A modern programming language that makes developers happier.

kotlinlang

Right now it does not. I am not sure about plans about that.  Do you plan to work on JVM or in multiplatform? I believe Commons-math supports sparse matrices. Not sure about arrays

And it would be nice if you could explain the use case for sparse arrays

JVM only for now.

A sparse matrix fulfills the need :)
Essentially I'm using it for sparse matrices where I sometimes extract it into arrays. But an array is really just a 1D-matrix so no worries
How about kmath, does that support it?

Not yet, since there was no request for it. We can add it to commons-math or nd4j wrappers (nd4j also has sparse matrices) if you could formulate what do you need in terms of operations.

It's mainly for feature vectors (like bag-of-words) and one-hot-encodings where there's more 0:s than 1:s.
Would be cool with int-matrices like multik support but float/double obviously works great too :)

So you need only vectors themselves? Do you need addition, subtraction, inversion etc?

In the library I also do addition, subtraction, SVD and much more. Would be nice to share library/data structure throughout everything

If you do not need operations, you can use <https://commons.apache.org/proper/commons-math/javadocs/api-3.6.1/index.html> for real-valued sparse vectors and <https://commons.apache.org/proper/commons-math/javadocs/api-3.6.1/index.html> for int-valued vectors. And we can add wrappers on top of it in kmath-commons.

I'll investigate tomorrow thank you :smile:

<@URGJLKD4N> torch has it: <https://pytorch.org/docs/stable/sparse.html>

<https://github.com/mipt-npm/kmath/tree/feature/noa>

I would advise to just simply write your calculation in python and torchscript it. You can then run that on the JVM.

Will check that too tomorrow! :slightly_smiling_face:

I think if your calculation is just about manipulating tensors, then instead of using all those libraries with multidimensional arrays we have in Kotlin, I would just torchscript the whole calculation in python and then run that on the JVM side.

with time I am just going to open a bit more of the functionality around that in kmath

There is an EJML wrapper <https://github.com/mipt-npm/kmath/blob/dev/kmath-ejml/src/main/kotlin/space/kscience/kmath/ejml/_generated.kt#L542>