Enable multiclass support for TPOT-NN

[rachitk]

What does this PR do?

Disables the requirement for binary output classes in TPOT-NN. Resolves #1149.

Where should the reviewer start?

Check tpot/builtins/nn.py and make sure the standard TPOT tests still pass.

How should this PR be tested?

Test against the standard nosetests suite and see if anything fails. There is currently a test for TPOT-NN specifically to ensure that multiclass problems are NOT supported that will fail as this PR disables that check; this can likely be modified to test that all estimators support multiclass problems.

Multiclass problems where the set of output classes does not exist in a sequence beginning from 0 will still fail - for example, a problem with classes [0, 2, 3, 5, 6] will fail due to not having class 1 and 4, and a problem with classes [1, 2, 3] will fail due to not having class 0. This will likely need to be addressed in a future PR.

The following code should test the multiclass functionality of the NN MLP classifier using TPOT's dependencies:

from tpot import TPOTClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split


if(__name__ == "__main__"):

	iris = load_iris()
	X_train, X_test, y_train, y_test = train_test_split(iris.data, iris.target, train_size=0.75, test_size=0.25)

	nn_classifier = TPOTClassifier(config_dict='TPOT NN', template='Selector-Transformer-PytorchMLPClassifier', 
		generations=3, population_size=3)

	nn_classifier.fit(X_train, y_train)
	print(nn_classifier.score(X_test, y_test))

	nn_classifier.export('pipeline_nn_iris_norestriction.py')

Any background context you want to provide?

This addresses #1149. Multiclass support was tested and validated using a variety of standard multiclass datasets, with no notable score degradation from base TPOT on multiclass problems.

What are the relevant issues?

#1149

Screenshots (if appropriate)

Questions:

• Do the docs need to be updated? No, but adding an example of using TPOT-NN for multiclass classification may be helpful.
• Does this PR add new (Python) dependencies? No.

[JDRomano2]

Please note that failed tests are not due to changes made in this PR - they fail elsewhere in the codebase. This work will be merged into master when the failed tests are resolved.