Voted imperative because computers are things we tell what to do.
But if we are talking about pure functions (no side effect, no change to state and so on) I would let the verb "compute" as implicit and use a name describing what have to be computed instead.
This is my usual naming approach but there are more corner cases that depend on context.
For example, when programming a domain model, specification methods are usually called `IsSatisfiedBy` since the class/interface express the conditions represented.
Similarly in DDD, I name most analysis method as `Analyse` as the interface/class name express the analysis that will be computed.