Proposal to Integrate ST-DBSCAN Algorithm into mlr3cluster

[MiboraMinima]

Hi everyone,

First of all, thank you so much for your work! I really enjoy using mlr3, and I appreciate that it integrates clustering into the same framework, which is fantastic for maintaining consistency in workflows.

Currently, there is no R package that implements ST-DBSCAN (as far as I know), and, therefore, no integrated framework that incorporates it.

In case you're unfamiliar with it, ST-DBSCAN is a variant of DBSCAN designed to find clusters in 3D space, typically spatio-temporal data (hence the "ST"). One common application is the clustering of stop-and-move behavior in GPS datasets.

While no packages currently implement this algorithm, there is an R code available here.

Would you be interested in integrating it into mlr3cluster? If so, how can I contribute to this effort?

[m-muecke]

Always open for taking feature requests. Alas, the current policy to integrate a new learner is that the package is on CRAN and well maintained and tested. Depending on complexity of the learner there might be considerations to directly add the implementation, which I don't prefer. I'm not familiar with ST-DBSCAN, but since you mentioned it's a variation of DBSCAN, I would suggest to open an issue in the dbscan package for possible integration, since that already implements DBSCAN, HDBSCAN, etc. and is also already included in mlr3cluster.

[MiboraMinima]

Thank you for your reply. Ok, I'll do that!

[m-muecke]

I had a look at the reference implementation, which I'm not too happy with. There is however a Python implementation that just extends calls to DBSCAN, which seems cleaner and we could implememt the same: https://github.com/eren-ck/st_dbscan.
There are two open questions:

1. Does ST-DBSCAN work on numeric features so that it can simply be integrated or not?
2. Where to put the implementation? I'm prone to either make a small package for CRAN and import {dbscan} or even better {dbscan} integrates it and we provide a PR with the implementation.

I won't have time for it right now, but should be nice for the near future.

Reference for future mhahsler/dbscan#83, most likely needs to be an extension package instead of a PR in {dbscan}.