DBSCAN multi threaded (?)

Hi All,

I’m trying to do outlier detection via DBSCAN, but i notice that it is very slow, and not optimised for multi-threading, some googling around, i’m not sure if it is really easy to make it multi-threaded…

are you guys having it in the roadmap somewhere, or is there an interesting workaround in doing this?

the nice thing that i like about DBSCAN is that it kind-of comes up with a most optimal nr of clusters, i’m not very fond of the elbow method to do the hyper-param optimisation, as i don’t have direct UI feedback and need to run them in batch…

would really appreciate your input on this!

Thx

H

Hi @hermyknime,

Have you benchmarked DBSCAN (3.7) from Weka on your dataset? You shouldn’t get your hope up, though, since the documentation states:

Basic implementation of DBSCAN clustering algorithm that should not be used as a reference for runtime benchmarks: more sophisticated implementations exist! Clustering of new instances is not supported

We currently don’t have an optimization of our DBSCAN implementation on the agenda…

Best,
Stefan

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.