Skip to content

Latest commit

 

History

History
59 lines (46 loc) · 1.74 KB

README.md

File metadata and controls

59 lines (46 loc) · 1.74 KB
Raw

gobl

Go machine learning library

Build Status

This is the base of a machine learning system for the game of Go. It provides components broken up into the following sections

goblb -- utilities for dealing with boards and playing games
gobld -- higher-level driver utilities for playing sgf's and
         iterating through games
gobll -- abstract classes for ML models
goblr -- utilities for creating representations or
         features used in the ML models
gobls -- utilities for parsing sgf files
goblu -- low-level utilities

Specifically, goblr provides utilities for creating the tensors that be used in the tensorflow, namely

  • liberties -- a group of matrices parameterized in the perspective (whose turn it is) and a level. Each number of liberties up to that level will have its own matrix, only the groups on the board with a number of liberties of that size will have a value of 1.0 in their respective locations in the matrix.
  • ko -- a matrix that has a value of 1.0 only in a place where there is currently an active ko.

Building

To build the library run

make lib

which will populate gobl/include with the library headers.

If the user has gtest installed on their system, a test suite can be built by setting GTEST_DIR to the location that gtest is installed

GTEST_DIR=/my/gtest/dir make gtest

Example

An example main has been provided that allows the user to play games against himself. There is also an option --parse filename to parse and then print an sgf file.

This can be built with

make main