baum_tiefe.png

def fig_baum_tiefe(X_train, X_val, y_train, y_val) -> None:
    """Train vs validation AUC over decision tree depth (overfitting curve)."""
    depths = [1, 2, 3, 4, 5, 6, 7, 8, 10, 12]
    train_aucs, val_aucs = [], []
    for d in depths:
        tree = DecisionTreeClassifier(max_depth=d, random_state=SEED)
        tree.fit(X_train, y_train)
        train_aucs.append(roc_auc_score(y_train, tree.predict_proba(X_train)[:, 1]))
        val_aucs.append(roc_auc_score(y_val,   tree.predict_proba(X_val)[:, 1]))

    fig, ax = plt.subplots(figsize=(7, 4.2))
    ax.plot(depths, train_aucs, "o-", color=PRIMARY, label="Trainings-AUC")
    ax.plot(depths, val_aucs,   "s-", color=EVENT,   label="Validierungs-AUC")
    ax.axvline(depths[int(np.argmax(val_aucs))], color=SECONDARY, lw=0.9,
               ls="--", label=f"Optimale Tiefe = {depths[int(np.argmax(val_aucs))]}")
    ax.set_xlabel("Baumtiefe (max_depth)")
    ax.set_ylabel("ROC-AUC")
    ax.set_title("Überanpassung: AUC über Baumtiefe")
    ax.legend(loc="lower right")
    save(fig, ASSETS / "baum_tiefe.png")