find_ab_params

cuml.manifold.umap.find_ab_params(spread=1.0, min_dist=0.1)

Fit a & b parameters for UMAP.

Selects a and b for the differentiable curve used in lower dimensional fuzzy simplicial complex construction. We want the smooth curve (from a pre-defined family with simple gradient) that best matches an offset exponential decay.

Borrowed from upstream umap: https://github.com/lmcinnes/umap.

Parameters:

spread: float (optional, default 1.0)

The effective scale of embedded points.

min_dist: float (optional, default 0.1)

The effective minimum distance between embedded points.

Returns:

a, bfloat

The a and b parameters for UMAP.