Closed schrum2 closed 5 years ago
Got this to work using the file DQN_cartpole.py from the Tidy RL repo. This uses tensorflow and does NOT use PyTorch. Hopefully we'll be able to run it on the campus machines. Since this is a working Deep RL agent, I'm closing this issue, but will soon make many more.
DQN is supposed to be a fairly basic Deep RL agent, so we should be able to get it working and running using PyTorch, and applied to Open AI Gym domains. The code from that Grokking RL Repo should work for this purpose, but the code also exists in other places (not always in PyTorch).
OpenAI actually has a repo that you might consider copying from: https://github.com/openai/baselines
Here is a PyTorch implementation of DQN: https://github.com/AndersonJo/dqn-pytorch
Here is another tensorflow option https://github.com/sarcturus00/Tidy-Reinforcement-learning
The idea is to copy as little code as possible to get ANYTHING working. Learn how to master Cartpole, and then also master another domain, like MountainCar. Once we are consistently learning in these simply domains, we can close this issue and move on to others.