Sure of Itself, and Wrong

A photo classifier reports how confident it is in each answer. A healthy model is very sure when it is right and less sure when it is wrong. The dangerous mistakes are the ones it made while still very sure.

Task: write confidence_check(y_true, y_pred, confidence, cutoff).

• y_true and y_pred are the real and predicted labels. confidence is the model's confidence in each prediction, between 0 and 1.
• Return a list of three values:
  1. the mean confidence on rows it got right, rounded to 3 decimal places,
  2. the mean confidence on rows it got wrong, rounded to 3 decimal places,
  3. how many wrong rows had a confidence strictly above cutoff, as a whole number.

Every test has at least one right row and one wrong row.