Closed itsb closed 7 years ago
I'm wondering if a better solution would be to set use cmt.utils.seed
directly in c2s-preprocess.py, and while we're at it set numpy.random.seed
as well.
The same approach can then be used for c2s-train
, we don't have to change the interface of every function in c2s.py, and we don't have to make sure the seed is passed to every c2s function call.
agree. this is cleaner. preprocess is working as before. assuming train also works.
Thanks!
Ability to set random seed is critical for debugging and reproducibility. See corresponding PR in CMT repo.