Open havietisov opened 1 year ago
I have been having the exact same issue. It happens usually at generation two for me if i start out with two hidden nodes. If i start out with one hidden nodes, the population size diverges from the configured number after 10-100 generations.
I also have had this happen when loading a checkpoint, NEAT overwriting it initially, stopping training and then re-load from that overwritten checkpoint.
I have hidden nodes set to 20, and it is immediately the wrong size (I want 250 per generation and got 546 in the first generation)
This software is completely unusable for me if this bug is not fixed.
when
initial_connection
value in the config set to fs_neat_hidden, partial_direct, full_direct, and andnum_hidden
is other than zero, creates population of size, different from what is set in config, on windows version specificallyTo Reproduce Steps to reproduce the behavior:
Expected behavior Population to be exactly of a size set in config
Screenshots I hope log would be enough
Desktop:
OS: windows 10
Config is in the attachment: config-feedforward.txt