ENH: enable skipna on groupby reduction ops

[jreback]

Edit[rhshadach]: The following methods do not have a skipna argument in groupby, but do have such an argument on the Series/DataFrame variant.

• max
• mean
• median
• min
• prod
• sem
• std
• sum
• var

---

#15674

In [19]: import pandas as pd
    ...: import numpy as np
    ...: d = {'l':  ['left', 'right', 'left', 'right', 'left', 'right'],
    ...:      'r': ['right', 'left', 'right', 'left', 'right', 'left'],
    ...:      'v': [-1, 1, -1, 1, -1, np.nan]}
    ...: df = pd.DataFrame(d)
    ...: 

In [20]: df.groupby('l').v.sum()
Out[20]: 
l
left    -3.0
right    2.0
Name: v, dtype: float64

In [21]: df.groupby('l').v.apply(lambda x: x.sum(skipna=False))
Out[21]: 
l
left    -3.0
right    NaN
Name: v, dtype: float64

ideally write [21] as
df.groupby('l').v.sum(skipna=False)

[mayukh18]

Can I take this up?

[jreback]

that would be great @mayukh18 !

[mayukh18]

@jreback I was thinking about putting a passthru check for specifically the how == 'add' case a level above and after the actual cython implementation. Does this seem like a good idea?

[jreback]

no, you simply need to add this on to _groupby_function near the top. Needs testing for all ops which support skipna (pretty much all numeric ops)

[mayukh18]

Adding this in _groupby_function will not effect any change for ops like mean, median etc as they are separately implemented. So, do those need the skipna too? Whereas, first and last ops don't seem like requiring skipna but they will be effected by modifying _groupby_function. I am naive so pardon me if I am asking silly stuff.

[jreback]

@mayukh18 yes the implementation is a bit muddled.

methods which are defined with an actual function (e.g. median/mean), this is very easy, just add to the validation function where numeric_only is now; that will pass it thru.

we can still use _groupby_function, what I would do is this.

add the default for skipna (so do this on all but first/last)
sum = _groupby_function('sum', 'add', np.sum, skipna=True)

then add skipna=None in the signature for _groupby_function, which will add the arg if its not None to kwargs (so it gets passed).