From 00fe41d2e7e40959b28bc2be30cf7ffeda687813 Mon Sep 17 00:00:00 2001 From: Jonathan Laurent Date: Mon, 30 Dec 2024 12:15:43 +0100 Subject: [PATCH] Update redesign/README.md Add an update in the redesign folder. --- redesign/README.md | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/redesign/README.md b/redesign/README.md index e7563fc3..22bcdb81 100644 --- a/redesign/README.md +++ b/redesign/README.md @@ -1,5 +1,9 @@ # AlphaZero.jl Redesign +## Update + +Andreas Spanopoulos made great [progress](https://github.com/AndrewSpano/AlphaZero.jl/tree/dev/redesign) on this redesign project during his 2023 GSOC. Fabrice Rosay ran promising experiments and suggested [improvements](https://github.com/jonathan-laurent/AlphaZero.jl/issues/217). Some work remains to integrate those contributions, add proper documentation and reach feature parity with the original AlphaZero.jl library. + ## Redesign Objectives - A codebase that is more accessible and easier to read @@ -102,4 +106,4 @@ By executing code directly in the editor window, the whole stack trace gets high - [Werner Duvaud's implementation](https://github.com/werner-duvaud/muzero-general) - [Duvaud's Tictactoe Params](https://github.com/werner-duvaud/muzero-general/blob/master/games/tictactoe.py) -Note that in the MuZero pseudocode, they seem to be updating the network every 1000 batch updates (batches have size 2048). There are 1e6 updates in total so this makes 1000 iterations. The buffer is surprisingly small with 1e6 samples. \ No newline at end of file +Note that in the MuZero pseudocode, they seem to be updating the network every 1000 batch updates (batches have size 2048). There are 1e6 updates in total so this makes 1000 iterations. The buffer is surprisingly small with 1e6 samples.