BUG: duplicate indexing on non-integer index with positional indexers failing in py3

[alexanderwhatley]

Code Sample, a copy-pastable example if possible

df = pd.DataFrame({"a" : [0,1,2], "b" : [1,2,3]})
df[["a", "a"]].apply(lambda x: x[0] + x[1], axis = 1)

Expected Output

0 0
1 2
2 4
dtype: int64

output of pd.show_versions()

Traceback (most recent call last):
File "C:\Users\Alexander\Anaconda3\lib\site-packages\pandas\indexes\base.py", line 1980, in get_value
tz=getattr(series.dtype, 'tz', None))
File "pandas\index.pyx", line 103, in pandas.index.IndexEngine.get_value (pandas\index.c:3332)
File "pandas\index.pyx", line 111, in pandas.index.IndexEngine.get_value (pandas\index.c:3035)
File "pandas\index.pyx", line 154, in pandas.index.IndexEngine.get_loc (pandas\index.c:3955)
File "pandas\index.pyx", line 169, in pandas.index.IndexEngine._get_loc_duplicates (pandas\index.c:4236)
TypeError: unorderable types: str() > int()

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "", line 1, in
File "C:\Users\Alexander\Anaconda3\lib\site-packages\pandas\core\frame.py", line 4061, in apply
return self._apply_standard(f, axis, reduce=reduce)
File "C:\Users\Alexander\Anaconda3\lib\site-packages\pandas\core\frame.py", line 4157, in _apply_standard
results[i] = func(v)
File "", line 1, in
File "C:\Users\Alexander\Anaconda3\lib\site-packages\pandas\core\series.py", line 583, in getitem
result = self.index.get_value(self, key)
File "C:\Users\Alexander\Anaconda3\lib\site-packages\pandas\indexes\base.py", line 2000, in get_value
raise IndexError(key)
IndexError: (0, 'occurred at index 0')

This is the error I get when running on Python 3.5.1/Pandas 0.18.1 on Windows. I get the expected output on Linux running Python 2.7.11/Pandas 0.18.1.

[alexanderwhatley]

Just as an update - it happens to work on Python 2.7.11/Pandas 0.18.1 on Windows as well.

[jreback]

The root example is here:

This works in py2

In [6]: s = Series([0,0],['a','a'])

In [7]: s[0]
Out[7]: 0

In [8]: s[1]
Out[8]: 0

We explicity catch the unordered exception and raise an error break out of the indexing code, but not sure what is happening after. This should work.

[jreback]

What you are writing is very fragile code anyhow, using getitem depends on the index, which in this case doesn't work (as they are strings), using .iloc makes this explicit. Of course you already know that you shouldn't be using .apply in this case (or in reality MOST cases). There is really no need.

In [1]: df = pd.DataFrame({"a" : [0,1,2], "b" : [1,2,3]})

In [2]: df[['a','a']].apply(lambda x: x.iloc[0] + x.iloc[1], axis=1)
Out[2]: 
0    0
1    2
2    4
dtype: int64

[topper-123]

Removing the 2/3 Compat label, as we've dropped Python2 support and the example still fails on master.

[simonjayhawkins]

this is now fixed on master

>>> import pandas as pd
>>>
>>> pd.__version__
'1.1.0.dev0+1045.g30724b9c6'
>>>
>>> df = pd.DataFrame({"a": [0, 1, 2], "b": [1, 2, 3]})
>>>
>>> df[["a", "a"]].apply(lambda x: x[0] + x[1], axis=1)
0    0
1    2
2    4
dtype: int64

[simonjayhawkins]

this is now fixed on master

fixed in #29700

7eb0db3 is the first new commit
commit 7eb0db3
Author: jbrockmendel [email protected]
Date: Mon Nov 25 15:45:27 2019 -0800

BUG: Index.get_loc raising incorrect error, closes #29189 (#29700)