As expected, the improvements are more significant for smaller datasets, and tend to increase with the number of leaves (e.g. the top row of Fig 4A and Fig 4B). For larger datasets, we start to see substantial improvements using HS at even larger numbers of leaves (Appendix S4.4).