A tool to help train reinforcement learning models to handle novel environments.

Polycraft AI Lab consists of a wrapper for Polycraft World game environments. These environments can used to train RL models that respond to novel tasks and scenarios.

First, download Polycraft AI Lab using pip:

pip install polycraft-lab

This downloads the polycraft-lab package from pip, which contains tools to easily set up and manage the Polycraft game client.

To install the most recent changes (experimental version):

pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://test.pypi.org/simple/ polycraft-lab

Alternatively, to install the latest version from source:

git clone https://github.com/PolycraftWorld/polycraft-ai-lab
pip install ./polycraft-ai-lab

If you installed PAL using method (1a), this isn't required.

Run the following command and follow the guided instructions:

pal init

This will automatically download Polycraft World to your local machine and initialize up other configuration files.

Now train your agent like you would do with any other gym-style environment:

from polycraft_lab import make

env = make('pogo_stick')
observation = env.reset()
for _ in range(1000):
    env.render()
    action = env.action_space.sample()  # your agent here (this takes random actions)
    observation, reward, done, info = env.step(action)

    if done:
        observation = env.reset()
env.close()

Polycraft AI Lab also contains a wrapper [WIP] to start experiment creation from the command line. The following begins the experiment creation process by launching Minecraft:

python -m polycraft_lab.ect --create EXPERIMENT_NAME --launch

Running python -m polycraft_lab.ect --create EXPERIMENT_NAME will allow you to create simpler experiments, allowing configuration of more high-level domain attributes, like action space and a preconfigured goal, such as finding diamonds as quickly as possible.

Clone out the repository:

git clone https://github.com/PolycraftWorld/polycraft-ai-lab.git

Optionally, you can create a virtual environment to store dependencies.

In any case, install the dependencies:

pip install -r requirements.txt

Alternatively, a virtual environment can be created with the necessary dependencies by running:

cd polycraft-ai-lab
pipenv install

Polycraft AI Lab will be distributed using pip.

The easy way to upload to upload to the test PyPI index:

./release.sh

To release to the live PyPI index:

./release.sh --release