Question on Supervised Dim Reduction

Nobody has claimed this yet.

I am working on a project where I have 100D navigation vectors of Wikipedia articles and I plan to cluster these vectors and then visualize the clusters. I wanted to use labels acquired from clustering in the process of reducing dimensions with UMAP.

But I have no idea if it it works or not. So my questions are:

1. Is there anyway to have any idea about the magnitude of the loss?
2. Is there anyway to see the definition of the loss function either in code or mathematically so I
can see how it utilizes the class labels?
3. If we cannot view the loss how would one approach parameter tuning? Right now I gauge the quality of my visualization by measuring trustworithness and checking it by eye.

Is the loss function listed [here](https://umap-learn.readthedocs.io/en/latest/how_umap_works.html)?

Kindly awaiting your answers! Thank you

Start with the linked “How UMAP Works” documentation and the supervised dimensional-reduction behavior described in the issue. A useful outcome would explain whether the loss is exposed, how labels affect it, and what guidance exists for parameter tuning and trustworthiness evaluation.

Written by the indexing model from the issue text.