Suppose, I have some points and I've run a clustering algorithm on them.

[michael@westergaard.social]

Suppose, I have some points and I've run a clustering algorithm on them. Now, I'd like to draw the clusters.

The points are in n-dimensional space, but I've run a PCoA (similar to a PCA) on them and reduced it down to two dimensions.

I'm thinking of drawing either a convex hull of the clustered points or drawing the shape ordering points by angle around cluster centroids. Is there a more canonical/better looking way I'm overlooking?

[drgroftehauge@sigmoid.social]

@michael t-SNE/UMAP

Just remember that nothing is real and hyperdimensional space is a bitch

[drgroftehauge@sigmoid.social]

@michael But t-SNE/UMAP is instead of PCA so not what you were asking, sorry.

[michael@westergaard.social]

I'm already doing a handful of different analyses, and UMAP looks like it might be interesting to add to the mix…

This is a PCoA (not PCA as I have rankings with distances) using convex hull visualization of DBSCAN clusters (DBSCAN run in full n-dimensional space, projected onto the two most significant dimensions).