fromnumeric.py compatibility with GroupBy, Window, and tslib functions

[gfyoung]

In #12810, it was decided that compatibility for groupby (including resample.py) and window functions would be left for a separate PR / discussion, which seems reasonable given how massive #12810 already is. This issue serves a reminder to tackle this after landing #12810, as it seems like this can be easily addressed afterwards.

[gfyoung]

@jreback : Adding timestamps and timedeltas as well to this issue given my question about tslib.pyx (i.e. what sort of compatibility should we give to methods with numpy counterparts?)

[gfyoung]

With my initial fromnumeric.py PR merged, it seems like a good idea to revisit this. The major files that I think merit examination for numpy compatibility are:

pandas/core/window/window.py
pandas/tseries/resample.py
pandas/core/groupby.py
pandas/tslib.pyx

[jreback]

how so, I don't really care to be compat with numpy for anything beyond very basic stuff. pls show an example.

[gfyoung]

Examples:

tslib.round(self, freq) vs. np.round(a, decimals=0, out=None)

window.max(self, how=None, **kwargs) vs. np.max(a, axis=None, out=None, keepdims=False)

resample.var(self, ddof=1) vs. np.var(a, axis=None, out=None, ddof=0, keepdims=False)

groupby.mean(self) vs. np.mean(a, axis=None, out=None, keepdims=False)

All I was thinking of doing was putting validation calls in the implementation, similar to what was done in my previous PR and nothing more than that. I'm also perfectly fine leaving them as is since numpy decoupling is also one of our objectives with pandas.

[jreback]

In [1]: df = DataFrame({'A' : [1,2,1], 'B' : [1,2,3]})

In [2]: g = df.groupby('A')

In [3]: g.mean()
Out[3]: 
   B
A   
1  2
2  2

In [4]: np.mean(g)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-fdcd38489fcb> in <module>()
----> 1 np.mean(g)

/Users/jreback/miniconda/lib/python2.7/site-packages/numpy/core/fromnumeric.pyc in mean(a, axis, dtype, out, keepdims)
   2878         try:
   2879             mean = a.mean
-> 2880             return mean(axis=axis, dtype=dtype, out=out)
   2881         except AttributeError:
   2882             pass

TypeError: mean() got an unexpected keyword argument 'axis'

[jreback]

ok that doesn't seem unreasonable

[jorisvandenbossche]

Do we actually want something like np.mean(g) to work?
A groupby object is not an array-like such as a Series. IMO we shouldn't put effort in enabling such usage

[gfyoung]

@jorisvandenbossche : I'll leave that for you to debate with @jreback . To reiterate, I am perfectly fine either way. This is not as serious a compatibility issue as the previous one I raised in #12644.