Store the training set internally so the user doesn't have to repeatedly pass it

[rhiever]

Since the training set is required to properly run the pipeline each time, this information should be stored internally so the user doesn't have to repeatedly pass it to TPOT (e.g., with the score() function).

[rhiever]

Actually, this is a bad idea. It can be disastrous to have multiple copies of a very large data set.

[rasbt]

Agreed! Plus, the training set may already be copied a bunch of times already (unless you disable the mutliprocessing in certain pipeline objects)

[rhiever]

I actually need to work through the pipeline operator code and probably remove all the copy() operations. I think they're unnecessary, but I'm actually not sure if the DataFrames are being passed by copy or reference.