valuepairs

[zsunberg]

Currently, we have actionvalues to get a vector of values, one for each action in the POMDP. Making sure these values always match up to the right actions seems rather error prone; it might be more helpful to have a valuepairs function that returns a list of action-value pairs so no one can get confused about which action each value corresponds to.

[rejuvyesh]

The vector representation is likely more efficient for DeepRL algorithms.

[MaximeBouton]

So far we managed the order internally by either having an actionmap field in the policy object or by making sure we use ordered_actions. I agree that it could be confusing, especially if actionindex does not agree with the ordering from whatever actions returns.
valuepairs would raise any ambiguity, not sure about the name.

[zsunberg]

The vector representation is likely more efficient for DeepRL algorithms.

Yes, though only very slightly if things are type-stable. I think we should have both.