Can cuml APIs be called inside numba kernals ?

[ammaryasirnaich]

Can we use cuml methods with numba kernals ?

[viclafargue]

You could very well call the cuML/RAFT APIs before or after the use of NUMBA kernels as long as you take care of streams synchronization. However, a Numba thread won't be able to call the cuML/RAFT APIs.

[dantegd]

@ammaryasirnaich what is your usecase? As @viclafargue numba and RAPIDS are compatible (we even have used numba in many parts of the stack), mainly through the use of cuda_array_interface, so operations are efficient without memory copies.

[ammaryasirnaich]

@dantegd actually i want to run the cuml.dask.cluster.DBSCAN() parallel on feature set (14000,10) row-wise. I don't think the function runs on a specific dimension if given. I had some experiments to run histograms in the same fashion, for that I had to write a cuda kernal using numba.

[ammaryasirnaich]

You could very well call the cuML/RAFT APIs before or after the use of NUMBA kernels as long as you take care of streams synchronization. However, a Numba thread won't be able to call the cuML/RAFT APIs.

Many thanks for the response

[github-actions[bot]]

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.