Basic-Policy-Gradient-Labs

A repo to study basic Policy Gradient algorithms (like REINFORCE) on classic control gym environments

Accompanying videos

A policy gradient class where the algorithms are explained:

https://www.youtube.com/watch?v=_RQYWSvMyyc

A video about phenomena studied with code from this repo:

https://www.youtube.com/watch?v=gLVodUwzHyU

A further video about the code itself: https://www.youtube.com/watch?v=ib8q9ReedbM

python version

Use python 3.

Installation

Main installs:

pip3 install -r requirements.txt

Install gym

Actually, the main install above does it, but if you want to do everything manually...

pip3 install gym

More information here:

https://gym.openai.com/docs/#installation

Install Continuous Cartpole Environment

pip3 install -e my_gym

And that should be it!

Example of command

python3 main_pg --env_name Pendulum-v0 --nb_repet 1 --nb_cycles 500 --max_episode_steps 200 --policy_type squashedGaussian

The list of possible arguments is found in arguments.py, together with the default values