[FEA] nan_euclidean or custom metric added as a metric in NearestNeighbors

[PolarBean]

sklearn has recently added support for nan_euclidean in nearest neighbors. This is described here: https://github.com/scikit-learn/scikit-learn/issues/25319

I see that cuml supports nan_euclidean in some places: ie; https://github.com/jakirkham/cuml/commit/d81465de97d3e62b0b20be3d25d42fd40ae6b215

however as of yet it does not seem to be available in NearestNeighbors. Is this possible to add, alternatively is it possible to add custom distance metrics defined by the user as is allowed in sklearn.

Thank you for your fantastic library

[dantegd]

Thanks for the request @PolarBean, it definitely is in our roadmap (both things actually), but we don't have a precise ETA currently. Will update this issue with more info as we progress things.

[PolarBean]

Great to hear. Thank you for your response. Happy New Year!

[akurniawan]

Hi @dantegd do you have any updates on this features? I'm currently interested to use DTW distance as the metric and hoping it could be implemented natively or as a custom metrics