Adjust UMAP hyperparameters given user data

[duhaime]

when running UMAP with very small or very large datasets, the default hyperparameters produce results that are too far apart or too tightly clustered. We can help fix this problem by setting UMAP's hyperparameters using insights from a user's dataset.

[pleonard212]

From testing it appears the spread variable could be key to visualizing small (1,000) datasets. Defaulting to 1, I've found that 0.4 works well with a dataset of 1,600 images. And conversely going above 1 might help with much larger datasets. It's described as "The effective scale of embedded points"; in pixplot it sort of has the effect of adjusting white space.