COMP34120 AI and Games Coursework

We will be running each within this specific Dockerfile image and with this specific Docker environment. So make sure your agent will work within this container environment and under the constrain.

Inside the Dockerfile, two versions of PyTorch and Tensorflow are available. Select the one you need based on the availability and use of the GPU.

To build Dockerfile.

docker build --build-arg UID=$UID -t hex .

The building process will take a while.

To run the container use:

docker run --cpus=8 --memory=8G -v "$(pwd)":/home/hex --name hex --rm -it hex /bin/bash

If you need GPU access, pass --runtime=nvidia to the docker run command.

The current repo will be mapped to /home/hex within the container. If you cd hex you should be able to see all your local file. Any changes made to that directory will reflect to your system directory. This will be the command we use to create the running environment for playing each game, so your agent can at most you 8 CPUs and uses 8 GB of memory.

To run a game of Hex, you can use:

python3 Hex.py

By default, two agents/DefaultAgents/NaiveAgent.py agent will play against each. To see all the available options and help message use python3 Hex.py --help.

To exit the docker container you can simply do exit. This will stop the container.

To enter the container again you can simply use:

docker start -i hex

To run the test suite, you can use:

python3 -m unittest discover

To use a GPU, use CUDA 12.3.0 with the preinstalled TensorFlow 2.19.0 and PyTorch 2.5.1+cu121, or ensure any other versions you install are compatible with them.

Misc

PDF doc link: typst