Is there a way to do <HDBSCAN> clustering using ko...

# datascience

[Dumitru Preguza]

Is there a way to do HDBSCAN clustering using kotlin tools ?
https://scikit-learn.org/stable/modules/clustering.html#clustering

[Dumitru Preguza]

@altavir maybe you know ?

[Yogeshvu]

Perhaps https://tribuo.org/learn/4.3/tutorials/clustering-hdbscan-tribuo-v4.html ?

[Dumitru Preguza]

Is something official from Kotlin team planed ?
I would like something more compatible with the Dataframe library
I guess it's a gap in the data tooling of kotlin

[altavir]

I do not think that Koltin team will do anything "official" for specific math/sciecntific tools. It is counterproductive. Scikit-learn is not "official" either, it is a community-driven project. It works the same way. The simplest way is to use one of several existing Java tools. You can also request a Kotlin wrapper so we can add it to KMath or like, or created your own simple wrapper. It does not make any sense to rewrite algorithms like this in Kotlin right now. I don't see any scenario where they are used outside JVM.

[Dumitru Preguza]

Okay, maybe just in my mind the python to -> kotlin ecosystem looks like this:
pandas -> kotlin dataframe
plotly -> kandy, plotly
Scikit-learn -> KotlinML, KMath, but still some gaps to support the full functionality, that's what I meant

I agree that wrappers are enough to not waste resources, I found Smile has some of the algrms. and a Kotlin wrapper: https://haifengl.github.io/clustering.html

[altavir]

Kandy is not a replacement for plotly. Kotlin ecosystem has plotly as well (https://github.com/SciProgCentre/visionforge/tree/master/plotly-kt).

The difference between Python and Kotlin ecosystem is that Kotlin already has a lot of libraries from Java. You do not have to reinvent the wheel, you just need to spend a bit of time to adapt a wrapper.

By the way, python libraries are mostly wrappers of C/Fortran/C++ libraries as well. In Kotlin it is just a lot easier to create a wrapper.

[Dumitru Preguza]

I know Kotlin ecosystem has plotly as well

[altavir]

Kandy tries to solve some problems that arise in Plotly and make plotting more or less framework-independent. But it has both good sides and bad sides.

[Dumitru Preguza]

I checked it, did not know you are the main contributor of visionforge, plotly kt

[Dumitru Preguza]

I did not use it extensively, I stick with the dataframe+kandy combo

[altavir]

Yes, Plotly-kt was a standalone project, but now it is a part of VisionForge since they share a lot of logic.