Applying transform() to mixed datasets.

[jackcturner]

Hello,

I've seen a few issues asking this same question, but they are unanswered and from some years ago, so I'm hoping I can get some advice.

I have two UMAP models generated using different data types of the same sample. I've taken the intersection of these two models to produce a single embedding which I'd now like to transform new data into. It's not clear to me how this can be done, as the intersection operation can only be applied to UMAP objects, not the arrays returned by the transform method.

Is there some way of doing this within the UMAP functionality, or are there alternative approaches?

Thanks,
Jack

[lmcinnes]

I'm afraid that's not supported. Once you have combined the models the transform is not really going to work. In principle it might be possible to do something, but it would involve calling code internals yourself and potentially a bit of work to get working. How badly do you need to be able to do this?

[jackcturner]

Thanks very much for the quick reply.

I'm trying to apply UMAP in a context I don't think its been used in before and want to ensure I'm making the most of its functionality. While being able to transform multiple types of data at once would be really useful, I'm sure I can get something interesting out regardless.

Besides, I'm not confident I understand enough about whats going on under the hood to figure this out for myself.

Thanks again.