cboittin/go-score-estimator

Neural network (using tensorflow) to predict the final score of a go game from the board position.

go-sequence

Neural network (using tensorflow) to predict the winner of a go game from the board position. Currently overfit on random positions of 10k high level games on tygem. Training for 400k games in is progress (using https://github.com/SThornewillvE/Pet-Project---Tygem-Fuseki-Web-Scraper-using-Python)

How to use

Requires python 3 with the numpy and tensorflow modules.

Training

Create an sgf folder with a collection of sgf files. Then, run parseSgf.py, which will create the records dataset from the sgf files. Only files that are valid and give a score will be parsed, the rest will be discarded.

Finally, run py scoreEstimation.py to train the network using the records dataset.

The program will load the session stored in the model folder. Rename it to archive it and start training from a newer one.

Testing

Simply run py scoreEstimation.py $SGF where $SGF is the path to your sgf file.

A positive score indicates black has a lead on the board, a negative one indicates white has a lead on the board. Komi is not taken into account and has to be subtracted from the score.

go-score-estimator's Issues

Can you provide a use of demo？

@cboittin
Thank you very much for your code. Can you provide a way to use it?