Possibly spurious GroupBy failure on Travis

[cpcloud]

Can't really investigate right now

https://travis-ci.org/pydata/pandas/jobs/19331373

[jreback]

I have seen this come up occasionally
hard to repro - not sure if it's an actual bug - maybe in the test data generation

[cpcloud]

ok i can look at it tonight

[jreback]

the test frame creation is subject (somewhat) to some inconcistency, if I run test_groupby alone it works fine, but in a certain chain it creates bigger test frames, not really sure why. so make sure that the input is consistent across runs...