Making categoricals comparable

[hayd]

another example:

http://stackoverflow.com/questions/23814368/sorting-pandas-categorical-labels-after-groupby

At the moment bins / buckets are simply strings, so you can't sort them properly.

A=pd.qcut(X['x'], [0,0.25,0.5,0.75,1.0])

In [28]: A[0]
Out[28]: '(0.635, 2.4]'

In [29]: type(A[0])
Out[29]: str

The dtype of the Index is object (because these are strings), maybe we could just have these as a subclass of strings where:

def __lt__(self, other):
    from ast import literal_eval
    return literal_eval(self) < literal_eval(other)

Can put together pr if seems reasonable.

[hayd]

Forgot to include motivating example for this: http://stackoverflow.com/questions/17156662/pandas-dataframe-unstack-changes-order-of-row-and-column-headers/17157110#17157110

[cpcloud]

hm why the ast? literal_eval('(1,2]') won't evaluate because that's not valid syntax. found this on ordering intervals which is a special case of partial ordering

[hayd]

oh sorry I meant literal_eval(self[1:-1])

[cpcloud]

that ordering is a partially ordered R2. does that make the most sense for intervals? it's probably the most intuitive...

[hayd]

Well, to be honest it could even just be the first element, since in this case bins are disjoint (is this the case for all categoricals?)