see pydata/xarray#7641 for an example for the conversion from loops to groupby.

The intro material should include tips on how to visualize those high-level objects (like, use .construct for rolling or call dict on groupby objects). I find that looking at a small sample of what they contain really helps when trying to come up with ways to transform the high-level object back to a Dataset / DataArray.

Also, maybe this is something for the main repository, but calling arr.rolling(...).construct("window_dim").isel(...) to get something that is easy to understand might be a bit much to write, so we probably should modify the repr of those objects to be a bit more informative?

Just chatted with @maxrjones .

Another concept to emphasize with rolling is views vs copies after calling .construct. And that if you have a vectorize-able function calling it on the output of .construct is probably a lot better than using Rolling.__iter__

You can also check the my links. We have used some of Xarray high computational patterns. I will be posting some more of my work which combine xarray with other packages like gdal