UMAP RGB / multi-channel images

[JesseRMeyer]

Hi,

The introductory material nicely showcases how UMAP will cluster 64x64x1 hand written digit images 'out of the box'. In my case, my images are not 1 channel, but we can consider the 2 channel case. I thought I could just pass into umap.fit(images_arr), but doing so raises the exception "Found array with dim 4. None expected <= 2". images_arr being a numpy array of (number_of_images, Y, X, number_of_channels). Referring to the API, the input parameter X is to be described by a shape of (n_samples, n_features) or (n_samples, n_samples). The hand written digit case I believe to be the latter.

Is the intention to use UMAP independently on each channel, and then somehow fit() these together afterward? I leafed through the Discussions and the Tutorial / User Guide for suggestions, and maybe I missed something, but I would appreciate guidance.

Thanks!

Jesse