BUG: ndim and shape for MultiIndex are wrong

[jtratner]

>>> from pandas import MultiIndex
>>> mi = MultiIndex.from_tuples(list(zip(range(10), range(10))))
>>> mi.shape
(0,)
>>> mi.ndim
1

I'd expect ndim to return 2 and shape to be (10, 2).

[cpcloud]

len is also incorrect :)

[jtratner]

@cpcloud it should be just shape[0] right?

[cpcloud]

yep or size doesn't really matter (size is about 100 ns faster)

[jtratner]

@cpcloud well, neither of them will be fast for MI, since they have to deal with labels. I'll pull up a pr

[jtratner]

remember that mi underlying is just an empty ndarray, since everything is just levels. Though I actually think that we could change MI's internal ndarray to be self.levels instead, since it will always be integers...

[jtratner]

sorry, I mean self.labels.

[cpcloud]

wait how are you thinking about this? i was just thinking mi.values.size...are you thinking something else?

[jtratner]

@cpcloud values is pretty slow though... that's a good interim though...

[jtratner]

side note here, does anyone actually use shape on MI? Is it necessary to define? Rather not support it if we don't have to. Thoughts? @jreback @cpcloud

[cpcloud]

i've used it in the past like mi.values.shape ... but that usage depends on the representation of MI which is not really so great (my dependence, not the representation)....how would define them ?

[jtratner]

For 0.14 b/c defining these causes segfaults when it subclasses ndarray.

[jtratner]

Update - not going to be fixed, not actually a bug (and not used anywhere anyways). Those attributes have always been meaningless, so not really necessary to define them.