API: implement __finalize__ for resample et al.

[JackKelly]

I'm really salivating at the chance to use the ._metadata and __finalize__ mechanisms in pandas-0.13-dev to create my own subclass of DataFrame and to have metadata propagate after calling functions inherited from DataFrame like dropna(), resample() etc.

Using the latest 0.13-dev version of Pandas, I think I might have bumped into a small bug (although I'm not sure if this is a bug or not??):

._metadata propagates after calling .resample(rule='D') and dropna() but not after calling .resample(rule='D', how='max').

More details, including the full code of my subclass, are given under the "experiments" heading of this issue: nilmtk/nilmtk#83

[jreback]

can you put up a small reproducible example ?

actually suprised it works at all with resample (finalize has to be explicity called, and I don't think was put in)

[JackKelly]

Sure, here's the small example:

import pandas as pd

class ElectricDataFrame(pd.DataFrame):
    def __init__(self, *args, **kwargs):
        super(ElectricDataFrame, self).__init__(*args, **kwargs)
        self._metadata = ['test']

    @property
    def _constructor(self):
        print("_constructor called")
        return ElectricDataFrame

edf = ElectricDataFrame([1,2,3,4,5], pd.date_range('2010', freq='D', periods=5))

Interactive:

In [5]: pd.__version__
Out[5]: '0.13.0-75-g7d9e9fa'

In [6]: edf._metadata
Out[6]: ['test']

In [7]: edf.dropna()._metadata
_constructor called
Out[7]: ['test']

In [8]: edf.resample(rule='D')._metadata
_constructor called
Out[8]: ['test']

In [9]: edf.resample(rule='D', how='max')._metadata
Out[9]: []

In [10]: edf.resample(rule='D', how='min')._metadata
Out[10]: []

In [11]: edf.resample(rule='D', how='mean')._metadata
Out[11]: []

[jreback]

thanks....I think since its daily data already and how is defaulted ,then its actually returning the same frame (that's why it works). This is not-implemented (though trivial to do)....will do for 0.13.1

[JackKelly]

since its daily data already and how is defaulted ,then its actually returning the same frame (that's why it works)

ah, interesting!

This is not-implemented (though trivial to do)....will do for 0.13.1

wonderful, thanks loads ;)

[jreback]

you could easily do it in your sub though...

class ElectricDataFrame(pd.DataFrame):

       def resample(self, *args, **kwargs):
           return super(ElectricDataFrame, self).resample(*args, **kwargs).__finalize__(self)

[JackKelly]

you could easily do it in your sub though...

Good point, thank you

[JackKelly]

awesome; thank you!