mrkwjc
commented
5 years ago Fot GA training in ffnet you do something like that (see examples):
net.train_genetic(input, target, individuals=20, generations=500)
The 'individual' here is a vector of network weights. This method is higly inefficient for larger networks. It can be useful however for searching a good starting point for other algorithms.
BTW: you cannot use custom fitness functions in ffnet (in an easy way at least), ffnet always minimizes sum of squared errors at network output.
Can anybody provide an example of how to train a network with a genetic algorithm using ffnet? Pseudocode for the GA part would suffice. I'd like to use DEAP (https://deap.readthedocs.io/en/master/) to train a NN using a simulation as fitness function. I just want to make sure I'm using ffnet the right (and best) way. Thanks.