Hi Bill and Berns,
The idea behind this is that you should run evaluation metrics on data that wasn't used in the training process. This is why there is the check box.
but the rationale is:
when you train a model, it is learning from your data. To know if it really learned properly, you should test it on data that it has never seen before. Otherwise it's like a student that studied with the test questions. They will do very well on the test but that's because they "cheated".
To do that in real life, we usually take some data from our training dataset and put it on a different sheet and don't use it on training, only for testing purposes.
Does it make sense?
Gus