BUG? when incrementing .loc[]-view

[immerrr]

Am I doing something wrong with indexing here?

PS. on the second thought it looks related to #3492, but I don't have a dev environment at hand to test and it fails for some reason when trying to build pandas via pip. If that's the case and it's already fixed, my apologies and feel free to close this one.

Steps to reproduce:

>>> df = DataFrame(index=np.arange(10))
>>> df['foobar'] = False
>>> df['a'] = np.arange(10)
>>> df
  foobar  a
0  False  0
1  False  1
2  False  2
3  False  3
4  False  4
5  False  5
6  False  6
7  False  7
8  False  8
9  False  9
# This works as expected
>>> df.a[df.a % 2 == 0] += 10
>>> df
  foobar   a
0  False  10
1  False   1
2  False  12
3  False   3
4  False  14
5  False   5
6  False  16
7  False   7
8  False  18
9  False   9
# This doesn't work, but expected due to copying on bool-indexing
>>> df[df.a % 2 == 0].a += 100
>>> df
  foobar   a
0  False  10
1  False   1
2  False  12
3  False   3
4  False  14
5  False   5
6  False  16
7  False   7
8  False  18
9  False   9
# But what does this?
>>> df.loc[df.a % 2 == 0, 'a'] += 1000
>>> df
  foobar     a
0  False  1010
1  False     1
2  False  1010
3  False     3
4  False  1010
5  False     5
6  False  1010
7  False     7
8  False  1010
9  False     9

>>> pandas.__version__
'0.11.0'

[cpcloud]

don't think this is a bug, see here

[immerrr]

@cpcloud , the page you've referenced mentions two out of three cases in my snippet, which, if you followed along the comments, were expected in the first place. It is in the 3rd case that where the magic starts: it's not a copy as changes are propagated to df.a, but the += operator goes nuts and instead of increasing every odd (even in 0-based notation) element by 1000, it only increases the first one and assigns the rest to it.

[cpcloud]

@immerrr whoops sorry didn't look at your last example close enough...i see now.

[cpcloud]

@immerrr this is fixed in git master

[jreback]

@immerrr, @cpcloud is right, this was fixed in master, #3533, introduced in 0.11.0, but
fixed in 0.11.1 (upcoming)

In [1]: df = DataFrame(index=np.arange(10))

In [2]: df['foobar'] = False

In [3]: df['a'] = np.arange(10)

In [4]: 

In [4]: df
Out[4]: 
  foobar  a
0  False  0
1  False  1
2  False  2
3  False  3
4  False  4
5  False  5
6  False  6
7  False  7
8  False  8
9  False  9

In [5]: df.loc[df.a % 2 == 0, 'a'] += 1000

In [6]: df
Out[6]: 
  foobar     a
0  False  1000
1  False     1
2  False  1002
3  False     3
4  False  1004
5  False     5
6  False  1006
7  False     7
8  False  1008

[immerrr]

@cpcloud @jreback
Awesome, thanks, looking forward to see it in pypi index. Closing the issue.