Open dkobak opened 3 months ago
Yes, we definitely need to add that! IMO, a general **kwargs would be messy, and something like like a knn_kwargs
argument would be better. Scikit-learn also does something similar in their KNN classes, but they call it "params". I prefer "kwargs". What do you think?
Yes, sounds good!
See title. We discussed it at Dagstuhl. Should be useful for PyNNdescent, but could potentially be useful for Annoy/HNSW/etc at well.
We could potentially allow Affinity classes to take **kwargs as well and pass them to the KNNIndex. Here I'm less sure if that's a good design idea.