Closed tztsai closed 1 year ago
Thanks for this! Right way forwards on the seeding issue. seed_all
doesn't currently get used -- in @CemGultekin1 's fork, torch.random_seed
is called early on in trainScript.py
:
We probably want to place a call to seed_all
around there. Otherwise, I'm happy with the shotgun approach of trying to seed as much as possible in case it crops up in any deeper calls.
I commented on #63 regarding the model final transformation issue. If you'd make a new branch just for the random seed changes, I'd gladly merge it in by itself (up to you).
Thanks! I'll make a new branch for the random seeding.
Both changes were split out into smaller PRs -- closing.
To address issue #44, I added a
random_state
argument to thesample
function in src/gz21_ocean_momentum/analysis/utils.py, and aseed_all
function in src/gz21_ocean_momentum/utils.py. To address issue #63, I movednet._final_transformation = lambda x: x
from the test file tests/models/test_fully_conv_net.py into the__init__
method ofFullyCNN
.