A gentle introduction to HDBSCAN and density-based clustering

towardsdatascience.com

55 points·polm23··9 comments

Good article.

I used HDBSCAN in my master's thesis. It works well with high dimensional data. If you're using it for high dimensional stuff I would recommend working with Uniform Manifold Approximation and Projection (UMAP) to visualise. I think it is made by the same author as HDBSCAN.

I wish they also talked about Density-Based Clustering Validation (DBCV) which can be used to calculate the mathematical stability of the clusters (for hyper parameters), apart from just looking at hierarchies.

This is sorta true, but not quite. Leland McInnes and John Healy (the creators of UMAP), do in-fact have an amazing paper on HDBSCAN, but it's not inventing it. In their paper, https://arxiv.org/pdf/1705.07321.pdf, they introduce AHDBSCAN which is a great extension of HDBSCAN to dramatically improve it's performance.

Their work is great but just wanted to save people a google in case they were interested.

Never heard of AHDBSCAN, thank you for sharing!

Is there a difference between HDBSCAN and DBSCAN (commonly used for point cloud processing). The idea seems the same but I guess I never heard it called HDBSCAN.

HDBSCAN == Hierarchical DBSCAN, an extension of DBSCAN that produces hierarchical clusters.

This is covered in the article.

any article on clustering should mention ANN benchmarks and state of the art http://ann-benchmarks.com/

Uhh, why? They're not really related to each other...

ANN is the building block of density estimation. If I recall correctly that's the bottleneck for density-based clustering as its various algorithms then take advantage of faster yet algorithms (sort, union-find, spanning-trees..). While very interesting I am not sure this gentle introduction is the best place to discuss ANN.

Reply on news.ycombinator.com