nshogi-engine: a USI shogi engine

nshogi-engine is a USI shogi engine built on the top of nshogi library. The engine implements an AlphaZero-style Monte-Carlo Tree Search (MCTS) algorithm, using multi-threading for search. In addition, this project includes a module for generating self-play data used for reinforcement learning in the AlphaZero style.

How to Build

Prerequisites

Ensure that the following are installed:

• make
• g++ or clang++
• For the TensorRT executor: CUDA and TensorRT
• For testing: googletest

Build Instructions

Open your terminal and run:

make [CUDA_ENABLED=1] EXECUTOR=(tensorrt|random) engine

Note: If you are using the TensorRT executor (EXECUTOR=tensorrt), you must specify CUDA_ENABLED=1.

The USI engine binary is built in ./build directory.

License

This repository is released under the MIT License.