Repository files navigation

Building tensorflow is a little bit complicate for me. So I made this.

• No tensorflow requirement
• No changes to AQ original code
• [optional] CUDA 7.0/8.0/9.0 CUDNN 6.0/7.0
• [optional] BLAS (OpenBLAS/Intel MKL/..)

mkdir buildcd buildcmake .. or cmake .. -G "Visual Studio 14 2015 Win64" -T host=x64 (windows)cmake --build . --config Release

• NN forwarding using modified dlib dnn (https://github.com/davisking/dlib)
• TF Graph parsing using modified protobuf (https://github.com/google/protobuf)
• add a CMakeLists.txt

AQ is a program of Go game with level of top players.
CGOS rating: 3952 (BayesElo)

• OS: 64-bit Windows 7 or later (NOT tested on 8.1/10)
• GPU: Nvidia GPU withCUDA capability of >=3.0
• CPU: CPU with SSE 4.2

• OS: 64-bit Windows 7 or later (NOT tested on 8.1/10)
• CPU: CPU with SSE 4.2

• OS: 64-bit Linux
• GPU: Nvidia GPU withCUDA capability of >=3.0
• CPU: CPU with SSE 4.2

Get themhere.

Set hardware and time control etc. in 'aq_config.txt.'

First of all, it is not recommended that users connect the released version as it is to CGOS. The developer is well aware of the capabilities of the released versions andtested it in advance with4thread/1GPU: i7-6700/GTX1080.
Of course, if you made your own changes to the source code or pb files, that is welcome. Please check your AQ's rating and send pull request!

The recommended settings are as follows.

-main time[sec] =900  -byoyomi[sec] =0  -emergency time[sec] =15 #set 60 when connecting from outside Japan

!!!Caution!!! This version was trained in Komi = 6.5 for the Japanese rule. So, AQ often loses 0.5 point at Black, but that is inevitable.

GoGui is a graphical interface to Go-engines (programs without own GUI), which use the Go Text Protocol (GTP).
See the 'GTP Shell' console to know AQ's thinking log.

command:(install directory)/AQ

(Ex.)  /home/user/gogui-1.4.9/AQ/AQ

command:(install directory)\AQ.exe

(Ex.)  C:\Users\user\gogui-1.4.9\AQ\AQ.exe

Build withbazel andTensorFlow.
Seethis instruction.

Build withcmake andTensorFlow.
Seethis instruction.

MIT

Yu Yamaguchi