Which digits get confused?

Graded on the total number of mistakes and the worst pair.

The script trains the closest-example model, predicts the test images, and prints the confusion matrix. The two loops then visit every cell of the matrix: true_digit picks the row and pred_digit picks the column. Two things in the starter may be new:

• continue means "skip the rest of this turn and go straight to the next one". Here it skips the diagonal cells, which count correct answers, so every cell that reaches the lines below it is a mistake.
• None means "nothing yet". worst_true and worst_pred hold None until the loop finds a mistake.

Finish the loop body: add each cell's count to mistakes, and whenever a cell's count is bigger than worst_count, remember that cell, as the comment in the starter describes. The last two lines then print the total and the worst pair.