Series.loc/iloc[x, y] does not raise exception

[pkch]

Code Sample, a copy-pastable example if possible

df = pd.DataFrame({'a': [10]})
# df['a'] is a series, can be indexed with 1 index only

# will raise IndexingError, as expected
df['a'].iloc[0, 0]
df['a'].loc[0, 0]

# will raise nothing, not as expected
df['a'].iloc[0, 0] = 1000 # equivalent to pass
df['a'].loc[0, 0] = 1000 # equivalent to df['a'].loc[0] = 1000

See this SO post

Expected Output

Expected an IndexingError exception in every case where two values are used to index a Series.

output of pd.show_versions()

pandas version 0.18.1, python 3.5

[jreback]

this is a bug for a non-MultiIndex. (for .loc). .iloc should raise as well.

BUT this IS valid for a MI.

In [10]: s = Series([1,2,3], index=pd.MultiIndex.from_tuples([(0,'a'),(0,'b'),(1,'a')]))

In [11]: s
Out[11]: 
0  a    1
   b    2
1  a    3
dtype: int64

In [12]: s.loc[0,'d'] = 4

In [13]: s
Out[13]: 
0  a    1
   b    2
1  a    3
0  d    4
dtype: int64

[jreback]

pull requests are welcome, keep in mind that there are hundreds of existing test cases that still must pass, so careful changing things.

Further note that you ARE chained indexing, and SETTING, which in this particular case is ok (as its a single dtype), but in general is NOT recommended, nor can you always detect it.

[jreback]

This does raise for a non-integer index already (for .loc)

In [22]: s = Series([1,2,3], index=list('abc'))

In [23]: s.loc[0,1] = 5
KeyError: '[0 1] not in index'

so this could use some more comprehensive testing.

[simonjayhawkins]

the iloc setitem case now raises as expected. on 0.25.3 gave IndexingError: Too many indexers same as the getitem cases. but on master now gives IndexError: too many indices for array

the loc setitem still does not raise.

[DriesSchaumont]

take