Gokaku is a computer Shogi program developed using deep reinforcement learning from randomly generated game records. The deep learning model of Gokaku incorporates large-kernel depthwise convolutions and multi-head attention, enabling it to efficiently grasp the overall state of the board. Its reinforcement learning procedure is inspired by approaches used in Katago and Gumbel AlphaZero, allowing it to efficiently learn a wide range of patterns. This page shows the improvement in playing strength through reinforcement learning and the results of self-play matches.
Gokaku is a sibling program of the computer Go program Maru. Gokaku shares the same deep learning model architecture, search algorithm, and reinforcement learning methodology as Maru.
If you have any questions, please contact Atsushi Takeda.
2025/05/05: Reinforcement learning of Gokaku 2.0 has started from game records with random moves.
2025/05/05: Training of the 1st model, which consists of 6 blocks and 96 channels (b5n1c96), has started.
2025/05/10: The 1st model training has been stopped when the number of generated game records reached 1M.
2025/05/10: Training of the 2nd model, which consists of 10 blocks and 128 channels (b8n2c128), has started.
2025/05/14: The Floodgate rating of the 1st model with the visits of 500 reached 2274.
2025/05/19: The Floodgate rating of the 1st model with the policy only reached 1883.
2025/05/23: The 2nd model training has been stopped when the number of generated game records reached 3M.
2025/05/23: Training of the 3rd model, which consists of 15 blocks and 192 channels (b12n3c192), has started.
2025/05/28: The Floodgate rating of the 2nd model with the visits of 500 reached 2568.
2025/06/01: The Floodgate rating of the 2nd model with the policy only reached 2032.
2025/06/13: The 3rd model training has been stopped when the number of generated game records reached 5M.
2025/06/14: Training of the 4th model, which consists of 20 blocks and 256 channels (b16n4c256), has started.
2025/06/21: The Floodgate rating of the 3rd model with the visits of 500 reached 2799.
2025/06/29: The Floodgate rating of the 3rd model with the policy only reached 2021.
Copyright (C) 2025 Atsushi TAKEDA