Store dimensionality reduction for feature visualization

[ziw-liu]

The speed of analysis will improve if we don't have to do the dimensionality reduction for each FOV.
czbiohub-sf/napari-iohub#13 (comment)

[ziw-liu]

@Soorya19Pradeep How should we do this? A script that stores the dataset-wide UMAP components in the prediction zarr store? Or should we just hardcode it in the prediction writer?

[Soorya19Pradeep]

@ziw-liu, I think it is better to do it through script as the UMAP components will change with run as well as number of components.

[ziw-liu]

UMAP components will change with run as well as number of components

The idea would be to store one UMAP per prediction run so it shouldn't change unless we need to tweak clustering parameters.

This is for the napari plugin so I'm not sure if we will ever use 3D UMAP.

[Soorya19Pradeep]

Ok. It makes sense for the Napari plugin as it makes it faster. Storing the UMAP components makes sense for this.