RaggedArray.__getitem__ fails on 2d lists

[justinrporter]

In numpy, it is possible to do the following:

>>> a = np.array(np.arange(9)).reshape(3, 3)
>>> a[list(zip( (1, 2), (2, 0), (0, 0) ))]

array([5, 6, 0])

Specifically, ndarray.__getitem__ handles the case where the requested indices are 2-element list, similar to the output of np.where, but a list rather than a tuple.

Similar code on a RaggedArray:

a = ra.RaggedArray([np.arange(3), np.arange(4), np.arange(2)])
a[list(zip( (1, 2), (2, 0), (0, 0) ))]

fails with

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-108-d8376781920d> in <module>()
      1 a = ra.RaggedArray([np.arange(3), np.arange(4), np.arange(2)])
----> 2 a[list(zip( (1, 2), (2, 0), (0, 0) ))]

/home/jrporter/modules/enspara/util/array.py in __getitem__(self, iis)
    455         elif (type(iis) is slice) or (type(iis) is list) \
    456                 or (type(iis) is np.ndarray):
--> 457             return RaggedArray(self._array[iis])
    458         # tuples get index conversion from 2d to 1d
    459         elif type(iis) is tuple:

IndexError: too many indices for array

The very-similar following succeeds:

>>> a = ra.RaggedArray([np.arange(3), np.arange(4), np.arange(2)])
>>> a[tuple(zip( (1, 2), (2, 0), (0, 0) ))]

array([2, 0, 0])

IIRC this tuple/list issue is one of the things that was gave me a lot of trouble when I wrote the __getitem__ we wound up not using. (I don't remember if I eventually figured it out or not.)

Anyway, it looks like the __getitem__ we currently have implemented attempts to use the type of the indexer ("iis") as an proxy for the meaning of the contained data. I would suggest using the length of the data instead. As a probably-flawed example of what I'm suggesting, if the list only has two elements of matching dimensions, then it's an np.where-result-like indexer. If it instead is a list of many elements of one dimension, then it can be interpreted as a list of indices in the appropriate dimension.