Is there a way to implement an OpenAI's environment, where the action space changes at each step?

Question

Is there a way to implement an OpenAI's environment, where the action space changes at each step?

Answer 1

Yes (though some of the premade agents may not work in this case).

@property
def action_space(self):
    # Do some code here to calculate the available actions
    return Something

The @property decorator is so that you can fit the standard format for a gym environment, where the action_space is a property env.action_space rather than a method env.action_space().