Game-AI Forum

flok wrote: ↑Wed Mar 15, 2023 6:45 pm Are you planning cutechess-cli functionality eventually? Like that it can do tournaments?

I am not planning this in the short term.

I have recently discovered that the Ataxx community is meeting in the Engine Programming discord. Here is an invite: https://discord.gg/Ym5cMVA7Nv There is an active community there. Note in particular: https://github.com/hzyhhzy/KataGo, a fork of KataGo, that supports multiple games, including Atax...

The 1.5.2 release is online: https://github.com/Remi-Coulom/gogui Don't use Quaqua anymore, which should fix some crashes on macOS (issue #18). Support for large boards up to 52x52. Support for games with moving pieces such as chess or ataxx Drag-and-drop of sgf files (thanks to @zakki) Fix wrong pl...

Here is an update on the 2023 International Computer Games Association events. The Advances in Computer Games conference will be held online November 28-30. The call for papers can be found here . We want to hear about your exciting, innovative work in game AI! The Computer Olympiad will be held at...

By Alexandre Trudeau and Michael Bowling https://arxiv.org/pdf/2302.12359.pdf AlphaZero is a self-play reinforcement learning algorithm that achieves superhuman play in chess, shogi, and Go via policy itera- tion. To be an effective policy improvement operator, AlphaZero’s search requires accurate v...

https://github.com/deepmind/mctx Mctx is a library with a JAX-native implementation of Monte Carlo tree search (MCTS) algorithms such as AlphaZero, MuZero, and Gumbel MuZero. For computation speed up, the implementation fully supports JIT-compilation. Search algorithms in Mctx are defined for and op...

https://www.lesswrong.com/posts/FF8i6SLfKb4g7C4EL/inside-the-mind-of-a-superhuman-go-model-how-does-leela-zero-2 We did some interpretability on Leela Zero, a superhuman Go model. With a technique similar to the logit lens, we found that the residual structure of Leela Zero induces a preferred basis...

flok wrote: ↑Tue Feb 28, 2023 5:10 pmEventually I'll move onto mcts and friends

Well, the idea can be useful in an alpha-beta search as well. If you give a bonus to a group with eyes, your search will work better.

I see. One very important heuristic is that you must somehow recognize eyes, and not fill them up. It is easy to detect an eye shape by looking at a 3x3 square around a point.

Have fun.

Rémi

I also run test-matches at home with my homegrown matcher (stats generated with ordo): # PLAYER : RATING POINTS PLAYED (%) 1 GnuGO level 0 : 3879.9 31492.0 31684 99 2 AmiGo : 3271.6 25484.5 31661 80 3 Stop : 2293.4 17685.5 31705 56 4 Daffy Baduck : 1997.6 11963.0 31657 38 5 dellabaduck_0e55ae0_1th ...