BUG: loc.__setitem__ over MultiIndex with slice key and step != 1 does strange things

[immerrr]

Here's how it looks like:

In [11]: s = pd.Series(np.arange(10), pd.MultiIndex.from_product(([0, 1], list('abcde'))))

In [12]: s
Out[12]: 
0  a    0
0  b    1
0  c    2
0  d    3
0  e    4
1  a    5
1  b    6
1  c    7
1  d    8
1  e    9
dtype: int64

In [13]: s.iloc[::4] = 100; s
Out[13]: 
0  a    100
0  b    100
0  c    100
0  d    100
0  e    100
1  a      5
1  b      6
1  c      7
1  d      8
1  e      9
dtype: int64

In [14]: s.loc[::4] = 200; s
Out[14]: 
0  a    200
0  b    200
0  c    200
0  d    200
0  e    200
1  a      5
1  b      6
1  c      7
1  d      8
1  e      9
dtype: int64

In [15]: s.ix[::4] = 300; s
Out[15]: 
0  a    300
0  b    300
0  c    300
0  d    300
0  e    300
1  a      5
1  b      6
1  c      7
1  d      8
1  e      9
dtype: int64

Regular setitem works fine:

In [16]: s[::4] = 10000; s
Out[16]: 
0  a    10000
0  b      300
0  c      300
0  d      300
0  e    10000
1  a        5
1  b        6
1  c        7
1  d    10000
1  e        9
dtype: int64

I have some ideas about iloc refactoring that will fix that for iloc, but loc/ix will still be affected.

[jreback]

This was prob never tested properly. That said for loc/ix. You could step 'logically', meaning you step the 'levels', but not sure if that is useful.

[jreback]

@immerrr able to get to this soon? otherwise I will bump to next version

[immerrr]

My schedule is flaky for ~2 more weeks, sorry.

[jreback]

np...take your time

[wesm]

I just looked at this. iloc behaves properly, but loc does not

In [8]: s
Out[8]: 
0  a    0
   b    1
   c    2
   d    3
   e    4
1  a    5
   b    6
   c    7
   d    8
   e    9
dtype: int64

In [9]: s.iloc[::4] = 100

In [10]: s
Out[10]: 
0  a    100
   b      1
   c      2
   d      3
   e    100
1  a      5
   b      6
   c      7
   d    100
   e      9
dtype: int64

In [11]: s.loc[::4] = 200

In [12]: s
Out[12]: 
0  a    200
   b    200
   c    200
   d    200
   e    200
1  a      5
   b      6
   c      7
   d    100
   e      9
dtype: int64

[simonjayhawkins]

I just looked at this. iloc behaves properly, but loc does not

changed title to reflect this and that .ix is deprecated.

[simonjayhawkins]

could add a test to confirm the .iloc behavior.

[mroeschke]

Looks like both loc and iloc work consistently now

In [10]: In [11]: s = pd.Series(np.arange(10), pd.MultiIndex.from_product(([0, 1], list('abcde'))))

In [11]: s.loc[::4] = 200

In [12]: s
Out[12]:
0  a    200
   b      1
   c      2
   d      3
   e    200
1  a      5
   b      6
   c      7
   d    200
   e      9
dtype: int64

In [13]: s.iloc[::4] = 100

In [14]: s
Out[14]:
0  a    100
   b      1
   c      2
   d      3
   e    100
1  a      5
   b      6
   c      7
   d    100
   e      9
dtype: int64

In [15]: pd.__version__
Out[15]: '1.3.0.dev0+1282.g33fec602ff'