Automatically determine num_actions and num_chance_outcomes in stochastic_muzero_policy

[carlosgmartin]

It is possible to automatically determine the num_actions and num_chance_outcomes parameters to stochastic_muzero_policy from the root, decision_recurrent_fn, and chance_recurrent_fn parameters, via ChanceRecurrentFnOutput.action_logits.shape and DecisionRecurrentFnOutput.chance_logits.shape. That would reduce the number of parameters that users need to pass in by two.

I suggest allowing num_actions and num_chance_outcomes to be None, making them None by default, and determining their values automatically if they're None. This preserves backward compatibility. Thoughts?

[fidlej]

Thanks for the suggestions. It sounds good. We do not need to preserve the backward compatibility with the None defaults. The stochastic_muzero_policy is still experimental.

Will you send a pull request or should I do that?

[fidlej]

Thanks for the improvements.