sigmazero

Implementation of the AlphaZero algorithm.

Feature highlights:

• Fast generation of legal moves using custom chess library libchess.
• Distributed training with one training process and multiple self-play processes.
• Batched Monte Carlo tree search.
• Communication with user interfaces and lichess.org using UCI.

Originally a group project in TDDE19 at Linköping University.

setup

First time:

# clone repository
git clone --recurse-submodules git@gitlab.liu.se:TDDE19-2021-1/tjack.git
# or `git submodule init && git submodule update` after regular clone

# use virtual environment (optional)
python3 -m venv venv
source venv/bin/activate

# install requirements
pip install -r requirements.txt

# setup build directory
meson build
# to choose non-default compiler like clang, do `CXX=clang++ meson <builddir>`
# to select build type, do `meson <builddir> --buildtype={debug,release}`
# if the build fails with some torch error, try `meson <builddir> -D_GLIBCXX_USE_CXX11_ABI=0

To build:

cd build
ninja

running

Run one selfplayer and one trainer:

model=model.pt
./selfplay $model | ./training $model

Run multiple selfplayers and one trainer:

model=model.pt
./training $model <(./selfplay $model) <(./selfplay $model) <(./selfplay $model)

lichess

Play with bot on lichess via lichess-bot:

# download lichess-bot somewhere
git clone https://github.com/ShailChoksi/lichess-bot.git <path/to/lichess-bot>

# export bot token token
export LICHESS_BOT_TOKEN=<token>

# run engine
cd <engine>
python3 <path/to/lichess-bot>/lichess-bot.py

Then challenge here:

https://lichess.org/?user=tjack-bot#friend

evaluate

Estimate Elo rating of of engine:

python3 scripts/elo_est.py build/<engine>