Documentation of `Dataset.count()` is ambiguous

[Articoking]

What is your issue?

Hello everyone!

I've noticed that the documentation for Dataset.count and DataArray.count is a little unclear with respect to what it actually does. In my experience, it's not immediately obvious when looking at this site whether zero values are counted or not.

The given example does not contain a 0 value that would clarify this, and what is worse is that the "see also" section points to np.count and to dask.array.count which do not exist. The closest NumPy equivalent is np.count_nonzero, which has a different behavior to the xarray count method. The xarray count is actually equivalent to Pandas DataFrame.count, which count all non-NA values including 0.

I would take a shot at writing a clearer docstring but I'm not sure how to do that using the generate_aggregations.py file...

[headtr1ck]

Thanks for the issue, many aggregations could indeed use some extra explanation.

Concerning the dead links, that's really not good. Probably we should look into letting sphinx report them.

[keewis]

that's what nitpicky is for, but we had a lot of these warnings the last time I checked, so we'd need to clean them up first (or mark them as ignored, some of them are descriptions where there should be type descriptions, which we mostly inherit from pandas / matplotlib / numpy).

[dcherian]

See also #7378

The xarray count is actually equivalent to Pandas DataFrame.count, which count all non-NA values including 0.

This is correct.

I'm not sure how to do that using the generate_aggregations.py file

I would start by adding a long_name and/or description property on Method that lets you add customizable text for each method.

what is worse is that the "see also" section points to np.count and to dask.array.count which do not exist.

The "See Also" piece is harder. Potentially instead of the current template we generate that too by adding a see_also_modules to Method. By default this would be ['numpy', 'dask.array', 'pandas.DataFrame'] but for count we can just have ['pandas.DataFrame'].

---

An alternate solution would be to manually fix the docstring for count and copy it to all the necessary classes. It is a bit of an outlier in that it's not a "standard" array-api type reduction.