Arun got it spot on. I've mentioned expected and predicted labels next to each other so at a quick glance the user can see how these differ. Never fit these transformations on test data.

train -> fit + transform

test -> transform