Series is missing real/imag attributes

[twmr]

The Series class in 0.10 had a real and a imag attribute but in 0.12dev these attrs are missing. This probably happend during the refactoring of the Series class for 0.12.

In [1]: pd.__version__
Out[1]: '0.10.0'

In [2]: a = np.arange(5)

In [3]: b = pd.Series(a + 4j*a)

In [4]: b.real
Out[4]: 
0    0
1    1
2    2
3    3
4    4

In [1]: pd.__version__
Out[1]: '0.12.0-381-g15d5f0a'

In [2]: a = np.arange(5)

In [3]: b = pd.Series(a + 4j*a)

In [4]: b.real
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-4-ec5ddb077f9e> in <module>()
----> 1 b.real

/usr/lib64/python2.7/site-packages/pandas-0.12.0_381_g15d5f0a-py2.7-linux-x86_64.egg/pandas/core/generic.pyc in __getattr__(self, name)
   1287             return self[name]
   1288         raise AttributeError("'%s' object has no attribute '%s'" %
-> 1289                              (type(self).__name__, name))
   1290 
   1291     def __setattr__(self, name, value):

AttributeError: 'Series' object has no attribute 'real'

[jreback]

you can access like below

These are easy to do (just a delegated property), but is this a typical access pattern?

should we make them methods? what do you do with the complex numbers?

In [1]: a = np.arange(5)

In [2]: 

In [2]: In [3]: b = pd.Series(a + 4j*a)

In [3]: b
Out[3]: 
0         0j
1     (1+4j)
2     (2+8j)
3    (3+12j)
4    (4+16j)
dtype: complex128

In [4]: b.values
Out[4]: array([ 0. +0.j,  1. +4.j,  2. +8.j,  3.+12.j,  4.+16.j])

In [5]: b.values.real
Out[5]: array([ 0.,  1.,  2.,  3.,  4.])

In [6]: b.values.imag
Out[6]: array([  0.,   4.,   8.,  12.,  16.])

[twmr]

I know that - actually, this is what I'm using as a workaround in 0.12dev. Is it intentional that the 'new' Series class does not contain these attrs anymore?

[jreback]

nope...they are apparently properties on numpy ararys; we didn't have any tests that use these so they weren't 'missing'. I'll add them..pretty easy

[jreback]

@Thisch ok...all fixed up...updated to current master and give a try..thanks for the report

[twmr]

Thank you.

Sry that I have not answered all your questions in your first reply. Atm I use complex numbers mainly for working with 2D positions (rotating these positions around a certain point is much easier if the are stored as complex numbers - exp(1j*phi) vs 2x2 Rotationmatrix)).

Simple delegation of the properties is fine

[twmr]

Support for abs() should be added as well. This makes sense for real valued data as well.

In 0.10:

In [4]: b.abs()
Out[4]: 
0     0.000000
1     4.123106
2     8.246211
3    12.369317
4    16.492423

In [5]: abs(b)
Out[5]: 
0     0.000000
1     4.123106
2     8.246211
3    12.369317
4    16.492423

I tried to fix this by adding

    @rwproperty.getproperty
    def __abs__(self):
        return abs(self.values)

to the series class. This has the effect that b.abs() now works but abs(b) doesn't (TypeError: 'numpy.ndarray' object is not callable). Any ideas ?

[jtratner]

why would that be a property?

[jreback]

This is already defined

In [1]:  a = np.arange(5)

In [2]:         b = Series(a + 4j*a)

In [3]: b
Out[3]: 
0         0j
1     (1+4j)
2     (2+8j)
3    (3+12j)
4    (4+16j)
dtype: complex128

In [4]: b.abs()
Out[4]: 
0     0.000000
1     4.123106
2     8.246211
3    12.369317
4    16.492423
dtype: float64

In [5]: np.abs(b)
Out[5]: 
0     0.000000
1     4.123106
2     8.246211
3    12.369317
4    16.492423
dtype: float64

[twmr]

You are right, the property decorator is wrong and both b.abs() an np.abs(b) already work. Some of my code uses abs(b) instead of np.abs(b) and this is raising an exception:

In [5]: abs(b)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-5-de066fb5b1d7> in <module>()
----> 1 abs(b)

TypeError: bad operand type for abs(): 'Series'

I can change my code to use np.abs instead - I just want you to know that abs(Series) is no longer working in 0.12dev.

[jtratner]

I have a PR brewing...it'll be in master shortly.

On Wed, Sep 11, 2013 at 7:09 PM, thomas notifications@github.com wrote:

You are right, the property decorator is wrong and both b.abs() an
np.abs(b) already work. Some of my code uses abs(b) instead of np.abs(b)and this is raising an exception:

In [5]: abs(b)

TypeError Traceback (most recent call last)
in ()
----> 1 abs(b)

TypeError: bad operand type for abs(): 'Series'

I can change my code to use np.abs instead - I just want you to know that
abs(Series) is no longer working in 0.12dev.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4819#issuecomment-24283691
.

[twmr]

Great. Thx for your quick fixes!