Compute_actions() for VectorEnv

How severe does this issue affect your experience of using Ray?

  • Medium: It contributes to significant difficulty to complete my task, but I can work around it.


I am trying to evaluate my policy trained on my custom VectorEnv.

I would like to provide the obs list coming from the vector_step() to the trainer and get back the actions for all environments. Is there a way to do this? I tried looking at compute_actions() but I do not know if it is what I need.

Thank you

Yes, compute_actions or compute_single_action is a way to go here.
But it depends on the algorithm you are using. If you use recurrent networks or attention, your view-requirements will mean that you need an input dict and you should use compute_actions_from_input_dict.