ENH: Add Index.normalize as a Series or DataFrame method

[cancan101]

Currently normalize does not work on a Series or a DataFrame:

pd.Series(range(5), index=pd.date_range('2013-1-1', periods=5,freq='D')).normalize()
AttributeError: 'Series' object has no attribute 'normalize'

However, changing the timezone, which I consider to be a similar type of operation, does work:

In [50]: pd.Series(range(5), index=pd.date_range('2013-1-1', periods=5,freq='D')).tz_localize('America/New_York')
Out[50]: 
2013-01-01 00:00:00-05:00    0
2013-01-02 00:00:00-05:00    1
2013-01-03 00:00:00-05:00    2
2013-01-04 00:00:00-05:00    3
2013-01-05 00:00:00-05:00    4
Freq: D, dtype: int64

My current workaround:

s = pd.Series(range(5), index=pd.date_range('2013-1-1', periods=5,freq='D'))
s.index = s.index.normalize()

[jreback]

related #4551

[hayd]

What are you expecting normalize to do?

Surely it would make more sense for it to be the following:

In [1]: df
Out[1]: 
   A  B
0  1  2
1  3  4
2  5  6

In [2]: df / np.sum(df.values)
Out[2]: 
          A         B
0  0.047619  0.095238
1  0.142857  0.190476
2  0.238095  0.285714

In [3]: df.div(df.sum())
Out[3]: 
          A         B
0  0.111111  0.166667
1  0.333333  0.333333
2  0.555556  0.500000

In [4]: df.div(df.sum(1), axis=0)
Out[4]: 
          A         B
0  0.333333  0.666667
1  0.428571  0.571429
2  0.454545  0.545455

[cancan101]

If we are going down that route, I would argue that normalize is pretty vague as it can currently be used. to_midnight is far more explicit. That being said, I would be fine with to_midnight being the new method added on a Series or DataFrame.

[hayd]

I think similar to Series string methods, we should put DatetimeIndex methods (like normalize) in a class. I'm not sure these make sense in DataFrame but certainly useful for Series.

[pwaller]

I just went looking for a normalize function, expecting df.div(df.sum()) (as in @hayd's comment). The problem I have is that df is a chained pipe expression, so I went with (for a series) s.pipe(lambda s: s.div(df.sum())). I did also try df.div(lambda df: df.sum()), but I don't know if that's in the right spirit or not.

[jbrockmendel]

-1, the API is too big as it is, plus it isn't obvious that it refers to the index and not the data

[jreback]

yep we already have .dt to direct this