Introducing "activation mutation" and "aggregation mutation"

[mdalvi]

Hello Colin,

Since you have started a major refactor of SharpNEAT here is something you might like.

The documentation belongs to a NEAT's implementation in python written by CodeReclaimers and what I want you to take a look at is there is a configurable option under [DefaultGenome] section which says activation_mutate_rate and aggregation_mutate_rate.

These options actually allow algorithm to pick different activations and aggregations from the specified activation_options and aggregation_options respectively at random during mutation.

While this sounds duuhhhh! but when I used them in some of my experiments I observed better convergence and faster convergence.

So instead of specifying same activation and aggregation for all nodes, this should be an option worth configurable in this refactor.

[colgreen]

Notes.

'Aggregation function' here refers to how the incoming signals to a node are combined. I.e. the standard approach is to sum the inputs before passing the result through a node activation function, however NEAT-python allows for other aggregations, see module-aggregations.