search

raymoo / NEET

Neuroevolution of Augmented Topologies (NEAT) -- in Haskell
GNU General Public License v3.0
12 stars 4 forks source link

Species turnover too high #2

Closed raymoo closed 9 years ago

raymoo commented 9 years ago

Too many species die and generate - over a hundred in just 20 generations.

raymoo commented 9 years ago

Found a few bugs, which helped a bit, the number of generated species is now below 100 for around 20 generations. But that's still way too high.

raymoo commented 9 years ago

Some clues: distance was returning NaN, because sometimes the shared connections were none. This was happening even in xorExperiment, where this should never happen since the initial population is fully connected.

raymoo commented 9 years ago

Actually no, I did switch it to sparse at some point. But the NaN thing was still a problem.

raymoo commented 9 years ago

The changes I've made recently seem to have made this less of a problem, though I'm not really sure if it's gone.

raymoo commented 9 years ago

The problem seems to be gone now, so I guess I will close this for now.