s-casci

5 ms·

1. 1.▲
   A simple game to have fun while learning more about soft-body physics(github.com)

   1 points by s-casci 3y ago|0 comments
2. 2.▲
   Show HN: Train a GPT to write like Shakespeare-from scratch, in one Python file(gist.github.com)

   2 points by s-casci 3y ago|0 comments
3. 3.▲
   by s-casci 3y ago

   The policy function outputs the probability of taking every possible (legal or illegal) action. Once you have a way of indexing those actions, both the policy and the game need to refer to the same thing when indexing the same number
4. 4.▲
   by s-casci 3y ago

   Good catches, I've added the missing information. Thanks
5. 5.▲
   by s-casci 3y ago

   AlphaZero has been made for perfect information games. That said, the Monte Carlo Tree Search in the library can be run with any agent that implements a value and policy function. So, while the AlphaZeroAgent in agents.py wouldn't fit
6. 6.▲
   by s-casci 3y ago

   Interesting, I didn't know about it... Modifying the existing environments' interfaces shouldn't be too difficult. Feel free to submit a PR!
7. 7.▲
   by s-casci 3y ago

   There certainly are other projects around AlphaZero, I'd say this is simpler and much more basic
8. 8.▲
   by s-casci 3y ago

   If you do that, please submit a PR!
9. 9.▲
   Show HN: Easily train AlphaZero-like agents on any environment you want(github.com)

   87 points by s-casci 3y ago|21 comments