Error using Series.subset() with NumPy 1.9

[freeman-lab]

Thunder's Series.subset() method relies on PySpark's rdd.takeSample(). Due to a recent patch to NumPy (numpy/numpy@6b1a120), takeSample is broken on NumPy 1.9 installations because it generates random seeds that frequently exceed the maximum bound of 2 ** 32.

As a result, the following example code:

data = tsc.makeExample('pca')
data.subset(10)

Will almost always produce:

ValueError: Seed must be between 0 and 4294967295

The underlying issue needs to be fixed in PySpark, but for now we can avoid the problem by explicitly specifying a seed in the correct range.