Keeping data-heavy objects after fit

[Jonathan Shor]

Would it be useful in some situations to keep some of the heavy internal data objects attached to the classifier, even after fit is done with them?

For example, for some visualization methods, keeping the synthetics created to include them might be interesting. Tho, there is a question of how you visualize 25 runs of synthetics, as they change on each run.

For ease of using included visualization methods, it might be useful to keep both the created synthetics and the original count matrix so all the data to plot is in the same place.

Perhaps a boolean parameter keep, defaulting to False, could control this.