Evidence for the Wine Model

A reviewer wants three things before trusting a wine classifier: how often it is right, how that compares to guessing, and what it leans on.

Task: write wine_evidence(n).

• Load load_wine(return_X_y=True, as_frame=True) and split with train_test_split(X, y, test_size=0.25, random_state=0, stratify=y).
• Fit make_pipeline(StandardScaler(), LogisticRegression(max_iter=1000)) on the training rows.
• Fit DummyClassifier(strategy="most_frequent") on the training rows as the baseline.
• Run permutation_importance(model, X_test, y_test, n_repeats=10, random_state=0) on the pipeline.
• Return a dict:
  • "model": the pipeline's test accuracy, rounded to 3 decimal places
  • "baseline": the dummy's test accuracy, rounded to 3 decimal places
  • "leans_on": the names of the n most important columns, highest first