Is there an example of action masking for continuous actions?
I have the following.
obs_space = spaces.Box(low=-1000,high=1000, shape=(30x50,))
act_space = spaces.Box(low=-0.05,high=0.05, shape=(30x3,))
so the 30 X 50 matrix actually is 20 X 50 and I pad the rest 10 X 50 with zeros. After that, I flatten it to get (1500,) dimension observations.
I want to add an action mask such that the model produces actions for 20 X 3 and the rest actions i.e. 10 X 3 are zero.
I am using PPO and ray2.0.0
How do I do this?
Thnx for your help in advance.