API/CLN: add in common operations to Series/Index, refactored as a OpsMixin

[jreback]

closes #4551
closes #4056
closes #5519

allow a Series to utilize index methods for its index type, e.g. Series.year is now defined
for a Series with a DatetimeIndex or a PeriodIndex; trying this on an Index type will
now raise a TypeError. The following properties are affected:
date,time,year,month,day,hour,minute,second,weekofyear
week,dayofweek,dayofyear,quarter,microsecond,nanosecond,qyear
and methods: min(),max(),

In [1]: s = Series(np.random.randn(5),index=tm.makeDateIndex(5))

In [2]: s
Out[2]: 
2000-01-03   -0.301523
2000-01-04    1.104868
2000-01-05    0.321592
2000-01-06   -0.565114
2000-01-07    0.888334
Freq: B, dtype: float64

In [3]: s.year
Out[3]: 
2000-01-03    2000
2000-01-04    2000
2000-01-05    2000
2000-01-06    2000
2000-01-07    2000
Freq: B, dtype: int64

In [4]: s.index.year
Out[4]: Int64Index([2000, 2000, 2000, 2000, 2000], dtype='int64')

In [5]: Series(np.random.randn(5)).year
TypeError: cannot perform an year operations on this type <class 'pandas.core.index.Int64Index'>

[cpcloud]

This looks cool. Slightly OT would it be useful to have a method to split dates into their parts? Something like split on datetime index? Would be useful for group by days months etc. although maybe there's a way to do that now

[cpcloud]

Or date_split since split is a bit too general sounding

[jreback]

you can just do index.date and index.time (which are not affected by this....)

I have to put a list of follow up stuff....

[jreback]

With this PR (note that DataFrame support missing for this, and I am not sure that we should add it
because then you could easily have name collisions with the method names, much less of an issue with Series)

In [4]: df.groupby(df.index.month).sum()
Out[4]: 
     0
1   12
2   14
3   16
4   18
5   20
6   22
7   24
8   26
9   28
10  30
11  32
12  34

[12 rows x 1 columns]

In [5]: s = Series(np.arange(24),index=date_range('20130101',periods=24,freq='MS'))

In [6]: s.groupby(s.month).sum()
Out[6]: 
1     12
2     14
3     16
4     18
5     20
6     22
7     24
8     26
9     28
10    30
11    32
12    34
dtype: int64

[jreback]

@jorisvandenbossche I have to push an update because the doc strings are not set...

so I can get

help(Series.month)

to work properly

but ipython help is odd

Series.month?

any idea how to do this?

[jorisvandenbossche]

@jreback Do you mean that you get a verbose output fget/fset for properties with the ipython help ?? This is at the moment also the case for other properties. But in the Docstring: .. part, there should be an explanation of the property.

[jreback]

yes exactly

hmm ok so this is a known issue then
ok

I put these in the API as well
hopefully will build

[jreback]

@cpcloud @jtratner @jorisvandenbossche any more comments on this?

[cpcloud]

Did you mean to use @wraps here?

[jreback]

I think I originally did, but sort of did it 'manually' by assigning name/doc string....bad?

[cpcloud]

Looking at the source for wraps it basically does that ... only diff is that it updates the __module__ attribute and copies the __dict__ attribute. Not "bad" per se.

[cpcloud]

This is fine ... I was just asking.

[jreback]

@cpcloud I took wraps out...not necessary as I am creating the function wrappers here from an accesssor, rather than wrapping another function