akshayka
commented
2 years ago Hi, thanks for the suggestion! We haven't looked into speeding up the nearest neighbor computation.
For speeding up nndescent in particular, I would suggest moving this discussion to the pynndescent Github repository.
Hi, I've seen this work by Intel: https://github.com/intel/scikit-learn-intelex and curious if anyone has looked into how it may compare / if it can be used as a speed up for nndescent? There is an example notebook here showing the speedup on a toy example: https://www.kaggle.com/kppetrov/fast-knn-using-scikit-learn-intelex-for-mnist - i believe there are also additional hooks for GPU methods: https://intel.github.io/scikit-learn-intelex/gpu.html