Ensure Randomness and reproducibility

[apmcleod]

Some types of randomness aren't guaranteed to be OS independent.

Also, we may want to use a numpy RandomState rather than seeds, but probably not.

[apmcleod]

Given that we will release the official ACME as a zipfile (or something), I don't think this is necessarily important. If people have slightly different random seeds, it's probably not a big deal.

[JamesOwers]

It's a shame not to be able to ensure full reproducibility given some set of parameters for the make_dataset.py script...but I don't think it's priority for the reason you give. Let's kick this for now.

[apmcleod]

Agreed. I'll remove the milestone.