Confusion matrix

A table of predicted class against actual class, so every kind of mistake is visible separately rather than blended into one score. The first thing to print when a classifier disappoints.

A definition is not an understanding. This is taught properly in Metrics in code, with the traps, part of Model evaluation & metrics.