I’m pretty sure the approach isn’t that novel, but I really rediscovered it on my own while exploring several compressions approaches (generic compressions with tailored dict like zstd, integer packing/compressions, compressed bitmap, …: I probably have my notes about these somewhere)
As such, I don’t have any name/papers to give you nor point you to similar application. But I would also be interested ^^
But don’t hesitate to reach out if you work on something similar and wanna discuss about it.
Ha, yeah, I remember reading about your project back in April (I think someone shared it on the GeoRust Discord).
Really cool stuff you have here!
Can't say I understand all the math behind it, as it's not my forte (even for H3, for the more numerical parts, I rely on the work of the original authors: I could never have come up with this myself), but your doc is really great!
For the follow-up article, I hope I can get to it eventually.
But spare time is a rare currency ^^
For my use case, the visual distortion of S2 was quite a no-go.
As for DB queries, it really depends on your use case and how you store your data, but you can get some good results.
But yeah, if you really need exact parent-child containment, S2 is easier to work with.
amo is using it quite a lot, mainly for the scratch map feature in the Bump Map application, but not only.
Use cases are:
- data storage
- data aggregation/clustering
- spatial indexing
- geometrical computation (as long as you're OK with approximation, you can speed up a lot of things by working with CellID instead of actual geometries)
- data visualization
I've seen it used by Databend, Helium, Breakroom (they did an Erlang binding on top of h3o), beaconDB, Greptime, Meilisearch.
But I don't exactly know what they are using it for (just that they pulled h3o in their projects).
That was the blog post for the initial release, and a lot of things have changed since then (definitely deserves a new blog post ^^).
The first big change happened six months after the release, when I rewrote most of the geometrical algorithms (leveraging the excellent geo crate) and got a massive boost in speed and reduction in memory usage which made it applicable at high resolutions and country-scale levels (e.g. some computation went from 15h to 7min, and from 18GB of RAM to 100MB). I also added support for alternative coverage methods (back then H3 only offered centroid containment).
Since then, the reference implementation has caught up in term of coverage predicate and even provides a new experimental coverage algorithm addressing some performance issue. I haven’t implemented yet but, IIRC, my current implementation still outperforms theirs (but less dramatically so).
I’ve also developed a little ecosystem of libraries around h3o:
- Tailored compression algorithm with h3o-zip (in optimal cases I’ve observed reductions from ~2GB to 100KB)
- Compact data structure for fast lookup with h3o-ice (based on FST)
- Map rendering with h3o-mvt
Most of these things run in production at Amo, where one of the main use cases is powering the Scratchmap feature, both client and server side, in the Bump app.
I’ve also seen adoption from other projects (bindings for R, Erlang, Polars, ...) and enterprises :)
As such, I don’t have any name/papers to give you nor point you to similar application. But I would also be interested ^^
But don’t hesitate to reach out if you work on something similar and wanna discuss about it.