grouped.filter with dropna option

[leo4183]

in Pandas 0.16.2:

grouped = a.groupby(idx)
grouped.filter(lambda x: x.mean()>0.0, dropna=False)

in case there was no group whose x.mean() was larger than zero, an AttributeError: 'list' object has no attribute 'astype' would be raised (rather than returning an all NaN dataframe/serie in previous versions).

[jreback]

pls show a complete example that can be copy pasted

[leo4183]

a concrete example:

import numpy as np
import pandas as pd

row = 9
a   = pd.Series(np.random.rand(row),index=np.repeat([1,2,3],3))
gp  = a.groupby(level=0)
empty_serie  = gp.filter(lambda x: x.mean()>1)
error_raised = gp.filter(lambda x: x.mean()>1, dropna=False)

where the empty_serie would be an empty serie as expected. however an attribute error would be raised for the last line

[jreback]

hmm, that's not very friendly!. pull-requests to fix welcome!