cestpasphoto / alpha-zero-general

A very fast implementation of AlphaZero, applied to games like Splendor, Santorini, The Little Prince, … Browser version available
MIT License
37 stars 11 forks source link

Amazing Repo, Thank You #7

Open PJF98 opened 4 days ago

PJF98 commented 4 days ago

Hi, just wanted to say thanks for all the effort you put into this repo.

I forked the repo and added 2 player Azul and have managed to train reasonably strong AIs (but not superhuman yet I don't think). I have learnt a lot from doing so. If you ever want a new challenge I would love to see the Azul AI get a lot stronger, I don't think I did a good job optimising the parameters.

cestpasphoto commented 3 days ago

Unfortunately I won't be able to contribute in the near future: I'm a new dad ... with twins. Wish me luck!

Tip: you may need to update the NN architecture as well, like increasing the width or depth.

PJF98 commented 3 days ago

Thanks, best of luck with your new twins!

I tried a couple of different NN architectures, a wider one seemed to work a bit better but I didn't experiment thoroughly.