ENH: Using built-in round on a series

[Dr-Irv]

Is there some philosophical reason that round(Series) and round(Dataframe) is not defined? See the example below. Being able to write round() as a function of a Dataframe or Series, as opposed to doing it afterwards would be nice. I'm willing to implement __round__() to make it work, but don't want to go down that path if there is some reason __round__() is not implemented.

Example:

    import pandas as pd
    import numpy as np
    df =  pd.DataFrame(np.random.randn(10, 3), columns=['A', 'B', 'C'])
    print (df.A.round())
    print (round(df))

Produces

0   -2
1    1
2   -1
3   -1
4    2
5    0
6    0
7    0
8   -2
9    0
Name: A, dtype: float64
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-8-9d2be9cbdc70> in <module>()
      3 df =  pd.DataFrame(np.random.randn(10, 3), columns=['A', 'B', 'C'])
      4 print (df.A.round())
----> 5 print (round(df))

TypeError: type DataFrame doesn't define __round__ method

[jreback]

sure you could define that to call the .round() method; do it in NDFrame (Panel will fail on this ATM so you could define .round(..) for that to call np.round(....), while Series/DataFrame have .round defined

[srib]

@jreback Series has .round supported. DataFrame doesn't seem to have one. Did you mean otherwise in your comment?

I would be happy to send a PR for this.

[jreback]

added in 0.17.0: http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.round.html?highlight=round#pandas.DataFrame.round

[srib]

@jreback Thanks!

[Dr-Irv]

@jreback So I have something that works fine with Python 3, but Python 2 presents a problem because in Python 2, round(class) doesn't call class.round() . In other words, you can't easily override the built-in round() behavior in Python 2. I'm not sure of the best way to handle this. I could just avoid the round() test in the testing code when doing a Python 2 test and we don't worry about supporting round(DataFrame) with Python 2. Other suggestions welcome.

[jreback]

Python 2.7.10 |Continuum Analytics, Inc.| (default, Oct 19 2015, 18:31:17)
Type "copyright", "credits" or "license" for more information.

Obviously you wouldn't actually call object.__round__, but this works in py2.

In [3]: class F(float):
   ...:     def __round__(self):
   ...:         return object.__round__(self)
   ...:     

In [4]: F(5.2)
Out[4]: 5.2

In [5]: round(F(5.2))
Out[5]: 5.0

[Dr-Irv]

@jreback I don't think your example applies, because we want round(DataFrame) to work correctly, and your example worked OK because you were subclassing a float object. If you changed return object.__round__(self) to return 98, you'd see that. The issue is that in py2, round() calls float() and then does the rounding. py3 allows round() to be overridden on a class by implementing __round__() One can override float() by implementing __float__(), but I don't think we want to override the float() behavior for pandas. Found this article on stackoverflow related to this. http://stackoverflow.com/questions/9751476/float-and-round-in-python-2-and-3

[jreback]

@Dr-Irv ok, that IS an ugly hack. So just test this on PY3, that is fine. This is really just a convenience feature anyhow, it really breaks up the flow, but if people really want to use it then it would work e.g.

df.round(....) vs round(df)