ERR: validate boolean indexing with iloc

[jreback]

xref #8956

This should raise a ValueError.

In [18]: s = Series(np.arange(5,1,-1))

In [19]: s
Out[19]: 
0    5
1    4
2    3
3    2
dtype: int64

In [20]: s.iloc[[True,False,True]]
Out[20]: 
0    5
2    3
dtype: int64

[jorisvandenbossche]

As I said in the other issue, numpy has the same behaviour as pandas (which is strange), although there docs say the same as we think it should be: "Boolean arrays must be of the same shape as the initial dimensions of the array being indexed" (http://docs.scipy.org/doc/numpy/user/basics.indexing.html#boolean-or-mask-index-arrays):

In [2]: np.__version__
Out[2]: '1.8.1'

In [3]: a = np.arange(5)

In [4]: a
Out[4]: array([0, 1, 2, 3, 4])

In [5]: a[np.array([True, False, True])]
Out[5]: array([0, 2])

[jorisvandenbossche]

In the meantime, numpy has changed this behaviour (already deprecated in 1.10, errors in 1.13), so we should follow I think. This should raise:

In [1]: np.__version__
Out[1]: '1.13.3'

In [2]: pd.__version__
Out[2]: '0.22.0.dev0+202.g492040b'

In [3]: a = np.arange(5)

In [4]: a[np.array([True, False, True])]
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-4-7375c598447e> in <module>()
----> 1 a[np.array([True, False, True])]

IndexError: boolean index did not match indexed array along dimension 0; dimension is 5 but corresponding boolean dimension is 3

In [5]: s = pd.Series(a)

In [6]: s.iloc[np.array([True, False, True])]
Out[6]: 
0    0
2    2
dtype: int64

[jreback]

yep agreed. I think they wrapped around the booleans before.