{
    "info": {
        "author": "Jarno Sepp\u00e4nen",
        "author_email": "azalea@meit.si",
        "bugtrack_url": null,
        "classifiers": [
            "License :: OSI Approved :: BSD License",
            "Operating System :: OS Independent",
            "Operating System :: POSIX :: Linux",
            "Programming Language :: Python :: 3"
        ],
        "description": "\n# Azalea\n\n> playing to learn to play\n\nAzalea is a reinterpretation of the [AlphaZero game AI](https://en.wikipedia.org/wiki/AlphaZero)\nlearning algorithm for the [Hex board game](https://en.wikipedia.org/wiki/Hex_(board_game)).\n\n## Features\n\n* Straightforward reimplementation of the AlphaZero algorithm except\n  for MCTS parallelization (see below)\n* Pre-trained model for Hex board game\n* Fast MCTS implementation through Numba JIT acceleration.\n* Fast Hex game move generation implementation through Numba.\n* Parallelized self play to saturate Nvidia V100 GPU during training\n* AI policy evaluation through round robin tournament, also parallelized\n* Tested on Ubuntu 16.04\n* Requires Python 3.6 and PyTorch 0.4\n\n## Differences to published AlphaZero\n\n* Single GPU implementation only - tested on Nvidia V100, with 8 CPU's\n  for move generation and MCTS, and 1 GPU for the policy network.\n* Only Hex game is implemented, though the code supports adding more\n  games. Two components are needed for a new game: move generator and\n  policy network, with board input and moves output adjusted to the\n  new game.\n* MCTS simulations are not run in parallel threads, but instead,\n  self-play games are played in parallel processes. This is to avoid\n  the need for a multi-threaded MCTS implementation while still\n  maintaining fast training speed and saturating the GPU.\n* MCTS simulation and board evaluations are batched according to\n  `search_batch_size` config parameter. \"Virtual loss\" is used\n  as in AlphaZero, to increase search diversity.\n\n# Installation\n\nClone the repository and install dependencies with Conda:\n\n    git clone https://github.com/jseppanen/azalea.git\n    conda env create -n azalea\n    source activate azalea\n\nThe default `environment.yml` installs GPU packages but you can choose\n`environment-cpu.yml` for testing on a laptop.\n\n## Playing against pretrained model\n\n    python play.py models/hex11-20180712-3362.policy.pth\n\nThis will load the model and start playing, asking for your move. The\ncolumns are labeled a\u2013k and rows 1\u201311. The first player, playing `X`'s,\nis trying to draw a vertical connected path through the board, while the\nsecond player, with `O`'s, is drawing a horizontal path.\n\n```\nO O O O X . . . . . . \n . . . . . . . . . . . \n  . . . . . . . . . . . \n   . . . . X . . . . . . \n    . . . . . X . . . . . \n     . . . . . . . . . . . \n      . . . . X . . . . . . \n       . . . . . . . . . . . \n        . . . X . . . . . . . \n  x      . . . . . . . . . . . \n o\\\\      . . . . . . . . . . . \nlast move: e1\nYour move? \n```\n\n## Model training\n\n    python train.py --config config/hex11_train_config.yml --rundir runs/train\n\n## Model comparison\n\n    python compare.py --config config/hex11_eval_config.yml --rundir runs/compare <mode1> <model2> [model3] ...\n\n## Model selection\n\n    python tune.py\n\n## References\n\n* [Mastering the Game of Go without Human Knowledge](https://deepmind.com/documents/119/agz_unformatted_nature.pdf)\n* [Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm](https://arxiv.org/abs/1712.01815)\n\n\n",
        "description_content_type": "text/markdown",
        "docs_url": null,
        "download_url": "",
        "downloads": {
            "last_day": -1,
            "last_month": -1,
            "last_week": -1
        },
        "home_page": "https://github.com/jseppanen/azalea",
        "keywords": "",
        "license": "BSD 3-Clause License",
        "maintainer": "",
        "maintainer_email": "",
        "name": "azalea",
        "package_url": "https://pypi.org/project/azalea/",
        "platform": "",
        "project_url": "https://pypi.org/project/azalea/",
        "project_urls": {
            "Homepage": "https://github.com/jseppanen/azalea"
        },
        "release_url": "https://pypi.org/project/azalea/0.1.0/",
        "requires_dist": [
            "numba",
            "numpy",
            "torch",
            "scipy",
            "pyyaml",
            "click",
            "tensorboardX"
        ],
        "requires_python": "",
        "summary": "Hex board game AI with self-play learning based on the AlphaZero algorithm",
        "version": "0.1.0"
    },
    "last_serial": 4137403,
    "releases": {
        "0.1.0": [
            {
                "comment_text": "",
                "digests": {
                    "md5": "862f162e30c023903fdc136022dd6c1c",
                    "sha256": "26d37f27c2e118c2f3d742efb87f152a9c24b1c5b691d06e9d4d2c2ea37be9c4"
                },
                "downloads": -1,
                "filename": "azalea-0.1.0-py3-none-any.whl",
                "has_sig": false,
                "md5_digest": "862f162e30c023903fdc136022dd6c1c",
                "packagetype": "bdist_wheel",
                "python_version": "py3",
                "requires_python": null,
                "size": 29059,
                "upload_time": "2018-08-05T10:08:39",
                "url": "https://files.pythonhosted.org/packages/83/10/f4a97bbd32aaf1467e489c658ceb7ef841f5ae00dbe9b04375db5f0971f4/azalea-0.1.0-py3-none-any.whl"
            },
            {
                "comment_text": "",
                "digests": {
                    "md5": "c558c419e4d36d6b33996d8e06a18372",
                    "sha256": "510629ce818a6cbede9d2837b42779c682b4f3a7d17f9c40199d6238b638cd45"
                },
                "downloads": -1,
                "filename": "azalea-0.1.0.tar.gz",
                "has_sig": false,
                "md5_digest": "c558c419e4d36d6b33996d8e06a18372",
                "packagetype": "sdist",
                "python_version": "source",
                "requires_python": null,
                "size": 22493,
                "upload_time": "2018-08-05T10:08:41",
                "url": "https://files.pythonhosted.org/packages/a6/b2/0f8876fd07cdace277d2f6c26005b75a1db0ae12410de16646f933b05a38/azalea-0.1.0.tar.gz"
            }
        ]
    },
    "urls": [
        {
            "comment_text": "",
            "digests": {
                "md5": "862f162e30c023903fdc136022dd6c1c",
                "sha256": "26d37f27c2e118c2f3d742efb87f152a9c24b1c5b691d06e9d4d2c2ea37be9c4"
            },
            "downloads": -1,
            "filename": "azalea-0.1.0-py3-none-any.whl",
            "has_sig": false,
            "md5_digest": "862f162e30c023903fdc136022dd6c1c",
            "packagetype": "bdist_wheel",
            "python_version": "py3",
            "requires_python": null,
            "size": 29059,
            "upload_time": "2018-08-05T10:08:39",
            "url": "https://files.pythonhosted.org/packages/83/10/f4a97bbd32aaf1467e489c658ceb7ef841f5ae00dbe9b04375db5f0971f4/azalea-0.1.0-py3-none-any.whl"
        },
        {
            "comment_text": "",
            "digests": {
                "md5": "c558c419e4d36d6b33996d8e06a18372",
                "sha256": "510629ce818a6cbede9d2837b42779c682b4f3a7d17f9c40199d6238b638cd45"
            },
            "downloads": -1,
            "filename": "azalea-0.1.0.tar.gz",
            "has_sig": false,
            "md5_digest": "c558c419e4d36d6b33996d8e06a18372",
            "packagetype": "sdist",
            "python_version": "source",
            "requires_python": null,
            "size": 22493,
            "upload_time": "2018-08-05T10:08:41",
            "url": "https://files.pythonhosted.org/packages/a6/b2/0f8876fd07cdace277d2f6c26005b75a1db0ae12410de16646f933b05a38/azalea-0.1.0.tar.gz"
        }
    ]
}