DataFrame.unstack() issues DeprecationWarning with MultiIndex

[xflr6]

>>> import warnings
>>> warnings.simplefilter('error')

>>> import pandas as pd
>>> pd.__version__
u'0.17.0rc1'

>>> df = pd.DataFrame({'A': list('aabb'), 'B': [0, 1, 0, 1], 'C': range(4)})
>>> gf = df.groupby(['A', 'B']).sum()
>>> gf
     C
A B   
a 0  0
  1  1
b 0  2
  1  3

>>> gf.unstack()
Traceback (most recent call last):
..., line 4274, in _reference_duplicate_name
    return np.sum(name == np.asarray(self.names)) > 1
DeprecationWarning: elementwise == comparison failed; this will raise an error in the future.

[jreback]

what version numpy?

there are 'lots' of these. We do have to fix this at some point. But its not a big deal.

I can't actually repro your result.

[jreback]

this is covered in #8537

pull-requests always welcome!

[xflr6]

Yes, sorry, I just realized this seems to be due to this in the numpy 1.10.0b1 I am using here.

The line in the traceback seems to involve a comparison like: -1 == np.asarray(['A', 'B'])

[jreback]

We actually test with this: https://travis-ci.org/pydata/pandas/jobs/81396483
numpy master with DeprecationWarnings turned on. I try to get rid of them; I think I got most but their are some (also the 3.4 build has some as well).

so eventually have to fix!

[rosnfeld]

Yup, I just hit this one today too when I did a fresh pull of the latest pandas and numpy.
Another repro case:

pd.DataFrame([[1, 2], [3, 4]], columns=['a', 'b']).groupby('a').b.value_counts().unstack()

[jreback]

yeh this is a numpy warning - never got around to tracking this one down

you are welcome to it!

[xflr6]

Maybe this should be re-opened (does #8537 mention mixed-type comparisons?).

I hope, we already tracked it: Wouldn't it be enough to replace this

return np.sum(name == np.asarray(self.names)) > 1

with this?

return sum(name == n for n in self.names) > 1