Hi,
I'm the maintainer for the umap-learn package, and this looks like a very powerful and useful approach to providing some feature attribution to (parametric) UMAP. I was hoping you might be willing to look at contributing a PR to the umap-learn package (https://github.com/lmcinnes/umap) to enable this for the parametric UMAP implementation there. I know you are working from the pytorch parametric UMAP in this case, but hopefully the keras based version in umap-learn might still be something you could adapt to.
Hi,
I'm the maintainer for the umap-learn package, and this looks like a very powerful and useful approach to providing some feature attribution to (parametric) UMAP. I was hoping you might be willing to look at contributing a PR to the umap-learn package (https://github.com/lmcinnes/umap) to enable this for the parametric UMAP implementation there. I know you are working from the pytorch parametric UMAP in this case, but hopefully the keras based version in umap-learn might still be something you could adapt to.