now it is implemented only for model, but shall we doe rank warning also for callbacks?

i think it needs to be a standalone package.

from pl import warnings

# from trainer.__init__
warnings.set_rank(rank)

# anywhere now including a lightning  module,   trainer, etc.
warnings.rank_zero_warn()

don't make it so we have to instantiate an instance... also instance doesn't make sense since we just need to call it from wherever.

Also, replace all the warnings across the project with this.

but the rank is depending on the backend so you cannot use a global rank value... can you?

Codecov Report

Merging #1428 into master will decrease coverage by 0%.
The diff coverage is 85%.

@@           Coverage Diff           @@
##           master   #1428    +/-   ##
=======================================
- Coverage      92%     92%    -0%     
=======================================
  Files          63      66     +3     
  Lines        3404    3508   +104     
=======================================
+ Hits         3123    3212    +89     
- Misses        281     296    +15     

warnings.rank_zero_warn()

but this has to have an extra parameter of the actual rank...
so we can have warning called from where ever but not with rank specific

warnings inside lightning are ALL rank specific.

Hello @Borda! Thanks for updating this PR.

• In the file pytorch_lightning/callbacks/model_checkpoint.py:

Line 206:111: E501 line too long (116 > 110 characters)

• In the file pytorch_lightning/trainer/trainer.py:

Line 957:111: E501 line too long (119 > 110 characters)

Comment last updated at 2020-04-09 17:26:48 UTC