Nan rows for filtered out data using bcolz backend

[tdennist]

This quirk was reported in this thread: https://groups.google.com/a/continuum.io/forum/#!topic/blaze-dev/BqJx4LAy0xE

In [28]: blaze.__version__
'0.8.0'

In [29]: bcolz.__version__
'0.8.1'

The bcolz backend to Blaze seems to display a nan entry for the filtered out rows:

In [22]: df = blaze.Data(bcolz.ctable([[1,1,3,3], [1,2,3,4]]))

In [23]: df = df[df['f1']==3]

In [24]: blaze.by(df[['f0']], Sum=df['f1'].sum())

   f0  Sum
0   1  NaN
1   3    3


The Pandas backend behaves as I would expect:

In [25]: df = blaze.Data(pandas.DataFrame(dict(f0=[1,1,3,3], f1=[1,2,3,4])))

In [26]: df = df[df['f1']==3]

In [27]: blaze.by(df[['f0']], Sum=df['f1'].sum())

   f0  Sum
0   3    3