Skip to content

Latest commit

 

History

History
65 lines (48 loc) · 1.27 KB

README.md

File metadata and controls

65 lines (48 loc) · 1.27 KB
Raw

Setup

Dependencies

cmake is required.

Python

with conda environment

conda create -n env python=3.8.2
pip install -r requirements.txt

c++

Download tensorflow c api from - https://www.tensorflow.org/install/lang_c and follow steps:

  • setup
  • extract
  • linker

build c++ files

mkdir -p build
cd build
cmake ..
make

After that 2 executables are compiled self_play and eval

from root directory

./build/self_play id

Will load model with id=id and play 300 games using model.

from root directory

./build/eval id1 id2 nr_sims

Will load models with id1 and id2 and play one game using nr_sims simulation for each monte carlo tree evaluation.

Human can play with ai using: from root directory

./build/eval id1

Model is then playing as 'x' using around 4200 simulations.

from root directory

python self_play_train.py id

Will start training procces from id. First model is either created if id=0 or loaded and trained on last 10 models data. Then new model with id+1 self play 300 games. This goes until stopped.

Models are excepted to be stored in models/mini-zero-id

Data from selfplay of model i is stored in data/i/

file utils.py provides some utilities to analise model.

#TODO: add example here or as jupyter notebook.