[pandas] groupby reversed

[Neal Becker]

The order produced by this groupby is the reverse of what I wanted. Since I can
do

for case,res in stuff:
...

I thought then stuff is a sequence, and reversed should work. But it doesn't

df = create_df (results)
stuff = df.groupby (['esno','order'])

for case, res in reversed (stuff):

for case, res in reversed (stuff):
File "/home/nbecker/.local/lib/python2.7/site-
packages/pandas/core/groupby.py", line 1871, in __getitem__
raise KeyError(str(key))
KeyError: '5'

2 questions:

1. Why didn't this work?

2. What's a better way?

[Wes McKinney]

I don't think GroupBy fully implements Python's iterator protocol to
work with reverse. You could do

for case, res in reversed(list(stuff)):
...

- Wes

[Neal Becker]

Can I convince groupby to sort differently? The doc seemed a bit unclear.

[Wes McKinney]

The sort=True/False option only controls whether each group key is
sorted or whether to use the observed order in the data. For example:

In [1]: s = Series(['d', 'c', 'd', 'c', 'b', 'a'])

In [2]: s.groupby(s).size()
Out[2]:
a 1
b 1
c 2
d 2

In [3]: s.groupby(s, sort=False).size()
Out[3]:
d 2
c 2
b 1
a 1


The keys are only descending in the last case because they first
appear in descending order in the data. I'm not opposed to different
iteration order options, but it would have to be "convince via pull
request".

- Wes