Do it yourself

Graded on four numbers the script prints.

This is your first complete classifier with two parts missing. It is a little shorter than the lesson: there is one split, into a training part and a test part, instead of three.

1. The split. Write one line that calls train_test_split on X and y with test_size=0.25 and random_state=0, and unpacks the four results into X_train, X_test, y_train, y_test, in that order. Step 2 of "Your first complete digit classifier" does the same thing with different names and settings.
2. The model. Create a KNeighborsClassifier with n_neighbors=3, call it model, and fit it on X_train and y_train.

Then run. The script prints the accuracy on the training images, the accuracy on the test images, the gap between them (training minus test: a large gap would mean the model does much better on images it stored than on new ones), and the always-one-digit baseline. Your model should score far above the baseline.