jonathan-laurent / AlphaZero.jl

A generic, simple and fast implementation of Deepmind's AlphaZero algorithm.

Home Page:https://jonathan-laurent.github.io/AlphaZero.jl/stable/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Maximum number of iterations

StepHaze opened this issue · comments

What maximum number of iterations have you done with AlphaZero.jl ?
Please share your experience

I personally have trained small games with up to 40 iterations but number of iterations is not a very interesting metric on its own. There is a tradeoff between doing a small number of iterations with a large number of simulated games each time and a large number of shorter iterations. As a rough approximation, I would say that the former is potentially more sample-efficient while the latter is more stable / easier to tune.