Every Combination, Five Folds Each

A clinic wants the best neighbour settings for a tumour classifier, chosen honestly: by cross-validation on the training rows, with the test rows opened only once at the end.

Task: write search_neighbours(neighbour_counts, weightings).

1. Load the data with load_breast_cancer(return_X_y=True, as_frame=True).
2. Split it with train_test_split(X, y, test_size=0.25, random_state=0, stratify=y).
3. Run a GridSearchCV over make_pipeline(StandardScaler(), KNeighborsClassifier()) with cv=5. Search every combination of neighbour_counts for the neighbour count and weightings for the weights setting.
4. Fit the search on the training rows.

Return a dict:

• "best": the search's best_params_, exactly as scikit-learn gives it,
• "cv": the best cross-validated score, rounded to 4 decimal places,
• "test": the refitted winner's accuracy on the test rows, rounded to 4 decimal places,
• "fits": how many models the search trained during cross-validation (combinations × folds), as a whole number.