Jan 9, 2023
Very cool! I understand that the GA parameters are not optimized, but I think the mutation ratio is a little too high. In general, we think about mutation at rates like 0.05 or less. The population is a little too small too, you could try 100 individuals, although your example is a very small ANN.
We also have some other advanced GAs that may work better, like this one (full disclosure, I'm one of the authors):
