot width (inches)", 4.0, 10.0, 5.5, 0.1)
+                plot_height = st.slider("Plot height (inches)", 2.5, 6.0, 3.6, 0.1)
+                plot_dpi_screen = st.slider("Screen DPI", 80, 200, 120, 10)
+                
+                export_dpi = st.selectbox("Export DPI (PNG)", [300, 600, 900, 1200], index=1)
+                
+                FIGSIZE = (plot_width, plot_height)
+
                 
                 # ROC curve plot (matplotlib)
                 roc = m["roc_curve"]
-                fig = plt.figure()
+                fig = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
                 plt.plot(roc["fpr"], roc["tpr"])
                 plt.plot([0, 1], [0, 1])
                 plt.xlabel("False Positive Rate (1 - Specificity)")
                 plt.ylabel("True Positive Rate (Sensitivity)")
                 plt.title(f"ROC Curve (AUC = {m['roc_auc']:.3f})")
-                st.pyplot(fig, clear_figure=True)
+                
+                render_plot_with_download(
+                    fig,
+                    title="ROC curve",
+                    filename="roc_curve.png",
+                    export_dpi=export_dpi
+                )
 
-                st.divider()
-                st.subheader("Precision–Recall (PR) Curve")
-                if "pr_curve" not in m:
-                    st.warning("PR curve not available in this model metadata. Retrain the model to generate it.")
-                else:
-                    pr = m["pr_curve"]
-                    c1, c2 = st.columns(2)
-                    c1.metric("Average Precision (AP)", f"{pr['average_precision']:.3f}")
-                    
-                    fig_pr = plt.figure()
-                    plt.plot(pr["recall"], pr["precision"])
-                    plt.xlabel("Recall")
-                    plt.ylabel("Precision")
-                    plt.title(f"PR Curve (AP = {pr['average_precision']:.3f})")
-                    st.pyplot(fig_pr, clear_figure=True)
+                #Precision recall curve
+                pr = m["pr_curve"]
+                fig_pr = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(pr["recall"], pr["precision"])
+                plt.xlabel("Recall")
+                plt.ylabel("Precision")
+                plt.title(f"PR Curve (AP = {pr['average_precision']:.3f})")
+                
+                render_plot_with_download(
+                    fig_pr,
+                    title="PR curve",
+                    filename="pr_curve.png",
+                    export_dpi=export_dpi
+                )
 
-                st.divider()
-                st.subheader("Calibration (Reliability Plot)")
                 
-                if "calibration" not in m:
-                    st.warning("calibration curve not available in this model metadata. Retrain the model to generate it.")
-                else:
-                    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']}")
-                    
-                    fig_cal = plt.figure()
-                    plt.plot(cal["prob_pred"], cal["prob_true"])
-                    plt.plot([0, 1], [0, 1])
-                    plt.xlabel("Mean predicted probability")
-                    plt.ylabel("Observed event rate")
-                    plt.title("Calibration curve")
-                    st.pyplot(fig_cal, clear_figure=True)
+                #Calibration plot
+                cal = m["calibration"]
+                fig_cal = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(cal["prob_pred"], cal["prob_true"])
+                plt.plot([0, 1], [0, 1])
+                plt.xlabel("Mean predicted probability")
+                plt.ylabel("Observed event rate")
+                plt.title("Calibration curve")
+                
+                render_plot_with_download(
+                    fig_cal,
+                    title="Calibration curve",
+                    filename="calibration_curve.png",
+                    export_dpi=export_dpi
+                )
 
 
-                st.divider()
-                st.subheader("Decision Curve Analysis (Clinical Usefulness)")
+                #Decision curve
+                dca = m["decision_curve"]
+                fig_dca = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(dca["thresholds"], dca["net_benefit_model"], label="Model")
+                plt.plot(dca["thresholds"], dca["net_benefit_all"], label="Treat all")
+                plt.plot(dca["thresholds"], dca["net_benefit_none"], label="Treat none")
+                plt.xlabel("Threshold probability")
+                plt.ylabel("Net benefit")
+                plt.title("Decision curve analysis")
+                plt.legend()
+                
+                render_plot_with_download(
+                    fig_dca,
+                    title="Decision curve",
+                    filename="decision_curve.png",
+                    export_dpi=export_dpi
+                )
 
-                if "decision_curve" not in m:
-                    st.warning("decision curve not available in this model metadata. Retrain the model to generate it.")
-                else:
-                    dca = m["decision_curve"]
-                    fig_dca = plt.figure()
-                    plt.plot(dca["thresholds"], dca["net_benefit_model"])
-                    plt.plot(dca["thresholds"], dca["net_benefit_all"])
-                    plt.plot(dca["thresholds"], dca["net_benefit_none"])
-                    plt.xlabel("Threshold probability")
-                    plt.ylabel("Net benefit")
-                    plt.title("Decision curve analysis")
-                    st.pyplot(fig_dca, clear_figure=True)
                 
                 st.caption(
                     "If the model curve is above Treat-all and Treat-none across a threshold range, "
@@ -967,48 +1023,72 @@ with tab_predict:
                 st.dataframe(cm_df)
         
                 # ROC plot
-                fig = plt.figure()
+                fig = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
                 plt.plot(fpr, tpr)
                 plt.plot([0, 1], [0, 1])
                 plt.xlabel("False Positive Rate (1 - Specificity)")
                 plt.ylabel("True Positive Rate (Sensitivity)")
                 plt.title(f"External ROC Curve (AUC = {roc_auc_ext:.3f})")
-                st.pyplot(fig, clear_figure=True)
+                
+                render_plot_with_download(
+                    fig,
+                    title="External ROC curve",
+                    filename="external_roc_curve.png",
+                    export_dpi=export_dpi
+                )
+
         
                 # PR plot
-                st.subheader("Precision–Recall (external)")
-                c1, c2 = st.columns(2)
-                c1.metric("Average Precision (AP)", f"{pr_ext['average_precision']:.3f}")
-                fig_pr = plt.figure()
-                plt.plot(pr_ext["recall"], pr_ext["precision"])
+                pr = m["pr_curve"]
+                fig_pr = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(pr["recall"], pr["precision"])
                 plt.xlabel("Recall")
                 plt.ylabel("Precision")
-                plt.title(f"External PR Curve (AP = {pr_ext['average_precision']:.3f})")
-                st.pyplot(fig_pr, clear_figure=True)
+                plt.title(f"PR Curve (AP = {pr['average_precision']:.3f})")
+                
+                render_plot_with_download(
+                    fig_pr,
+                    title="PR curve",
+                    filename="pr_curve.png",
+                    export_dpi=export_dpi
+                )
+
         
                 # Calibration plot
-                st.subheader("Calibration (external)")
-                c1, c2 = st.columns(2)
-                c1.metric("Brier score", f"{cal_ext['brier']:.4f}")
-                c2.write(f"Bins: {cal_ext['n_bins']} | Strategy: {cal_ext['strategy']}")
-                fig_cal = plt.figure()
-                plt.plot(cal_ext["prob_pred"], cal_ext["prob_true"])
+                cal = m["calibration"]
+                fig_cal = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(cal["prob_pred"], cal["prob_true"])
                 plt.plot([0, 1], [0, 1])
                 plt.xlabel("Mean predicted probability")
                 plt.ylabel("Observed event rate")
-                plt.title("External Calibration curve")
-                st.pyplot(fig_cal, clear_figure=True)
+                plt.title("Calibration curve")
+                
+                render_plot_with_download(
+                    fig_cal,
+                    title="Calibration curve",
+                    filename="calibration_curve.png",
+                    export_dpi=export_dpi
+                )
+
         
                 # DCA plot
-                st.subheader("Decision Curve Analysis (external)")
-                fig_dca = plt.figure()
-                plt.plot(dca_ext["thresholds"], dca_ext["net_benefit_model"])
-                plt.plot(dca_ext["thresholds"], dca_ext["net_benefit_all"])
-                plt.plot(dca_ext["thresholds"], dca_ext["net_benefit_none"])
+                dca = m["decision_curve"]
+                fig_dca = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+                plt.plot(dca["thresholds"], dca["net_benefit_model"], label="Model")
+                plt.plot(dca["thresholds"], dca["net_benefit_all"], label="Treat all")
+                plt.plot(dca["thresholds"], dca["net_benefit_none"], label="Treat none")
                 plt.xlabel("Threshold probability")
                 plt.ylabel("Net benefit")
-                plt.title("External Decision curve analysis")
-                st.pyplot(fig_dca, clear_figure=True)
+                plt.title("Decision curve analysis")
+                plt.legend()
+                
+                render_plot_with_download(
+                    fig_dca,
+                    title="Decision curve",
+                    filename="decision_curve.png",
+                    export_dpi=export_dpi
+                )
+
         
             except Exception as e:
                 st.error(f"Could not compute external validation metrics: {e}")
@@ -1190,7 +1270,7 @@ with tab_predict:
             st.markdown(f"### Global SHAP summary (first {batch_n} rows)")
         
             # BAR SUMMARY
-            fig_bar = plt.figure()
+            fig_bar = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
             shap.summary_plot(
                 shap_vals_batch,
                 features=X_dense,
@@ -1198,20 +1278,33 @@ with tab_predict:
                 plot_type="bar",
                 max_display=max_display,
                 show=False,
+                plot_size=FIGSIZE
+            )
+            render_plot_with_download(
+                fig_bar,
+                title="SHAP bar summary",
+                filename="shap_summary_bar.png",
+                export_dpi=export_dpi
             )
-            st.pyplot(fig_bar, clear_figure=True)
+
         
             # BEESWARM SUMMARY (optional)
-            if show_beeswarm:
-                fig_swarm = plt.figure()
-                shap.summary_plot(
-                    shap_vals_batch,
-                    features=X_dense,
-                    feature_names=names,
-                    max_display=max_display,
-                    show=False,
-                )
-                st.pyplot(fig_swarm, clear_figure=True)
+            fig_swarm = make_fig(figsize=FIGSIZE, dpi=plot_dpi_screen)
+            shap.summary_plot(
+                shap_vals_batch,
+                features=X_dense,
+                feature_names=names,
+                max_display=max_display,
+                show=False,
+                plot_size=FIGSIZE
+            )
+            render_plot_with_download(
+                fig_swarm,
+                title="SHAP beeswarm",
+                filename="shap_beeswarm.png",
+                export_dpi=export_dpi
+            )
+
         
             st.markdown("### Waterfall plots (batch)")