broadinstitute/ml4h

seed tensorflow random number generator

Closed this issue · 1 comments

What
currently numpy random is seeded with an arg but tensorflow is not. we should seed tensorflow as well

Why
it's important to be able to reproduce results of an experiment, to that end, numpy is seeded but TF is not, lets seed both

How
https://stackoverflow.com/questions/38469632/tensorflow-non-repeatable-results may be more involved but attempt to just tf.set_seed()

Acceptance Criteria
reproducible results (after removing nondeterminism from parallelism)

didn't improve much during testing