Japanese
AobaFuribisha is a distributed Deep reinforcement learning for Shogi Ranging Rook without human knowledge.
If you are interested, please join us. Anyone can contribute using Google Colab.

GitHub Source and Windows binary. GitHub(Japanese top page)

2024-08-12 Web site is opened.

2026-03-13 19:29 JST(update every 30 minutes)
In past an hour   2, 8 games
In pash 24 hours 3, 35 games
Total 22027752 games. Latest weight= w2195. Next is in 372.2 hours. Thank you for your contribution!
In past   7000 games, average of move 166.4 moves, Sente winrate 0.558, draw 0.043
In past 100000 games, average of move 165.9 moves, Sente winrate 0.558, draw 0.054
Elo progress. Elo is based on floodgate.

AobaFuribisha 100playout/move vs Suisho5(7.50) 2000 nodes/move. 400 match games with even evaluated opening.

Game samples without noise. There is no ELO adjustment for black player.
Self-play games without noise. Each game uses same weight.

Game records
From
no000000000000.csa to
no000000100028.csa
 are generated by not using neural network, but random function.
The first game that is generated by neural network is
no000000100029.csa
128x10block, replay buffer is past 100,000 games.
Weights
Network size is 256 x 20 block (ResNet). AlphaZero style.
w002  ... 128x10b,minibatch  128, learning rate 0.01,     wd 0.0002, momentum 0.9,   100000 games.