Overview

The purpose of this repository is to provide meaninful baselines for a variety reinforcement learning approaches. The RL approaches are evaluated with an emphasis on the average performance accross multiple agents.

Q-learning algorithms

Recipies

OpenAI cartpole v1
Description: See https://gym.openai.com/envs/CartPole-v1/ for more information.
Agent description	Representative parameters	Mean performance accross thirty agents
SGD with feedforward ANN
Colorado State Univ cartpole swing-up and balance task
An inverted pendulum on a cart initially developed by Chuck Anderson (chuck.anderson@colostate.edu). An evaluation episode begins with the pole pointing down, the cart in the center of the track, with both the cart and pole with zero velocity.
Agent description	Representative parameters	Mean performance accross thirty agents
SGD with feedforward ANN	Adam with feedforward ANN

Carpole

Other stuff:

How to create package

python3 setup.py sdist bdist_wheel

python3 -m twine upload dist/*

To create a private release

$ python3 setup.py sdist bdist_wheel