vincentschen commented 4 years ago

Description of proposed changes

Update dev + setup.py requirements for torch
Use native torch.utils.tensorboard and remove tensorboardX dep
Moved lr_scheduler.step() after optimizer.step()
Replace deprecated uses of .byte() with .bool()

Related issue(s)

Fixes https://github.com/snorkel-team/snorkel/issues/1558

Test plan

All unit tests pass with torch==1.2.0, 1.3.0, 1.4.0
Updated Slicing convergence test for longer fit.
tox -e {complex,spark} passes with torch==1.2.0, 1.3.0, 1.4.0

Checklist

Need help on these? Just ask!

[x] I have read the CONTRIBUTING document.
[x] I have updated the documentation accordingly.
[x] I have added tests to cover my changes.
[x] I have run tox -e complex and/or tox -e spark if appropriate.
[x] All new and existing tests passed.

codecov[bot] commented 4 years ago

Codecov Report

Merging #1565 into master will not change coverage by %. The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1565   +/-   ##
=======================================
  Coverage   97.13%   97.13%           
=======================================
  Files          56       56           
  Lines        2091     2091           
  Branches      342      342           
=======================================
  Hits         2031     2031           
  Misses         31       31           
  Partials       29       29

Impacted Files	Coverage Δ
...ssification/training/loggers/tensorboard_writer.py	`100.00% <100.00%> (ø)`
snorkel/classification/training/trainer.py	`89.83% <100.00%> (ø)`
snorkel/labeling/model/label_model.py	`95.54% <100.00%> (ø)`
snorkel/slicing/utils.py	`94.66% <100.00%> (ø)`

henryre commented 4 years ago

@vincentschen can we update the Test Plan with details from the multiple version tests?