Nick Becker

[FEA] Reduce peak memory requirements of HDBSCAN fit with large min_samples

1

cuML requires significantly more memory to fit HDBSCAN as `min_cluster_size` increases than the CPU version does. This can cause users to unexpectedly run into memory issues when switching from CPU...

[FEA] Support for HDBSCAN relative_validity

8

The CPU hdbscan [library provides](https://github.com/scikit-learn-contrib/hdbscan/blob/c5fcf4b3829d391eadd14598736a763952790a82/hdbscan/hdbscan_.py#L1457) a `relative_validity_` attribute, which is a "[fast approximation of the Density Based Cluster Validity (DBCV) score](https://hdbscan.readthedocs.io/en/latest/api.html#id92)". It would be nice to support this in cuML,...

[FEA] Convert Dask into an optional dependency

1

Dask is the primary runtime through which cuML provides multi-GPU functionality in Python. As a result, it's historically been a required cuML dependency. This is a non-issue for users who...

[QST] Should NVTX annotations of estimators always be available (rather than opt-in)

1

Currently, it looks like most NVTX annotations for estimators are opt-in, based on the `NVTX_BENCHMARK` environment variable. https://github.com/rapidsai/cuml/blob/697a37c68a1cffe08c5ec12fac5961cf1d7aed7c/python/cuml/internals/base.pyx#L244-L246 Should we instead always annotate? Or, do we expect this would add...

We frequently investigate the memory behavior of workflows. Like https://github.com/rapidsai/cudf/pull/15866 , we should implement configurable RMM-bsaed memory profiling tied to NVTX annotations.

[DOC] Document which algorithms expect Fortran vs. C contiguous data

1

For many algorithms, whether the input data is C or Fortran contiguous determines whether an expensive memory copy needs to be made. While this seems innocuous, it can have significant...

As noted in #5929, for many algorithms, whether the input data is C or Fortran contiguous determines whether an expensive memory copy needs to be made. While this seems innocuous,...

[FEA]: nx-cugraph should intercept built-in constructors like `from_pandas_edgelist` if `NETWORKX_BACKEND_PRIORITY=cugraph`

5

### Is this a new feature, an improvement, or a change to existing functionality? New Feature ### How would you describe the priority of this feature request High ### Please...

[QST]: Should nx-cugraph GPU graph caching be opt-out rather than opt-in?

2

### What is your question? Converting a CPU NetworkX graph to a GPU graph can be slow, so nx-cuGraph provides a configuration option that caches the converted graph so users...

[BUG]: Nx-cuGraph caching not working with documentation example

2

### Version 24.06 ### Which installation method(s) does this occur on? Conda ### Describe the bug. When I ran the documentation example multiple times, I expected performance to be significantly...