ENH: Unify API of groupby.sum and sum

[phil20686]

Is your feature request related to a problem?

The level keyword in sum is marked as deprecated, but the suggested replacement with groupby.sum does not support skipna, therefore, some behaviours are impossible to replicate easily.

Describe the solution you'd like

grouby.sum should accept the skipna=False keyword,

API breaking implications

Currently the API is inconsistent, since
df.sum(axis=1, level=0, skipna=False) has no equivalent since df.groupby(axis=1, level=0).sum(skipna=False) is not available.

Describe alternatives you've considered

One can of course do: df.groupby(axis=1, level=0).apply(lambda x : x.sum(axis=1, skipna=False)) but this cannot be the intended use.

Additional context

# Your code here, if applicable

[samukweku]

@phil20686 kindly add an example to explain better

[simonjayhawkins]

@phil20686 kindly add an example to explain better

and if you could add some references to support the discussion and other readers that'll be great too. e.g. the deprecation and alternatives, where this was discussed etc.

and update the issue title to be more informative too!

[rhshadrach]

I believe df.sum(axis=1, level=0, skipna=True) should always be equivalent to df.groupby(axis=1, level=0).sum(); can you provide an example where this is not the case?

[phil20686]

I meant false sorry, I have added an example. There seems to be no way to eradicate this future warning without changing the behaviour, except resorting to something unpleasant like using apply instead of sum.

[rhshadrach]

@phil20686 - can you add a plaintext example rather than a screenshot; this allows devs to copy and paste rather than having to retype the entire code.

[rhshadrach]

One can of course do: df.groupby(axis=1, level=0).apply(lambda x : x.sum(axis=1, skipna=False)) but this cannot be the intended use.

This is just about how pandas implements df.sum(level=0); only difference is that pandas uses .agg instead of .apply.

[rhshadrach]

I'm +0 on adding skipna to groupby. The deprecated level argument does make users supply a replacement that is a bit more verbose, but just as efficient, as pandas is using today. While certainly not great, I also think it's not terrible. If we do go this route, then it shouldn't just be sum of course, but (at least) all ops of DataFrame that have skipna.

It seems to me to be desirable to add this as a fully baked in feature of the groupby code, rather than bolted on using agg/apply, but I'm not sure how difficult that might be.