Enables the removal of training data from fitted R models while retaining predict functionality. The purged models are more portable as their memory footprints do not scale with the training sample size.