Breaking examples due to resample refactor

[jorisvandenbossche]

While using master a bit, I discovered some more cases where the new resample API breaks things:

• Plotting. .plot is a dedicated groupby/resample method (which adds each group individually to the plot), while I think it is a very common idiom to quickly resample your timeseries and plot it with (old API) eg s.resample('D').plot().
  Example with master:

  In [1]: s = pd.Series(np.random.randn(60), index=date_range('2016-01-01', periods=60, freq='1min'))
  
  In [3]: s.resample('15min').plot()
  Out[3]:
  2016-01-01 00:00:00    Axes(0.125,0.1;0.775x0.8)
  2016-01-01 00:15:00    Axes(0.125,0.1;0.775x0.8)
  2016-01-01 00:30:00    Axes(0.125,0.1;0.775x0.8)
  2016-01-01 00:45:00    Axes(0.125,0.1;0.775x0.8)
  Freq: 15T, dtype: object
  

  

  while previously it would just have given you one continuous line.
  This one can be solved I think by special casing plot for resample (not have it a special groupby-like method, but let it warn and pass the the resample().mean() result to Series.plot() like the 'deprecated_valids')

• When you previously called a method on the resample result that is also a valid Resampler method now. Eg s.resample(freq).min() would previously have given you the "minimum daily average" while now it will give you the "minimum per day".
  This one is more difficult/impossible to solve I think? As you could detect that case if you know it is old code, but cannot distinguish it from perfectly valid code with the new API. If we can't solve it, I think it deserves some mention in the whatsnew explanation.

• Using resample on a groupby object (xref #12202). Using the example of that issue, with 0.17.1 you get:

  In [1]: df = pd.DataFrame({'date': pd.date_range(start='2016-01-01', periods=4,
  freq='W'),
  ...:                'group': [1, 1, 2, 2],
  ...:                'val': [5, 6, 7, 8]})
  
  In [2]: df.set_index('date', inplace=True)
  
  In [3]: df
  Out[3]:
        group  val
  date
  2016-01-03      1    5
  2016-01-10      1    6
  2016-01-17      2    7
  2016-01-24      2    8
  
  In [4]: df.groupby('group').resample('1D', fill_method='ffill')
  Out[4]:
                  val
  group date
  1     2016-01-03    5
    2016-01-04    5
    2016-01-05    5
    2016-01-06    5
    2016-01-07    5
    2016-01-08    5
    2016-01-09    5
    2016-01-10    6
  2     2016-01-17    7
    2016-01-18    7
    2016-01-19    7
    2016-01-20    7
    2016-01-21    7
    2016-01-22    7
    2016-01-23    7
    2016-01-24    8
  
  In [5]: pd.__version__
  Out[5]: u'0.17.1'
  

  while with master you get:

  In [29]: df.groupby('group').resample('1D', fill_method='ffill')
  Out[29]: <pandas.core.groupby.DataFrameGroupBy object at 0x0000000009BA73C8>
  

  which will give you different results/error with further operations on that. Also, this case does not raise any FutureWarning (which should, as the user should adapt the code to groupby().resample('D').ffill())

[jreback]

ok

1. just need to define .plot(...) on the Resampler to actually call .mean().plot(...) (and have a nice warning message)

similar / better than this

In [5]: s.resample('15min',how='sum').plot()
/Users/jreback/miniconda/bin/ipython:1: FutureWarning: how in .resample() is deprecated
the new syntax is .resample(...).sum()
  #!/bin/bash /Users/jreback/miniconda/bin/python.app
Out[5]: <matplotlib.axes._subplots.AxesSubplot at 0x11bd82450>

3. need to provide warnings here as well (this is handled by the .groupby(...).resample(...) which actually calls things, but hits a different path

[jreback]

2. I think you are right, have to doc this. It should break code loudly though as the previous API would return a scalar, this will return a Series.