Param `mean` in Random.multivariateNormal doesn't need to be a KtNDArray

[alshan]

It is 1-D array_like thing and can be just a std Kotiln collection.

Instead of

Random.multivariateNormal(mean=array(listOf(-2,0)), ...

would be nice to have:

Random.multivariateNormal(mean=listOf(-2,0), ...

[devcrocod]

Alright. I thought that the user will work mainly with ndarray, but in cases where mean is calculated in ndarray, it will be possible to convert using toList().

[alshan]

If KtNDArray implements Iterable then even toList wont be necessary, providing that mean argument had type Iterable.

[devcrocod]

There is one more problem. If we use ndarray, then we simply pass a pointer to it in python, and if we use jvm structure (array, list), then we need to convert it to python array. This is additional overhead.

[alshan]

Isn't the overhead there in any case? mean=array(listOf(-2,0)

[alshan]

Would be nice also to relax the type of parameter cov because now I have to create KtNDArrays when I don't really need them.

Compare how Python is so much more concise than Kotlin:

Py: https://nbviewer.jupyter.org/github/JetBrains/lets-plot/blob/master/docs/examples/jupyter-notebooks/density_2d.ipynb Kt: https://nbviewer.jupyter.org/github/JetBrains/lets-plot-kotlin/blob/master/docs/examples/jupyter-notebooks/density_2d.ipynb

[devcrocod]

Isn't the overhead there in any case?

Yes, there is overhead. Only in case of reuse mean will give an improvement.

I agree that it is necessary to support iterable for mean. I will save the signature with KtNDArray and add method with signature for Iterable. I will also find similar methods where this is necessary.

Of course, all this would be easier if KtNDarray implemented the Iterable interface. Here I described what are the difficulties of this.