An open-source implementation of the AlphaGoZero algorithm
Switch branches/tags
Nothing to show
Clone or download
Failed to load latest commit information.
cc Set CUDA device before creating builder (TensorRT 5 requirement). Oct 12, 2018
cluster V13 image builds (#484) Oct 6, 2018
minigui addressing review comments Oct 3, 2018
oneoffs Merge remote-tracking branch 'official/master' into more_refactoring11 Oct 3, 2018
proto Inference worker can optionally monitor a training checkpoint directo… Jul 12, 2018
ratings break pb export out of dual_net export (#481) Oct 3, 2018
rl_loop V13 image builds (#484) Oct 6, 2018
sgf Added comments that games are public domain Feb 1, 2018
testing update test docker image Oct 1, 2018
tests V13 image builds (#484) Oct 6, 2018
tools Moving .bazelrc according to best practices. Sep 21, 2018
.gitignore Moving .bazelrc according to best practices. Sep 21, 2018
.pylintrc Pylint and code license headers pass Oct 1, 2018 Add code of conduct. Jan 30, 2018 Several python cleanups. Feb 1, 2018
LICENSE initial commit Nov 10, 2017 Update (#488) Oct 6, 2018 update RESULTS describing v5 and v7a (#236) May 24, 2018
WORKSPACE Adding TensorRT DualNet implementation. TensorRT requires the input Oct 12, 2018 Several python cleanups. Feb 1, 2018 Pylint and code license headers pass Oct 1, 2018 Glue all the RL loop bits back together (#315) Jul 16, 2018 Replaces pygtp with our own simple GTP engine. (#214) May 9, 2018 Correctly log summary.scalar step when training on TPU Oct 5, 2018 fix up bugs with main(argv) Sep 21, 2018 Cleanups (#308) Jul 10, 2018 Flake8 fixes (#320) Aug 10, 2018 Factor out; unify handling of --verbose flags Sep 8, 2018 address review comments Oct 3, 2018 Improvements to several debug print statements (#366) Aug 11, 2018 V13 image builds (#484) Oct 6, 2018 V13 image builds (#484) Oct 6, 2018 Add tiebreak to pick_move (#464) Sep 26, 2018
notes.txt Create dockerfiles and deployments for tpu selfplay (#310) Jul 13, 2018 Replaces pygtp with our own simple GTP engine. (#214) May 9, 2018 Separate out; refactor --model_dir vs --working_dir; combine… Sep 8, 2018
requirements-analysis.txt Added requirements-analysis.txt, added tqdm for long for loops Feb 1, 2018
requirements.txt a couple of fixes Oct 10, 2018 Another pass over Make a standalone utility Oct 3, 2018 Flake8 fixes (#320) Aug 10, 2018 Merge branch 'master' into more_refactoring10 Oct 1, 2018 Flake8 fixes (#320) Aug 10, 2018 Add documentation for prep_flags and flagfile strategy Oct 1, 2018 Pylint and code license headers pass Oct 1, 2018
tsconfig.json address review comments Oct 3, 2018 pylint over test files (#380) Aug 14, 2018 Another pass over Make a standalone utility Oct 3, 2018

Minigo: A minimalist Go engine modeled after AlphaGo Zero, built on MuGo

This is a pure Python implementation of a neural-network based Go AI, using TensorFlow. While inspired by DeepMind's AlphaGo algorithm, this project is not a DeepMind project nor is it affiliated with the official AlphaGo project.

This is NOT an official version of AlphaGo

Repeat, this is not the official AlphaGo program by DeepMind. This is an independent effort by Go enthusiasts to replicate the results of the AlphaGo Zero paper ("Mastering the Game of Go without Human Knowledge," Nature), with some resources generously made available by Google.

Minigo is based off of Brian Lee's "MuGo" -- a pure Python implementation of the first AlphaGo paper "Mastering the Game of Go with Deep Neural Networks and Tree Search" published in Nature. This implementation adds features and architecture changes present in the more recent AlphaGo Zero paper, "Mastering the Game of Go without Human Knowledge". More recently, this architecture was extended for Chess and Shogi in "Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm". These papers will often be abridged in Minigo documentation as AG (for AlphaGo), AGZ (for AlphaGo Zero), and AZ (for AlphaZero) respectively.

Goals of the Project

  1. Provide a clear set of learning examples using Tensorflow, Kubernetes, and Google Cloud Platform for establishing Reinforcement Learning pipelines on various hardware accelerators.

  2. Reproduce the methods of the original DeepMind AlphaGo papers as faithfully as possible, through an open-source implementation and open-source pipeline tools.

  3. Provide our data, results, and discoveries in the open to benefit the Go, machine learning, and Kubernetes communities.

An explicit non-goal of the project is to produce a competitive Go program that establishes itself as the top Go AI. Instead, we strive for a readable, understandable implementation that can benefit the community, even if that means our implementation is not as fast or efficient as possible.

While this product might produce such a strong model, we hope to focus on the process. Remember, getting there is half the fun. :)

We hope this project is an accessible way for interested developers to have access to a strong Go model with an easy-to-understand platform of python code available for extension, adaptation, etc.

If you'd like to read about our experiences training models, see

To see our guidelines for contributing, see

Getting Started

This project assumes you have the following:

  • virtualenv / virtualenvwrapper
  • Python 3.5+
  • Docker
  • Cloud SDK
  • Bazel v0.11 or greater

The Hitchhiker's guide to python has a good intro to python development and virtualenv usage. The instructions after this point haven't been tested in environments that are not using virtualenv.

pip3 install virtualenv
pip3 install virtualenvwrapper

Install TensorFlow

First set up and enter your virtualenv and then the shared requirements:

pip3 install -r requirements.txt

Then, you'll need to choose to install the GPU or CPU tensorflow requirements:

  • GPU: pip3 install "tensorflow-gpu>=1.9,<1.10".
    • Note: You must install CUDA 9.0. for Tensorflow 1.5+.
  • CPU: pip3 install "tensorflow>=1.9,<1.10".

Setting up the Environment

You may want to use a cloud project for resources. If so set:


Then, running

source cluster/

will set up other environment variables defaults.

Running unit tests


Automated Tests

Test Dashboard

To automatically test PRs, Minigo uses Prow, which is a test framework created by the Kubernetes team for testing changes in a hermetic environment. We use prow for running unit tests, linting our code, and launching our test Minigo Kubernetes clusters.

You can see the status of our automated tests by looking at the Prow and Testgrid UIs:


All commands are compatible with either Google Cloud Storage as a remote file system, or your local file system. The examples here use GCS, but local file paths will work just as well.

To use GCS, set the BUCKET_NAME variable and authenticate via gcloud login. Otherwise, all commands fetching files from GCS will hang.

For instance, this would set a bucket, authenticate, and then look for the most recent model.

# When you first start we recommend using our minigo-pub bucket.
# Later you can setup your own bucket and store data there.
export BUCKET_NAME=minigo-pub/v9-19x19
gcloud auth application-default login
gsutil ls gs://$BUCKET_NAME/models | tail -4

Which might look like:


These four files comprise the model. Commands that take a model as an argument usually need the path to the model basename, e.g. gs://$BUCKET_NAME/models/000737-fury

You'll need to copy them to your local disk. This fragment copies the files associated with $MODEL_NAME to the directory specified by MINIGO_MODELS:

mkdir -p $MINIGO_MODELS/models
gsutil ls gs://$BUCKET_NAME/models/$MODEL_NAME.* | \
       gsutil cp -I $MINIGO_MODELS/models


To watch Minigo play a game, you need to specify a model. Here's an example to play using the latest model in your bucket

python3 \
  --verbose=2 \
  --num_readouts=400 \

where READOUTS is how many searches to make per move. Timing information and statistics will be printed at each move. Setting verbosity to 3 or higher will print a board at each move.

Playing Against Minigo

Minigo uses the GTP Protocol, and you can use any gtp-compliant program with it.

# Latest model should look like: /path/to/models/000123-something
LATEST_MODEL=$(ls -d $MINIGO_MODELS/* | tail -1 | cut -f 1 -d '.')
BOARD_SIZE=19 python3 --load_file=$LATEST_MODEL --num_readouts=$READOUTS --verbose=3

After some loading messages, it will display GTP engine ready, at which point it can receive commands. GTP cheatsheet:

genmove [color]             # Asks the engine to generate a move for a side
play [color] [coordinate]   # Tells the engine that a move should be played for `color` at `coordinate`
showboard                   # Asks the engine to print the board.

One way to play via GTP is to use gogui-display (which implements a UI that speaks GTP.) You can download the gogui set of tools at See also documentation on interesting ways to use GTP.

gogui-twogtp -black 'python3 --load_file=$LATEST_MODEL' -white 'gogui-display' -size 19 -komi 7.5 -verbose -auto

Another way to play via GTP is to watch it play against GnuGo, while spectating the games:

BLACK="gnugo --mode gtp"
WHITE="python3 --load_file=$LATEST_MODEL"
TWOGTP="gogui-twogtp -black \"$BLACK\" -white \"$WHITE\" -games 10 \
  -size 19 -alternate -sgffile gnugo"
gogui -size 19 -program "$TWOGTP" -computer-both -auto

Training Minigo


The following sequence of commands will allow you to do one iteration of reinforcement learning on 9x9. These are the basic commands used to produce the models and games referenced above.

The commands are

  • bootstrap: initializes a random model
  • selfplay: plays games with the latest model, producing data used for training
  • train: trains a new model with the selfplay results from the most recent N generations.

Training works via tf.Estimator; a working directory manages checkpoints and training logs, and the latest checkpoint is periodically exported to GCS, where it gets picked up by selfplay workers.

Configuration for things like "where do debug SGFs get written", "where does training data get written", "where do the latest models get published" are managed by the helper scripts in the rl_loop directory. Those helper scripts execute the same commands as demonstrated below. Configuration for things like "what size network is being used?" or "how many readouts during selfplay" can be passed in as flags. The utility helps ensure all parts of the pipeline are using the same network configuration.

All local paths in the examples can be replaced with gs:// GCS paths, and the Kubernetes-orchestrated version of the reinforcement learning loop uses GCS.


This command initializes your working directory for the trainer and a random model. This random model is also exported to --model-save-path so that selfplay can immediately start playing with this random model.

If these directories don't exist, bootstrap will create them for you.

export MODEL_NAME=000000-bootstrap
BOARD_SIZE=19 python3 \
  --work_dir=estimator_working_dir \


This command starts self-playing, outputting its raw game data as tf.Examples as well as in SGF form in the directories.

BOARD_SIZE=19 python3 \
  --load_file=outputs/models/$MODEL_NAME \
  --num_readouts 10 \
  --verbose 3 \
  --selfplay_dir=outputs/data/selfplay \
  --holdout_dir=outputs/data/holdout \


This command takes a directory of tf.Example files from selfplay and trains a new model, starting from the latest model weights in the estimator_working_dir parameter.

Run the training job:

BOARD_SIZE=19 python3 \
  outputs/data/selfplay/* \
  --work_dir=estimator_working_dir \

At the end of training, the latest checkpoint will be exported to. Additionally, you can follow along with the training progress with TensorBoard. If you point TensorBoard at the estimator working directory, it will find the training log files and display them.

tensorboard --logdir=estimator_working_dir


It can be useful to set aside some games to use as a 'validation set' for tracking the model overfitting. One way to do this is with the validate command.

Validating on holdout data

By default, MiniGo will hold out 5% of selfplay games for validation. This can be changed by adjusting the holdout_pct flag on the selfplay command.

With this setup, rl_loop/ will validate on the same window of games that were used to train, writing TensorBoard logs to the estimator working directory.

Validating on a different set of data

This might be useful if you have some known set of 'good data' to test your network against, e.g., a set of pro games. Assuming you've got a set of .sgfs with the proper komi & boardsizes, you'll want to preprocess them into the .tfrecord files, by running something similar to

import preprocessing
filenames = [generate a list of filenames here]
for f in filenames:
        preprocessing.make_dataset_from_sgf(f, f.replace(".sgf", ".tfrecord.zz"))

Once you've collected all the files in a directory, producing validation is as easy as

BOARD_SIZE=19 python3 \
  validation_files/ \
  --work_dir=estimator_working_dir \

The will glob all the .tfrecord.zz files under the directories given as positional arguments and compute the validation error for the positions from those files.

Running Minigo on a Kubernetes Cluster

See more at cluster/