https://github.com/jliphard/DeepEvolve

Found a genetic algorithm that finds out the best parameters for a
neural network at github:

	https://github.com/harvitronix/neural-network-genetic-algorithm

Modified it in order to get it compatible with the KerasROOTClassification class.

Stop training when a monitored quantity has stopped improving.

* store history information to json
* shuffle training data before fit
* include sample_weight
* separate load function to plot from previously trained setup
* weights and history are still stored after keyboard interrupt