search

ukuleleplayer / pureples

Pure Python Library for ES-HyperNEAT. Contains implementations of HyperNEAT and ES-HyperNEAT.
MIT License
113 stars 36 forks source link

Question about hyperparameter tuning / memory consumption #24

Closed pablogranolabar closed 2 years ago

pablogranolabar commented 2 years ago

Hi again @ukuleleplayer

I've got a collection of flat binary inputs of self.observation_space = spaces.Box(low=0., high=1., shape=(163840,)) which by usual standards in Gym is not that large I don't think; however when evolving even the smallest parameter ANN, I'm getting OOM within seconds.

Do you have any thoughts on hyperparameter tuning with PUREPLES as relates to memory consumption? I get OOM even with a 64GB instance using this observation space.

TIA

ukuleleplayer commented 2 years ago

Hi,

That sounds like something is off in the code somewhere - if you configure the smallest possible ES network, well, then your bottleneck in memory consumption should definitely not be there.. I can't say be sure what's happening, but maybe look elsewhere than your parameter tuning. Check first whether it is even possible to run in a normal NEAT setup.

Cheers.