[isabellat]

Since its just a dot product between the learned weights and the feature vector, we do this in the application layer as lacksconfidence surmised.