MultiIndex loc for value not in index

[hayd]

example:

In [11]: s = pd.Series({(0, 0): 1, (0, 1): 2, (0, 3): 3, (1, 0): 1, (1, 2): 4, (3, 0): 5})

In [12]: s.loc[(1, 0):]
Out[12]:
(1, 0)    1
(1, 2)    4
(3, 0)    5
dtype: int64

In [13]: s.loc[(1, 1):]
KeyError: 'start bound [(1, 1)] is not the [index]'

Should this work? cc @jreback

From SO question.

[jreback]

This looks legit, you are asking for a starting tuple with no stopping tuple, so you get the rest of the index
and it correctly start/stops including the endpoints (if provided)

In [4]: s.loc[(1,0):]
Out[4]: 
(1, 0)    1
(1, 2)    4
(3, 0)    5
dtype: int64

In [5]: s
Out[5]: 
(0, 0)    1
(0, 1)    2
(0, 3)    3
(1, 0)    1
(1, 2)    4
(3, 0)    5
dtype: int64

In [6]: s.loc[(1,0):(1,2)]
Out[6]: 
(1, 0)    1
(1, 2)    4
dtype: int64

In [7]: s.loc[(0,3):(1,2)]
Out[7]: 
(0, 3)    3
(1, 0)    1
(1, 2)    4
dtype: int64

[hayd]

@jreback but don't we expect:

In [13]: s.loc[(1, 1):]
Out[13]:
(1, 2)    4
(3, 0)    5
dtype: int64

rather than KeyError?

[jreback]

nope....that's according to spec, then endpoints must be included

I get that this since there is a defined ordering that this should work (and might in cases of datetimes), where lookup is 'approximate', e.g. s['2000'] would start at the first datetime of 2000 (even if its much later)

but s.loc['2000'] would not

I think OP is best by doing this using a multi-index, a list of tuples is essentially (but not exactly the same)
and in this case, I suppose you could argue its a bug....

In [15]: s2 = s.copy()

In [16]: s2.index = pd.MultiIndex.from_tuples(s.index)

In [17]: s2
Out[17]: 
0  0    1
   1    2
   3    3
1  0    1
   2    4
3  0    5
dtype: int64

In [18]: s2[(1,0):]
Out[18]: 
1  0    1
   2    4
3  0    5
dtype: int64

In [19]: s2[(1,1):]
Out[19]: 
1  2    4
3  0    5
dtype: int64

[hayd]

Ha! So the error is that this totally isn't a MultiIndex. Sorry

[hayd]

is it weird that using loc this gives an error?

In [92]: s[:(1,1)]
Out[92]: 
0  0    1
   1    2
   3    3
1  0    1
dtype: int64

In [93]: s.loc[:(1,1)]
KeyError: 'stop bound [(1, 1)] is not in the [index]'

[jreback]

that's correct
I you specify an endpoint in a slice it must be included in the index

[hayd]

not sure I follow why this works without loc but not with loc...

[jreback]

loc is very strict about this, while I think getitem just searches. I think that's the intent, though maybe these should be the same