Hi Tomas,

I think I must have changed the hyper-parameters before publishing the post, and didn’t re-run all the experiments again to ensure the same results for all the exploration techniques. I am going to go back and attempt to find the original hyper-parameters I used for those results. In my experience CartPole is somewhat difficult to solve with DQN, and there is a narrow space within which good hyperparameters exist.

Generally using a smaller learning rate and larger number of annealing steps will yield more robust improvements though.

PhD. Interests include Deep (Reinforcement) Learning, Computational Neuroscience, and Phenomenology.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store