Using (Double/Dueling) Deep-Q Networks to play Super Mario Bros.

Use virtualenv to contain the Python environment to a single local installation of python3:

To setup the virtual environment:

virtualenv -p python3 .env
source .env/bin/activate

When you've concluded the session:

deactivate

requirements.txt lists the Python dependencies for the project with frozen versions. To install dependencies:

python -m pip install -r requirements.txt

NOTE if you're NOT using virtualenv, ensure that python aliases python3; python2 is not supported.

The following instructions assume you have a shell running at the top level directory of the project. For comprehensive documentation on command line options, run the following:

python . -h

To execute the unittest suite for the project run:

python -m unittest discover .

To play games with an agent that makes random decisions:

python . -m random -e <environment ID>

• <environment ID> is the ID of the environment to play randomly.

For instance, to play a random agent on Pong:

python . -m random -e Pong-v0

To train a Deep-Q agent to play a game:

python . -m train -e <environment ID>

• <environment ID> is the ID of the environment to train on.

For instance, to train a Deep-Q agent on Pong:

python . -m train -e Pong-v0

To run a trained Deep-Q agent on validation games:

python . -m play -o <results directory>

• <results directory> is a directory containing a weights.h5 file from a training session

For instance, to play a Deep-Q agent on Pong:

python . -m play -e results/Pong-v0/DeepQAgent/2018-06-07_09-24