search

seallard / walker

NEAT
MIT License
0 stars 0 forks source link

Stuck in update #48

Closed seallard closed 3 years ago

seallard commented 3 years ago

After ca 20 000 evaluations (in total over multiple genomes and generations), the program is stuck in the while-loop of network.activation.

seallard commented 3 years ago

Both of the input links have been disconnected somehow. image

seallard commented 3 years ago

The bug is not present if crossover is disabled, so it probably has to do with the crossover code.

seallard commented 3 years ago

The network does not have any links! Does the genome which the network was created from contain any links? image

seallard commented 3 years ago

Yes, the genome does have two links but none of them are enabled. How does this happen?

seallard commented 3 years ago

If the links have matching id:s they are randomly chosen from either parent. If one disabled version of each original link is present in the parents and the more fit parent does not have any other genes, the offspring might end up without active links.