So far the action `x` is assumed to be a 1dim continuous variable. Implementing `s_wise_max` for discrete action will be straightforward. The issue will be how to design the `ContinuousDP` type.