search

seallard / walker

NEAT
MIT License
0 stars 0 forks source link

Species converge on the same non-optimal fitness #63

Closed seallard closed 3 years ago

seallard commented 3 years ago

It seems like the different species reach a maximum fitness of 8.9999 after a large number of generations and do not improve beyond that. The population consists of 5 different species containing 30 individuals. Each of them are structurally different, as expected. This probably reflects some fundamental underlying problem... But what? And why?

All of the networks can solve all of the XOR-inputs except for (1,1,0).

seallard commented 3 years ago

The zeroth node is a bias node with value of 1. Species 1: image

Species 2: image

Species 3: image

Species 4: image

Species 5: image

seallard commented 3 years ago

There were several bugs in the mutation methods. Also, the sigmoid had to be steepened for the networks to be able to solve XOR.