IPython Parallel map mysteriously turns pandas Series into numpy ndarray

[samuela]

If you run

from IPython.parallel import Client
c = Client()
v = c[:]

with v.sync_imports():
    from pandas import Series
    import numpy as np

poop = v.map(lambda x: Series({'a': x}), range(5))
[type(x) for x in poop]

The result is

[numpy.ndarray,
 numpy.ndarray,
 numpy.ndarray,
 numpy.ndarray,
 numpy.ndarray]

Whereas running

poop = [Series({'a': x}) for x in range(5)]
[type(x) for x in poop]

returns

[pandas.core.series.Series,
 pandas.core.series.Series,
 pandas.core.series.Series,
 pandas.core.series.Series,
 pandas.core.series.Series]

as expected.

[minrk]

Yes, a known bug due to an inappropriate use of isinstance, since Series is a subclass of ndarray. This is fixed a while back in master, when I rewrote the parallel serialization.

[samuela]

Great, thanks!