fgxaos / rl-atari-experiments

Project for CentraleSupelec's Reinforcement Learning course

RL Atari experiments

Deep Reinforcement Learning project with Atari games for the Reinforcement Learning course at CentraleSupelec

Description

For this project, we implemented two different models: DQN and MNF-DQN.

Thanks to OpenAI's gym environment, any Atari environment can be used to train one of the two aforementioned models.

How to run

• Install the required libraries

pip install -r requirements.txt

• Set the configuration file cfg.yml to run the desired experiment

• Install ROMs

In order to import ROMS, you need to download Roms.rar from the Atari 2600 VCS ROM Collection and extract the .rar file. Once you've done that, run:

python -m atari_py.import_roms <path to folder>

This should print out the names of ROMs as it imports them. The ROMs will be copied to your atari_py installation directory.

• Run the experiment

python main.py

Note: in our experiments we used only three Atari environments (Freeway-v0, Skiing-v0, MsPacman-v0); but it is possible to run an experiment with any other Atari game (as long as it is available in the gym environment).

References

In this project, we used:

• some functions from OpenAI's baselines;
• code from PyTorch's tutorial on the DQN model;
• code from Facebook Research's MNF-DQN model.