improved RollingFunctions.jl integration

[simeonschaub]

RollingFunctions.jl already works with AbstractDimArrays, but unfortunately the dimension is lost in the process. Python's xarray has a method rolling which keeps the original dimensions of the array and even has a particularly useful center argument to specify that the labels should be set at the center of the window.
It also supports a dim argument which RollingFunctions.jl doesn't have AFAIK, but I'm thinking that this might not be necessary and with #850, one could just use eachslice for this.
Would it be reasonable to add an extension here supporting such a use case?

[lazarusA]

mmm... I thought this was covered by https://rafaqz.github.io/DimensionalData.jl/stable/groupby#binning and https://rafaqz.github.io/DimensionalData.jl/stable/groupby#Select-by-Dimension, maybe we could show it more, if indeed is the same.

[simeonschaub]

IIUC binning is not the same as applying a rolling function though, since bins are disjoint whereas the point of rolling functions is that there are sliding windows that overlap.

[lazarusA]

well, for moving windows (xarray-like) we have this in yaxarrays, but is not documented yet.

JuliaDataCubes/YAXArrays.jl#459

[rafaqz]

I was thinking we needed an extension, but actually Stencils.jl doesn't and still returns a DimArray;

https://github.com/rafaqz/Stencils.jl/blob/main/test%2Farray.jl#L170-L172

So this is a RollingFunctions.jl problem. To fix it generically for all input array types, they probably need to use similar to generate the output array from the input, and then write into it.

[rafaqz]

This is what Stencils does:

https://github.com/rafaqz/Stencils.jl/blob/main/src%2Fmapstencil.jl#L28-L31

They can just copy that and the tests

[simeonschaub]

well, for moving windows (xarray-like) we have this in yaxarrays, but is JuliaDataCubes/YAXArrays.jl#459 yet.

JuliaDataCubes/YAXArrays.jl#459

I looked into this, but the mapCubes API seems very complex for simply calculating a rolling mean

[simeonschaub]

I ended up creating my own package in https://github.com/simeonschaub/RollingWindowArrays.jl for rolling window operations since I wasn't really satisfied with RollingFunctions.jl because its implementation is quite complex and hard to extend. I think this issue can probably be closed.