Introduction

This repository mostly consists of third-party code related to

• (1) Move Evaluation in Go using Deep Convolutional Neural Networks
• (2) Mastering the game of Go with deep neural networks and tree search
• (3) Mastering the game of Go without human knowledge

Currently, the code implements supervised learning applied to high-level human play from the KGS server (data found here). This is mostly explored in (1) (although it is also used in (2)). The neural network used is a clone of the network from (3).

Data

To download the data, run

bin/download_kgs.sh

To setup the data for training, run

python -m deepgo prepare

Training

Run

python -m deepgo train --batch 512 --checkpoint checkpoint --keep_all_checkpoints --workers 8

Prototype C++ Version

04/08/2019 In deepgo/cpp/external/ wget https://download.pytorch.org/libtorch/nightly/cu100/libtorch-shared-with-deps-latest.zip

rm -rf build mkdir build cd build cmake -DCMAKE_PREFIX_PATH=/home/bryanhe/deepgo/cpp/external/libtorch .. make

https://discuss.pytorch.org/t/libtorch-cmake-issues/28246 grep culibos find .

pytorch/examples#506

export LD_LIBRARY_PATH=/home/bryanhe/deepgo/cpp/external/libtorch/:${LD_LIBRARY_PATH}