search

USGS-R / river-dl

Deep learning model for predicting environmental variables on river systems
Creative Commons Zero v1.0 Universal
21 stars 15 forks source link

reproducible results with random seed #39

Closed jsadler2 closed 4 years ago

jsadler2 commented 4 years ago

I've been setting a random seed in the training, but I am still getting different results. I'd like to modify the code to be able to produce completely reproducible results.

jsadler2 commented 4 years ago 
[jsadler][IGSARMEWLTJMS] $ sha256sum trained_weights_seed_245a/.data-00000-of-00001
292e1382afc069fc79867122ce3df548ec1117da4bb55b24c28cb986174061cb *trained_weights_seed_245a/.data-00000-of-00001

[/d/onedrive/OneDrive - DOI/research/drb_ml_model/data/out] (master)
[jsadler][IGSARMEWLTJMS] $ sha256sum trained_weights_seed_245b/.data-00000-of-00001
292e1382afc069fc79867122ce3df548ec1117da4bb55b24c28cb986174061cb *trained_weights_seed_245b/.data-00000-of-00001

[/d/onedrive/OneDrive - DOI/research/drb_ml_model/data/out] (master)
[jsadler][IGSARMEWLTJMS] $ sha256sum trained_weights_no_seed/.data-00000-of-00001
a9f57afdda33f8aeb24beb4d8f9ece369ab5a3a7594952d9659eb75b36dd4fef *trained_weights_no_seed/.data-00000-of-00001