Comment by EdSchouten
Comment by EdSchouten 4 days ago
Exactly. But my concern is that this is not any stronger/better than what Prolly trees already offer, which is why I'm disappointed that they are mentioned under "related work", but not discussed/compared in more detail.
You're right, we should delve into a comparison more with respect to prolly trees. We actually have a lot of experience with prolly trees, and have found, in practice, that you need to do a lot of the things that folks like dolt have had to do to make them work nicely. Whereas with G-trees, the basic implementation turns out to be quite nice (and extremely easy to reason about).
One of the biggest benefits of G-trees in my mind, is their ease of implementation. Additionally, we did a lot of work to explore their statistical properties, which doesn't exist for prolly trees (though in hindsight, we have done this, so should probably write it up formally).