FIX raise when groupby selecting cols not in frame

[hayd]

KeyError part of #5264

[jreback]

rebase on master .....setuptools had changed so build was failing

the 2.6 failure is real though...

[hayd]

lol to the failure, similar to me just claiming was an issue in 2.6. Silly.

[hayd]

@jreback Thinking about this, do you know if there is a helper function in loc already that might be better than this hack?

[jreback]

maybe slightly more intuitive, but yours pretty self exaplanatory

diff --git a/pandas/core/groupby.py b/pandas/core/groupby.py
index f0ee01d..fce8947 100644
--- a/pandas/core/groupby.py
+++ b/pandas/core/groupby.py
@@ -2579,8 +2579,8 @@ class DataFrameGroupBy(NDFrameGroupBy):
             raise Exception('Column(s) %s already selected' % self._selection)

         if isinstance(key, (list, tuple, Series, np.ndarray)):
-            if len(self.obj.columns.intersection(key)) != len(key):
-                bad_keys = list(set(key).difference(self.obj.columns))
+            if self.obj.columns.isin(key).sum() != len(key):
+                bad_keys = Index(key)-self.obj.columns
                 raise KeyError("Columns not found: %s"
                                % str(bad_keys)[1:-1])
             return DataFrameGroupBy(self.obj, self.grouper, selection=key,

[hayd]

subtle improveent is that yours doesn't allow for repeats... whhich is not supported... should it be :s (rabbit hole?)

[jreback]

hmm

I suppose that should technically be allowed
so go with yours

[hayd]

No methods work at the moment (as you're reindexing by a dupe), will have a little think.

[jreback]

the reindexing could be fixed, but i think you are right, it opens a big can of worms.So pls add a check and test for dups (just raise ValueError) for now.

[hayd]

Now I think about it, not tested axis=1...at all (I'm guessing this won't work, I could just raise as NotImplemented ) :S

[jreback]

more tests on this for axis=1? release notes?

[jreback]

ping!

[jreback]

ping

[jreback]

@hayd ping....need to get this in ASAP

[jreback]

this looks fine. anything else on this? give ok and I'll merge it (I want to update the release note a bit)

[hayd]

Will have some time for this late tomorrow (and wed and thurs), sorry been swamped recently, need to pull finger out on many PRs lined up mañana.

I see point on axis=1, I'm worried machinery is different... and not played with this.

[jreback]

ping!