"Black-white No-score 2024" For games which black and white plays alternately

For board games with multi-stage moves or irregular move order For example: Chess(regard one move as two stages: choose and place), Amazons(3 stages), Ataxx(sometimes 1 stage, some times 2 stages), ConnectSix, ...

Latest KataGo engine for Gomoku and Renju. Also called "KataGomo" and "KataGomoku".
Recently a b28c512nbt net is training under the sponsorship of Zhizi(智子)(www.zhizigo.cn www.zhizigomoku.cn), which provides online paid KataGo services.
The training started in 2024.5 and will finish in 2024.11. ~30*RTX4090 for 10 hours every day. The strongest net is only available on Zhizi.

An experiment of combining NNUE and NN. (not very success) Gomoku-NNUE: A fast and small network with special structure for gomoku.
Gomoku-NNUE is firstly created by me (https://github.com/hzyhhzy/gomoku_nnue/tree/multiRules) and improved by dblue and used in Rapfi(https://github.com/dhbloo/rapfi)
Input the result of a small MCTS (~300 nodes) of NNUE to the big katago net.
Result: The katago net become "as strong as NNUE+MCTS" in a very short time, however further progress is very slow. If use transfer-training from a trained katago net, it become slightly stronger than the origin net with same nodes, but not worth its large CPU cost.

A branch to test MCTS of NNUE. Regarding the result of a small MCTS (~300 nodes) of NNUE as a "meta-node", then apply MCTS to meta-nodes. Katago NN is not used here. Only to use katago's MCTS

https://en.wikipedia.org/wiki/Hex_(board_game)

Latest KataHex code Hex2024 with move limit settings, to estimate the length of the optimal game of Hex board on N*N board
Result: f(N) ~ 0.6 * N^1.9
https://mathoverflow.net/questions/302821/length-of-optimal-play-in-hex-as-a-function-of-size/477037#477037

Latest release of Katago for Hex game (KataHex) https://github.com/hzyhhzy/KataGo/releases/tag/Hex_20240812

A special engine for "edge template". Special network is needed.
This branch is to estimate the min width of 8~11th row "edge template" of Hex
Result: f(n)>=9*n-46 for n>=8, and probably f(n)=9*n-46 for n<=11
https://www.hexwiki.net/index.php/Open_problems_about_edge_templates

A special engine for single-stone escape problems (HexTemplate branch is more specific for "edge template"). Special network is needed.
Some analysis and results: https://mathoverflow.net/questions/470376/connection-properties-of-a-single-stone-on-an-infinite-hex-board

Including Caro, ConnectSix, and maybe some math problems

KataGo for Connect Six The 2022 ver has shown that black has great advantage on 21x21 or larger boards, but 19x19 was unknown. Recently @eeoo has found that black has advantage on 19x19, so I started a new run in 2024.8. I decide to strictly prove that black wins in Connect Six on 19x19 boards, collaborating with dblue using his ConnectSix Rapfi.

Caro is a modified Gomoku rule, the only difference is that fives with two terminals blocked are not a win (OXXXXXO is not a win)
Latest release: https://github.com/hzyhhzy/KataGo/releases/tag/Caro_20240804

We all know black wins in 5-in-a-row(Gomoku) on an infinite board. And more than 5 in a row is very difficult which means easy to draw.
Although it's very easy even for a beginner to draw in 6-in-a-row game, 7-in-a-row has not been proven to be a draw in infinite boards and remains to be an open problem for decades, showing a big gap between "theory" and "practice".
https://arxiv.org/abs/2107.05363 This article shows a method to reduce the infinite board to 4*n boards. If we can prove that there exist such one n, that white can draw on a 4*n board (with some kinds of rules), then "7-in-a-row is a draw" is proved. In this article the authors tried n<=14 but failed.
This branch is to solve the game mentioned in this article on 4*n boards with large n.
However katago shows that the min n is around 30, which means it is almost impossible to strictly solve the 4*n game. So this method is almost dead.
More details was written here: https://mathoverflow.net/questions/471302/can-gomokufive-in-a-row-draw-on-an-infinite-board-what-about-other-m-n-k-game

It's a game that black and white alternately play edges. It is hard to represent edges in Katago, so I uses a (2H-1)*(2W-1) board to represent the Dots-and-boxes board Result: This is a very chaotic game, no strategies, just brute-force search. (probably because it is almost an "impartical game"). Katago can learn almost nothing in the first 1/3 game and there is only some strategies near the end of the game

2023.1.8 Engine based on Katago 1.12 start Modifying

2022.1 Engine based on Katago TensorRT+GraphSearch

Updated on 2022.4.9

Gomoku/Renju: "GomDevVCN" branch
https://github.com/hzyhhzy/KataGo/tree/GomDevVCN

TODO:
先做改动小的，再做改动大的

完全没改的分支：Go2022

Updated on 2021.4.5 Now Gomoku/Renju code is uploading in "Gomoku" branch.

----------- 2020.7.26 ----------

Now there are gomoku(freestyle,standard,renju,caro), reversi, connect6, breakthrough, hex, "four in a row", Chinese checkers, and many board games few people play.

Now it's the strongest program on this games as I know: all kinds of gomoku(compared with Embryo) connect6(compared with gzero) hex(compared with gzero and) Chinese checkers(compared with Shangxin Tiaoqi)