BUG: Various inconsistencies in DataFrame __getitem__ and __setitem__ behavior

[stephenwlin]

setitem fails with row slices supported by getitem

In [27]: df=p.DataFrame(
   ....:              {"alpha": ["a", "b", "c", "d", "e"],
   ....:              "beta":[1,2,3,4,5]},
   ....:              columns=["alpha","beta"],
   ....:              index=[1.0, 2.5, 3, 4.5, 5.0])

In [28]: df[0:1] # does work, slices on rows
Out[28]: 
  alpha  beta
1     a     1

In [29]: df[0:1] = 0 # fails
(exception)

setitem fails on boolean indexing unless key is ndarray, while getitem casts appropriately

In [49]: df=p.DataFrame(
   ....:              {"alpha": ["a", "b", "c", "d", "e"],
   ....:              "beta":[1,2,3,4,5]},
   ....:              columns=["alpha","beta"],
   ....:              index=[1.0, 2.5, 3, 4.5, 5.0])

In [50]: df[np.array([True, False, False, False, True])]
Out[50]: 
  alpha  beta
1     a     1
5     e     5

In [51]: df[[True, False, False, False, True]]
Out[51]: 
  alpha  beta
1     a     1
5     e     5

In [52]: df[np.array([True, False, False, False, True])] = 4 # does work

In [53]: df[[True, False, False, False, True]] = 3 # does not work
(exception)

getitem does not reindex boolean Series key but setitem does

In [54]: df=p.DataFrame(
   ....:              {"alpha": ["a", "b", "c", "d", "e"],
   ....:              "beta":[1,2,3,4,5]},
   ....:              columns=["alpha","beta"],
   ....:              index=[1.0, 2.5, 3, 4.5, 5.0])

In [55]: s=p.Series([True, True, False, False, False],
   ....:              index=[5.0, 4.5, 3, 2.5, 1.0]) # row-reversed index

In [56]: df[s] # gets first two rows (no reindex)
Out[56]: 
    alpha  beta
1.0     a     1
2.5     b     2

In [57]: df[s] = 1 # assigns last two rows (reindexed)

In [58]: df
Out[58]: 
    alpha  beta
1.0     a     1
2.5     b     2
3.0     c     3
4.5     1     1
5.0     1     1

getitem does not allow DataFrame key when columns are MultiIndex, but setitem does

In [60]: c2=p.MultiIndex(levels=[['foo', 'bar', 'baz', 'qux'],
   ....:                        ['one', 'two', 'three']],
   ....:                labels=[[0, 0, 0, 1, 1, 2, 2, 3, 3, 3],
   ....:                        [0, 1, 2, 0, 1, 1, 2, 0, 1, 2]],
   ....:                names=['first', 'second'])

In [61]: df2=p.DataFrame(np.random.randn(3, 10),
   ....:              index=['A','B','C'], columns=c2)

In [62]: df2[df2 > 0] # fails
(exception)

In [63]: df2[df2 > 0] = 4 # ok

setitem and getitem reindex DataFrame keys when shapes differ, but do not reindex when shapes are the same and indexes/columns are different

In [68]: df3=p.DataFrame(np.random.randn(5, 2),
   ....:              index=[1.0, 2.5, 3, 4.5, 5.0])

In [69]: key = (df3 > 0).reindex(df3.index[:-1]) # first four rows

In [70]: df3[key] = 5 # key is reindexed because shape does not match (ok)

In [71]: df3
Out[71]: 
            0         1
1.0 -1.599142  5.000000
2.5  5.000000 -0.180330
3.0 -0.568206  5.000000
4.5 -0.340465 -0.070105
5.0 -0.474955  1.928842

In [72]: key = key.reindex(key.index[::-1]) # reverse rows in key

In [73]: df3[key] = 4 # key is reindexed because shape does not match (ok)

In [74]: df3
Out[74]: 
            0         1
1.0 -1.599142  4.000000
2.5  4.000000 -0.180330
3.0 -0.568206  4.000000
4.5 -0.340465 -0.070105
5.0 -0.474955  1.928842

In [75]: df3=p.DataFrame(np.random.randn(5, 2),
   ....:              index=[1.0, 2.5, 3, 4.5, 5.0])

In [76]: key = (df3 > 0) # all rows

In [77]: df3[key] = 5 # no reindex required

In [78]: df3
Out[78]: 
            0         1
1.0  5.000000 -0.233779
2.5  5.000000 -0.141962
3.0 -0.232551  5.000000
4.5 -1.663034  5.000000
5.0 -0.653200 -0.365681

In [79]: key = key.reindex(key.index[::-1]) # reverse rows in key

In [80]: df3[key] = 4 # key is not reindexed, even though rows are reversed

In [81]: df3
Out[81]: 
            0         1
1.0  5.000000 -0.233779
2.5  5.000000  4.000000
3.0 -0.232551  4.000000
4.5  4.000000  5.000000
5.0  4.000000 -0.365681