Description

Hi! I was wondering to know the difference between hidden property in the training configuration of dqn and fcnet_hiddens of model configuration.

Reviewing the discuss forum (reference) I found that the documentation is not complete. hidden is described as "Dense-layer setup for each the advantage branch and the value branch" but felt one line more: "in a dueling architecture." to be more precise an avoid confusions.

Best

Link

https://github.com/ray-project/ray/blob/master/rllib%2Falgorithms%2Fdqn%2Fdqn.py#L175