sort=False ignored in Series groupby on MultiIndex levels

[cottrell]

Example:

import pandas
def geta():
    i = pandas.MultiIndex.from_tuples([(1, 2, 'a', 0),
                              (1, 2, 'a', 1),
                              (1, 1, 'b', 0),
                              (1, 1, 'b', 1),
                              (2, 1, 'b', 0),
                              (2, 1, 'b', 1)], names=['a', 'b', 'c', 'd'])
    a = pandas.Series([0, 1, 2, 3, 4, 5], index=i)
    return(a)

for dosort in [True, False]:
    a = geta()
    b = a.groupby(level=['a', 'b'], sort=dosort).first()
    a = None
    print('%s, sort=%s, \n%s' % (geta.__name__, dosort, b))

Output:

geta, sort=True,
a  b
1  1    2
   2    0
2  1    4
dtype: int64
geta, sort=False,
a  b
1  1    2
   2    0
2  1    4
dtype: int64

Also see: #9076
Possibly related to: #8868

[cottrell]

This seems to fix it. There might be a few more sort args missing elsewhere, for example in groupby._indexer_from_factorized.

diff --git a/pandas/core/groupby.py b/pandas/core/groupby.py
index cb5dedc..06fbb55 100644
--- a/pandas/core/groupby.py
+++ b/pandas/core/groupby.py
@@ -1378,7 +1378,7 @@ class BaseGrouper(object):
         else:
             if len(all_labels) > 1:
                 group_index = get_group_index(all_labels, self.shape)
-                comp_ids, obs_group_ids = _compress_group_index(group_index)
+                comp_ids, obs_group_ids = _compress_group_index(group_index, sort=self.sort)
             else:
                 ping = self.groupings[0]
                 comp_ids = ping.labels

Output after patch:

geta, sort=False,
a  b
geta, sort=True,
a  b
1  1    2
   2    0
2  1    4
dtype: int64
geta, sort=False,
a  b
1  2    0
   1    2
2  1    4
dtype: int64

[jreback]

try on master - iIIRC this all is fixed

[cottrell]

Just tried on master at the commit below and I am seeing no difference.

commit 107cb10
Merge: c4a996a f9031c2
Author: Joris Van den Bossche [email protected]
Date: Sun Feb 8 14:35:47 2015 +0100

Merge pull request #9417 from cel4/plot_error

improved error message for invalid chart types

[jreback]

@cottrell can you do a separate PR for this (with a test), (and a release note). You can then rebase the to_coo (#9076) on top. thxs.

[jreback]

this might interact with #9445