louism33 / rl-starter-files

RL starter files in order to immediatly train, visualize and evaluate an agent without writing any line of code

With thanks to Lucas Willems (lcswillems), original author of https://github.com/lcswillems/rl-starter-files

This is designed to be used with https://github.com/louism33/gym-minigrid to train agents on the Scalable Supervision gridworld. This was for my thesis.

Easiest is to run: ./trainPPOLoop.sh of ./trainDQNLoop.sh Make sure to change the underlying gridworld hyperparameters in https://github.com/louism33/gym-minigrid/blob/master/gym_minigrid/envs/scalableoversight.py.

You will need to use this in conjunction with https://github.com/louism33/torch-ac and https://github.com/louism33/gym-minigrid