[ENH] Support for scattered precomputed distance matrix for MNMG DBSCAN

rapidsai / cuml

cuML - RAPIDS Machine Learning Library

https://docs.rapids.ai/api/cuml/stable/

Apache License 2.0

4.2k stars 525 forks source link

[ENH] Support for scattered precomputed distance matrix for MNMG DBSCAN #3615

Open Nyrio opened 3 years ago

Nyrio commented 3 years ago

Support for precomputed distance matrices in DBSCAN was added in #3585 The precomputed distance matrix is quadratic in size, which means two things:

If you can fit it in memory, your dataset is probably too small to take advantage of MNMG.
Sending the complete matrix to all the nodes is long and therefore MNMG is currently extremely slow.

Is there a use case for supporting a distance matrix that is scattered across the nodes (solves both points above)? It will require changing the API a bit, as well as the kernels.

github-actions[bot] commented 3 years ago

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

github-actions[bot] commented 2 years ago

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.