We are developing a library for statistical plotti...

# datascience

[Igor Alshannikov]

We are developing a library for statistical plotting which is pure Kotlin and multiplatform: https://github.com/JetBrains/lets-plot

The library includes some statistical algorithms - in particular, smoothing algorithms. We've picked some low hanging fruit - linear model (or lm) smoothing - although we have not yet implemented more complex algorithms such as loess, glm or gam.

Is anybody aware of a pure-Kotlin statistical library in existence or in development which might include said algorithms?

[altavir]

The nexus of univariate statistics in kotlin is @thomasnield's kotlin-statistics: https://github.com/thomasnield/kotlin-statistics. If you need only only univariate case, please put an issue there and I will work on it. The loess interpolator is implemented in commons-math and I can easily port it to multiplatform. I plan to add interpolation API to kmath: https://github.com/mipt-npm/kmath/issues/40, but it will take more time because I want to develop robust API for generic multiavariate case. We can start with the simple algorithm in kotlin-statistics and then back-port it to the new kmath API. In either case please open an issue.

[altavir]

Also Lets-plot is very important for scientific ecosystem, but we need to wrap it in more kotlinized API.

[Igor Alshannikov]

Sounds very good, I've submitted the issue to both projects:
https://github.com/thomasnield/kotlin-statistics/issues/44
https://github.com/mipt-npm/kmath/issues/79

[altavir]

thanks a lot. I will try to find the time to work on Interpolation API. When I am done, it will be quite ease to add loess.

[thomasnield]

@altavir please keep me posted.

[Igor Alshannikov]

Guys, we've managed loess, check this out: https://nbviewer.jupyter.org/github/JetBrains/lets-plot/blob/master/docs/examples/jupyter-notebooks/geom_smooth.ipynb

[altavir]

OK, sorry, I did not manage to make it past the conversion stage. Can you give a reference to the actual code implementing the feature?

[altavir]

I've assigned a student for this task. Some interpolation API is ready, but not yet tested: https://github.com/mipt-npm/kmath/blob/dev/kmath-functions/src/commonMain/kotlin/scientifik/kmath/interpolation/SplineInterpolator.kt

[Igor Alshannikov]

Cool, somebody asked me about spline interpolation on line plots. May be I can use this when ready.

[Igor Alshannikov]

This is the local regression entry point: https://github.com/JetBrains/lets-plot/blob/master/plot-base-portable/src/commonMain/kotlin/jetbrains/datalore/plot/base/stat/regression/LocalPolynomialRegression.kt

[altavir]

Oh, it is much more simple than the commons-math code I was porting. My current problem is that I want to do it in a generic way, so we could use interpolation for any value with defined field, but of course it is not what you need for simple plotting. I will try to dedicate more time to the problem, saly I do not have a lot of it right now.