feat: implement cross-validated predictions caching and update confusion matrix function

app/training/generate_figures.py

@@ -94,14 +94,24 @@ def _load_csv_data(feature_cols):
     return X, y
 
 
-def fig_confusion_matrix(results: dict) -> None:
+def _get_cv_predictions(model, X_scaled, y, cache: dict) -> tuple:
+    """Cross-validated predictions with caching across figures."""
+    key = id(model)
+    if key in cache:
+        return cache[key]
+    cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
+    y_prob = cross_val_predict(
+        clone(model), X_scaled, y, cv=cv, method="predict_proba", n_jobs=-1,
+    )[:, 1]
+    y_pred = (y_prob > 0.5).astype(int)
+    cache[key] = (y, y_pred, y_prob)
+    return y, y_pred, y_prob
+
+
+def fig_confusion_matrix(results: dict, y_true: np.ndarray, y_pred: np.ndarray) -> None:
     """Confusion matrix for the best model (CV predictions)."""
     best = results.get("_best_model", "XGBoost")
-    data = results.get(best)
-    if not data:
-        return
-    y_true = np.array(data["y_true"])
-    y_pred = np.array(data["y_pred"])
+    data = results.get(best, {})
     cm = confusion_matrix(y_true, y_pred)
 
     fig, ax = plt.subplots(figsize=(6.5, 5.5))
@@ -109,9 +119,12 @@ def fig_confusion_matrix(results: dict) -> None:
         "auris", [PALETTE["bg"], PALETTE["primary"]],
     )
     im = ax.imshow(cm, cmap=cmap, aspect="auto")
+    acc = data.get("accuracy", (y_true == y_pred).mean())
+    f1v = data.get("f1", 0.0)
+    aucv = data.get("roc_auc", 0.0)
     ax.set_title(
         f"Karışıklık Matrisi — {best}\n"
-        f"Accuracy: {data['accuracy']:.1%}  F1: {data['f1']:.3f}  AUC: {data['roc_auc']:.3f}",
+        f"Accuracy: {acc:.1%}  F1: {f1v:.3f}  AUC: {aucv:.3f}",
         fontsize=13, fontweight="bold",
     )
     classes = ["İnsan / Human", "AI"]