Changing columns with `+=` gives UnboundLocalError

[wcbeard]

In [1]: import numpy as np
In [2]: import pandas as pd
In [3]: np.random.seed(10)
In [4]: a = np.random.randint(0, 20, (6, 5))
In [5]: df = pd.DataFrame(a, columns=list('abcde'))
In [6]: df
Out[6]:
    a   b   c   d   e
0   9   4  15   0  17
1  16  17   8   9   0
2  10   8   4  19  16
3   4  15  11  11   1
4   8   4  14  17  19
5  13   5  13  19  13

Changing columns with + is fine

In [7]: df.columns = df.columns + '_x'
In [8]: df
Out[8]:
   a_x  b_x  c_x  d_x  e_x
0    9    4   15    0   17
1   16   17    8    9    0
2   10    8    4   19   16
3    4   15   11   11    1
4    8    4   14   17   19
5   13    5   13   19   13

The += operator on the other hand correctly changes the columns,

In [9]: df.columns += '_y'
In [10]: df.columns
Out[10]: Index([u'a_x_y', u'b_x_y', u'c_x_y', u'd_x_y', u'e_x_y'], dtype=object)

but gives an error when trying to display the whole dataframe:

In [11]: df
Out[11]: ---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)

Traceback is here. Not sure if it's an ipython or pandas issue. pd.__version__ == '0.12.0-371-g1434776'

[jreback]

this is a trivial fix, just define __iadd__ = __add__ in core/index.py

need a test, and I think it makes sense to disable methods that don't make sense on an index

mul,trudev,floordiv,pow
imul,itruediv,iflloordiv,ipow

disabling is easy (just set them like incore/base.py/FrozenNDArray )

do a PR?

[wcbeard]

do a PR?

I'd love to, but I'll probably have to wait till after work tomorrow if there's no rush.

[jreback]

no problem...

[wcbeard]

@jreback I think disabling __floordiv__ breaks pandas.tools.tests.test_pivot.TestPivotTable (there's a data.index // 5 line in there). How about disabling all but that one?

[jreback]

just make that

data.index.to_series() // 5

that is an incorrect usage (and the reason these should be disabled!)

[jreback]

pls also add a note in doc/source/release.rst (API section about this change, referencing the issue)..thanks

[wcbeard]

about the __iadd__ fix, or just the disabled methods?

[jreback]

you fixed iadd? I meant also isub too (__isub__ = __sub__

[wcbeard]

Well, fixed insofar as I aliased to add. I'll do isub too.

Sent from my iPhone

On Sep 28, 2013, at 7:35 PM, jreback notifications@github.com wrote:

you fixed iadd? I meant also isub too (isub = sub

—
Reply to this email directly or view it on GitHub.

[jreback]

@d10genes sorry..didn't mean the ?. I saw that you fixed iadd, wanted you to add isub (which you just said you are doing)...so good!

[wcbeard]

Also, just so I understand you, redefining __isub__ this way will only make a difference for set difference between the index and another iterable (since there's not really a string - analogue to string concatenation with +)?

__isub__ currently breaks for set differences like this, and your suggestion would fix it, just want to make sure this was your intention (I didn't know about the set +/- operations with Index before...)

[jreback]

yep....+/- so set ops on Index (with a single string being coerced to an index). Possibly an odd design, but it does make sense for this type of object.

[wcbeard]

Ok, I understand, but I don't think subtracting a single string coerces it to an index:

In [1]: import pandas as pd

In [2]: a = pd.Index(list('abc'))

In [3]: a
Out[3]: Index([u'a', u'b', u'c'], dtype=object)

In [4]: a - 'a'

gives a Exception: Input must be iterable! (Index.diff does a check for __iter__ attribute at the beginning). Should this behavior (not coercing a string to index for -) be filed separately? Or did you mean something else?