Rllib extremely complex action space Possible?

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

  • High: It blocks me to complete my task.


how would I have an action space that represents two different SETS of actions. One, a set of 3 possible discrete actions (0, 1 and 2). And two, an action represented by a continuous value. And then the possible actions could be any combination of the discrete actions and the continuous value, for example,

Action 0: (1, 12.3)
Action 1: (2, 3.0)
Action 2: (0, 33.5)
Action n: (0, 999.0)

Please advise.

Thank you.

@hridayns , I think You can interpret action outputs as two continous values (similar like in robot environments) but the first of them can be rounded to 3 possible values.