Parameterize TPOT's GP operations: initialization, selection, mutation, and crossover

I would like to have more control wrt the DEAP settings.
For example, the standard is now to use the NSGA2 algorithm for selection, but I would like to experiment with other selection criteria also. Also the initialization method,crossover and mutation methods can be opened up.

Another issue is the score that comes out of the algorithm. When I asked about the final score not being what I see during the runs I got the following answer: . The score achieved during training is an internal cross-validation score, whereas the score at the end is based on the holdout testing data. That's why you see a difference in the score. In the next version of TPOT, we're removing that information about score throughout the run because it seems to be confusing people for no good reason.

I would suggest not to take the information away but to mention that the final score is on the holdout test data.