sum() or prod() on an empty of all-NaN series

[karimpedia]

In the documentation on Working with Missing Data in the latest release (0.22) it is mentioned that:

With sum or prod on an empty or all-NaN Series, or columns of a DataFrame, the result will be all-NaN.
Warning: These behaviors differ from the default in numpy where an empty sum returns zero.

while in fact, sum or prod on an empty or all-NaN Series behaves exactly similar to numpy, and this is also the case with the code sample given there:

In [1]: pd.Series([np.nan]).sum(), pd.Series([]).sum()
Out [1]: (0.0, 0.0)

In [2]: np.nansum([np.nan]), np.sum([])
Out [2]: (0.0, 0.0)

In [3]: pd.Series([np.nan]).prod(), pd.Series([]).prod()
Out [3]: (1.0, 1.0)

In [4]: np.nanprod([np.nan]), np.prod([])
Out [4]: (1.0, 1.0)

I wonder whether that's a typo in the docs or is it just not clear enough ?

[jorisvandenbossche]

@karimpedia Thanks for spotting that!
That's an oversight and should be updated (we changed the behaviour in the 0.22 release (https://pandas-docs.github.io/pandas-docs-travis/whatsnew.html#v0-22-0-december-29-2017), but clearly forgot to update the docs in that place)

[arminv]

On the same page (just above the section mentioned by @karimpedia) it reads:

Methods like cumsum and cumprod ignore NA values, but preserve them in the resulting arrays.

Isn't it better to say something like:

Cumulative methods like cumsum and cumprod ignore NA values by default, but preserve them in the resulting arrays. To override this behaviour and include NA values, use skipna=False.

[pulkitmaloo]

Can I work on this?

[jorisvandenbossche]

@karimpedia where you already working on a PR? Otherwise, @pulkitmaloo sure!

@arminv that's a good suggestion.

[arminv]

@jorisvandenbossche Very minor but I noticed that some methods are hyperlinked to their docstrings (isna, notna, dropna, etc) while others are not (cumsum, cumprod, ffill, bfill, etc). Is there a good reason for this inconsistency? If not then maybe add links for these too?

[jorisvandenbossche]

Yes, in general always welcome to add more such links.

[karimpedia]

@jorisvandenbossche no I was not, so @pulkitmaloo please go ahead !

[pulkitmaloo]

Please review PR #20424
This is my first PR so I'm not sure if I documented the Pull Request in a proper format, please review the comments.