Should be very similar to the random forest aggregations but need to weight the contributions from each estimator appropriately.