Right now Anneal Evolution supports Monte Carlo trials and can tune the deviation for gaussian sampling. Neuro Evolution includes features for genetic algorithms and neural network input. It may be more convenient to place all of these features in one evolution class, so all controllers can use all learning methods.
Thoughts on whether it is worth the work (1 - 2 hours) to combine them? Other thoughts on the architecture of the learning library? I could also see arguments for simplifying the classes (one method per class) to clarify implementation.
In addition to the learning algorithms, NeuroEvolution does not print out its parameters like AnnealEvolution, making it more difficult to restart a run. This feature should be added in either case.
Right now Anneal Evolution supports Monte Carlo trials and can tune the deviation for gaussian sampling. Neuro Evolution includes features for genetic algorithms and neural network input. It may be more convenient to place all of these features in one evolution class, so all controllers can use all learning methods.
Thoughts on whether it is worth the work (1 - 2 hours) to combine them? Other thoughts on the architecture of the learning library? I could also see arguments for simplifying the classes (one method per class) to clarify implementation.