alec-tschantz / rl-inference

Reinforcement Learning through Active Inference
67 stars 21 forks source link

Reinforcement Learning through Active Inference

License: MIT

Repository for ICLR 2020 submission to Bridging AI and Cognitive Science (BAICS)

[Paper / Presentation]

Requirements

The environment.yml contains the exact package versions, but recent versions of the following packages should be sufficient:

If you wish to use mujoco or dm_control environments you will also require:

Running

git clone https://github.com/alec-tschantz/rl-inference.git
cd rl-inference
python scripts/train.py --config_name="mountain_car" --logdir="mountain_car" --strategy="information" --seed=0

The provided configs are [mountain_car, cup_catch, half_cheetah_run, half_cheetah_flip ant_maze, reacher]

Acknowledgments

The code for the ensemble model was adapted from the max repo