How did you pick hyperparameter for the final run?

google-deepmind / dqn_zoo

DQN Zoo is a collection of reference implementations of reinforcement learning agents developed at DeepMind based on the Deep Q-Network (DQN) agent.

Apache License 2.0

451 stars 78 forks source link

How did you pick hyperparameter for the final run? #20

Closed hmishfaq closed 1 year ago

hmishfaq commented 1 year ago

Do you choose the hyperparameters based of the different seeds from the ones you use for evaluation? I just want to understand how the hyperparameters were chosen and after choosing the hyperparameter if the algorithms were rerun.

GeorgOstrovski commented 1 year ago

dqn_zoo attempts to provide exact reproductions of previously published DQN-based algorithms - all hyperparameters in this repository were chosen to match those in the respective publications, and we did not perform any of our own tuning.

If your question is about the original publications - the answer likely varies, and you'd have to go back to those papers to get details on how hyperparameter tuning was performed in each case.