Add random_state option to benchmark function

[utf]

Would be nice to add support for the random_sample variable of pandas.DataFrame.sample() for pipeline benchmarking.

E.g. implemented for this line:

testdf, traindf = np.split(df.sample(frac=1), [int(test_spec * len(df))])

Would allow for random but deterministic sampling of the dataframe when choosing the test/train split. This way you can benchmark two models on the same dataframe and automatically have the same test/train split. Can just override the benchmark test_spec variable again (e.g. if test_spec is an int or numpy.random.RandomState object then use it as the random_sample variable).

[ardunn]

sounds good to me!

[ardunn]

@utf current benchmarking implementation has you pass in a sklearn kfold (or StratifiedKFold) to the benchmarking function. the kfold object can accept a random state param, so this issue is essentially closed.

[utf]

Sounds great