You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There have been certain spectral partitioning bugs that have popped up in CUDA 11.4 nightlies. While they are being investigated at the moment in #2600, we are approaching burndown quickly and need certain PRs to go in to unblock downstream work.
Follow-up PR to simplify (and make more robust) the parameter handling in cuml UMAP cpp. This would be removing duplicate parameters, and making graph_degree set to n_neighbors.
Removing the now unnecessary slicing
Some customer-facing testing for performance and memory usage.
I4. mproving cuml UMAP's default behavior to better dispatch build algo based on data size and input type. This isn't really blocked on this upstream work, but would be nice to test in concert with it.
The text was updated successfully, but these errors were encountered:
There have been certain spectral partitioning bugs that have popped up in CUDA 11.4 nightlies. While they are being investigated at the moment in #2600, we are approaching burndown quickly and need certain PRs to go in to unblock downstream work.
Specifically in cuML (mostly blocked on #2586):
I4. mproving cuml UMAP's default behavior to better dispatch build algo based on data size and input type. This isn't really blocked on this upstream work, but would be nice to test in concert with it.
The text was updated successfully, but these errors were encountered: