Corey J. Nolet

While doing the migration from RAFT, I found there are still several header files in `cuvs/include` which define structs and/or classes and they include templates and implementations. We need to...

This can often ease development times, but it creates a situation where any collisions in functions or object names can create very challenging, if not impossible to debug situations. While...

[FEA] In-place k-select for sparse knn on string matching

7

One of the challenges with providing optimized primitives for sparse datasets is the large number of different degree distributions that can be encountered in practice. In some recent discussions with...

Native support for cuPy/cuDF backed Anndata

6

This would be immensely useful for the GPU data science community as it would start to enable pipelines fully on GPU.

RAFT has a `runtime_exception` that we try to catch and throw everywhere but it's more of a blanket for anything that can go wrong at runtime and it doesn't actually...

[FEA] Filtered CAGRA to automatically use pre-filtered brute-force when the filter ratio reaches a particular theshold.

3

While CAGRA offers the capability to specify a pre-filter, it is limited in that heavy filters can end up reducing recall significantly after a particular point. For this reason, we...

Currently, all of the Ann indexes inherit from a common base class, but the base class doesn't have any virtual methods or state attached to it, so any APIs coded...

Our serialization tools from RAFT are able to read and write numpy files and it's really trivial to write numpy `ndarray` tensors to files on disk. If we can automatically...

We've migrated over most of the important VSS/clustering APIs to cuVS but a few remain: - [ ] Random ball cover (Corey) - [ ] Sparse distances (Divye?) - [...