[QST] UMAP Inference (without Python)

[ryanccarelli]

From the documentation, I could only find support for pickling the UMAP model. I would like to export to a format that supports (GPU) inference without Python. Something like Onnx or the Treelite solution recently added for random forests.

Does something like this exist for UMAP? Any suggestions would be welcome, thanks

[viclafargue]

Unfortunately this feature is not yet available, but that is something we are thinking about. However, if you are interested, there's already a POC PR that allows one to train a LinearRegression model with Python and serialize the model as a protobuf file that is then recoverable from C++. The design is far from definitive and we might very well not use Protobuf in the end.

[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.