search

CodeReclaimers / neat-python

Python implementation of the NEAT neuroevolution algorithm
BSD 3-Clause "New" or "Revised" License
1.4k stars 488 forks source link

Support Batching when evaluating genome fitness #245

Closed ephi closed 2 years ago

ephi commented 2 years ago

python neat supports fitness evolution by sending one example at a time. This is not very practical when one want to evaluate fitness over a big dataset.

So instead of just sending one example at time to the evolved network via activate(..) method. it'd be better if activate could accept a matrix N X D and output matrix N X D2 where D = the input dim (number of input neurons), D2 = the output dim (number of output neurons).

TheMightiestCarrot commented 2 years ago

How did you resolve this problem?

ephi commented 2 years ago

Not directly resolved this, I batched over HyperNEAT substrate instead. Though I'd be happy if batching were to be supported natively.