https://github.com/stefandeveloper/bomberman

Reinforcement Learning for Bomberman at University Heidelberg
https://github.com/stefandeveloper/bomberman

bomberman bomberman-ai bomberman-game heidelberg heidelberg-university reinforcement-learning

Last synced: about 1 month ago
JSON representation

Reinforcement Learning for Bomberman at University Heidelberg

• Host: GitHub
• URL: https://github.com/stefandeveloper/bomberman
• Owner: stefanDeveloper
• License: mit
• Created: 2021-02-20T19:22:19.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-03-28T17:02:58.000Z (over 3 years ago)
• Last Synced: 2023-03-25T19:50:44.367Z (over 1 year ago)
• Topics: bomberman, bomberman-ai, bomberman-game, heidelberg, heidelberg-university, reinforcement-learning
• Language: Python
• Homepage:
• Size: 122 MB
• Stars: 2
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Brooklyn Bang Bang

> Setup for a project/competition amongst students to train a winning Reinforcement Learning agent for the classic game Bomberman.

## Approaches

* [Terry Jeffords](agent_code/terry_jeffords/README.md) simple neural network approach

* [Amy Santiago](agent_code/amy_santiago/README.md) is an agent, based on [Pytorch Example](https://pytorch.org/tutorials/intermediate/reinforcement_q_learning.html)

* [Rosa Diaz](agent_code/rosa_diaz/README.md) agent with pre-selection of valid/invalid and preferred/discouraged actions

* [Jake Peralta](agent_code/jake_peralta/README.md)

* [Nikolaj Boyle](agent_code/nikolaj_boyle/README.md)

* [Scully](agent_code/scully/README.md) agent with full sight but convolution layers in the network

* [Blindfisch](agent_code/blindfisch/README.md) agent with restricted sight and convolution of input features

* [Hitchcock](agent_code/hitchcock/README.md) imitates the rule based agent

## Github Actions

> Training agents can be annoying. To ease our work, we have integrated a Github Action to train our agents automatically. Sadly, my server does not have a graphic card, so we still use the CPU to train. Well, I ain't rich :D 

Run [Github Runner](https://docs.github.com/en/actions/hosting-your-own-runners/about-self-hosted-runners) using a [Docker Runner](https://hub.docker.com/r/tcardonne/github-runner/) image:

```shell

docker run -it --name github-runner \

    -e RUNNER_NAME=private \

    -e GITHUB_ACCESS_TOKEN=... \

    -e RUNNER_TOKEN=... \

    -e RUNNER_REPOSITORY_URL=https://github.com/stefanDeveloper/bomberman \

    -v /var/run/docker.sock:/var/run/docker.sock \

    tcardonne/github-runner:ubuntu-20.04

```

Troubleshooting occurred when we ran the latest image. However, in [this](https://github.com/tcardonne/docker-github-runner/issues/22) pull request, they added the ubuntu support. Using this tag, we can set up our python environment. 

```shell

Error: Version 3.9 with arch x64 not found

```

## Tips & Tricks

Setup environment:

```python

pip install -r requirements.txt

```

Train agent:

```python

python -m main play --agents scully --train 1 --n-round 2500 --no-gui

```

Play agent:

```python

python -m main play --agents scully

```