search

jonathan-laurent / AlphaZero.jl

A generic, simple and fast implementation of Deepmind's AlphaZero algorithm.
https://jonathan-laurent.github.io/AlphaZero.jl/stable/
MIT License
1.24k stars 140 forks source link

When should I stop learning? #194

Open StepHaze opened 1 year ago

StepHaze commented 1 year ago

If (after self-play and learning) the network has stopped replacing itself (<55% points ?) during 5-10 iterations, should I stop learning?

jonathan-laurent commented 1 year ago

There is still a small possibility the network is going to improve later in this case but I would probably stop the experiment and investigate better hyperparameters in such a case.

StepHaze commented 1 year ago

Thanks Jonathan. What hyperparameters should I pay attention first of all?