I don't really know what to call this project as of now. I want to present cute graphics in a kawaii webpage and all that. The goal is to compare at least two algorithms:

• Distributed Deep Q-Network
• Emergent Tangled Program Graphs ^1
• Neuroevolution

and common to all:

• use a world model.

In a first instance, these will be tested against a single environment, ViZDoom. Some time later I'd love to do put these to the test and multi-task with OpenAI's Universe.

All shall be published in a webpage hosted by a branch of this repo.

Dockerfiles are in the /docker folder and follow mainly what's written in the vizdoom docker directory.

So after building it, to run it, run

git clone https://gitlab.com/bgalvao/rl.git
cd rl/docker/vizdoom/
sudo docker build -t rl .
cd ../..
sudo docker run  -ti --net=host -e DISPLAY=${DISPLAY} --privileged --rm -v $(pwd):/home/vizdoom/dev rl

Inside the docker image, remember to invoke python3 cxxx.py.

^1 Stephen Kelly and Malcolm I. Heywood. 2017. Multi-task learning in Atari video games with emergent tangled program graphs. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO '17). ACM, New York, NY, USA, 195-202. DOI: https://doi.org/10.1145/3071178.3071303