Hacker News new | past | comments | ask | show | jobs | submit login

If you like DBSCAN, I would recommend checking out OPTICS. It is not a clustering algorithm per se, but generates information from which clusterings for multiple settings of DBSCAN parameters can be "queried" cheaply. The DBSCAN and OPTICS papers share an author. One my favorite algorithms. This is the original paper - [1], and this looks like a helpful presentation on the topic - [2]. Since you mention k-means, I would point out that unlike k-means which finds convex-shaped clusters only, DBSCAN/OPTICS identify clusters of any shape.

[1] http://www.dbs.ifi.lmu.de/Publikationen/Papers/OPTICS.pdf

[2] https://www.cse.buffalo.edu/faculty/azhang/cse601/density-ba...

That definitely looks very interesting, I'll look into that.

Clustering on non linearly separable clusters is also one of the reasons we chose DBSCAN back then :)

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact