This MR adds caching to the hyperpe model. The probability is cached for each of the models individually.