Reinforcement Learning Intro

These educational materials originate from the OpenAI 2018 Hackathon.

This repo contains slides and code for a 30-minute "drink from the firehose" intro to RL, given at the OpenAI Hackthon on March 3rd, 2018. The slides contain a mathemiatical and programming introduction to reinforcement learning from the context of deep neural networks.

The file pg_cartpole.py contains an 80-line, bare-bones implementation of policy gradient which seems to work in CartPole.

The file dqn_cartpole.py contains a short but somewhat messy implementation of DQN, which was developed during a Deep Q-Learning workshope at the Hackathon. Cheers to the participants of the workshop, who successfully debugged it!

Run the Samples

The samples require the python packages gym, tensorflow and numpy.

pip install tensorflow numpy gym
python pg_cartpole.py
python dqn_cartpole.py

Gym may require additional dependencies for a full installation. See here: https://github.com/openai/gym#installation