search

schrum2 / MM-NEAT

Modular Multiobjective (Hyper) Neuro-Evolution of Augmenting Topologies + MAP-Elites: Java code for evolving intelligent agents in Ms. Pac-Man, Tetris, and more, as well as code for Procedural Content Generation in Mario, Zelda, Minecraft, and more!
http://people.southwestern.edu/~schrum2/re/mm-neat.php
Other
50 stars 20 forks source link

See if old experiment batch files are broken for not specifying crossover correctly or other issues #952

Open schrum2 opened 1 year ago

schrum2 commented 1 year ago

MMNEAT assumes default crossover is TWEANNCrossover, which should be fine for older NEAT tasks, but will not work for evolution of real vectors. For a while, these just defaulted to using ArrayCrossover, but that needs to be explicit in the batch file now.

schrum2 commented 1 year ago

I added a separate arrayCrossover command line parameter that is used by the array-based genotypes now. All other genotypes seem to be networks, but they still just use the parameter called "crossover"

This was the solution to TWEANNPlusParameter genotypes that needed to perform both network crossover and array crossover on two separate components of the genotype.

This seems to make everything work, though I am still double-checking

schrum2 commented 1 year ago

otherStats have proven to be a big issue for some of the MAP Elites runs, as well as the requirement to only track one fitness function now that MOME is around.

I've checked everything at this point except for UT2004 and I also need to re-check Minecraft, but that should be fine