Anupam

Follow up to #6486 Scope: This should essentially be spectral embedding + kmeans Will need to do the cpp layer in cuvs and expose in cuml Python API to mirror...

Parent tracker issue for the entire spectral embedding feature

Tracking issue for python API for spectral embedding API should be similar to `sklearn.manifold.SpectralEmbedding`

Instantiate only specific RAFT linewise kernels

1

Depends on https://github.com/rapidsai/raft/pull/2701 and Resolves https://github.com/rapidsai/raft/issues/2682

Enable dice metric for UMAP and pairwise_distances

6

Resolves #5129 Depends on https://github.com/rapidsai/cuvs/pull/971 Adding new distance types in cuvs bloats the binary size.

Follow up to #6743 We have seen the spectral initialization fail sometimes and need to find out where the problem is coming from. # Todo * Extract solver input data...

Moving dice distance from raft to cuvs and support for half types

8

Resolves #966

Once we are able to support `int64_t` in Spectral Embedding https://github.com/rapidsai/cuml/issues/7225 we can support `int64_t` in Spectral Clustering.

We should remove the contents under `include/embed` and `src/embed` since we have introduced a new Spectral Clustering under `include/cluster`

UMAP fix int32 overflow causing illegal mem access

3

Resolves #7517 The `optimize_batch_kernel` and `optimize_batch_kernel_reg` use the following condition to check for out of bounds `while (row < nnz)` Inside the loop `row += skip_size` is used for iteration....