Umap: How does UMAP estimates/evaluates variance?

[cartal]

Hi,

Thanks for develop UMAP. Is such a superb tool.

My question is regarding how much variance can be explained by UMAP. I have been through he documentation, and is possible that this is explained somewhere in the preprint, but I may have missed it.

Can you provide with an explanation or a place where I can find this?

Thanks!

[lmcinnes]

As a non-linear manifold learning technique that ultimately works with
metric spaces at its heart rather than feature based data UMAP doesn't
really have a notion of explained variance the way algorithms like PCA do.
A colleague is working on developing some alternative measures of how well
an embedding has performed, but that is still preliminary work and we don't
have any published code for that yet -- ultimately UMAP remains a research
project with additional features and utilities still being developed.

On Mon, Aug 13, 2018 at 7:02 AM Carlos Talavera-López <
[email protected]> wrote:

Hi,

Thanks for develop UMAP. Is such a superb tool.

My question is regarding how much variance can be explained by UMAP. I
have been through he documentation, and is possible that this is explained
somewhere in the preprint, but I may have missed it.

Can you provide with an explanation or a place where I can find this?

Thanks!

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/lmcinnes/umap/issues/122, or mute the thread
https://github.com/notifications/unsubscribe-auth/ALaKBSRs72ucVLk9Ref5Nu_WHR4R0_sgks5uQVzRgaJpZM4V6TT7
.

[cartal]

Thank you so much!