Change dangerous default random seed selection #18846
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
awaelchli
requested review from
carmocca,
justusschock and
williamFalcon
as code owners
carmocca
approved these changes
Oct 24, 2023
mergify
bot
added
ready
|
@carmocca I added this to the changed section already. Is there anything else I should do to communicate this change? A warning that the feature changed would be possible to add but then we are again offending all these users who don't want to see false-positive warnings. |
|
Nothing I can think of. I just wanted to point it out |
tsenst
pushed a commit
to tsenst/lightning
that referenced
this pull request
Oct 26, 2023
Borda
pushed a commit
that referenced
this pull request
Nov 2, 2023
(cherry picked from commit 6bfde6a)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
The
seed_everythingfunction can be called without arguments. In this case, it will take the seed value from the environment variablePL_GLOBAL_SEED. But if the environment variable is not set, it falls back to selecting the seed randomly. This isn't usually a problem if Lightning is in control of launching the processes, because child processes will inherit the seed value from the environment variable. However, in cases where processes are launched externally (e.g. SLURM, torchelastic etc.), this default behavior is dangerous because each process will independently choose a random seed. This can affect sampling, randomized validation splits, and other behaviors that rely on each process having the same seed.This PR removes this random behavior and makes it so that the
seed_everythingfalls back to setting a fixed seed 0. This change only affects users who have intentionally calledseed_everythingwith no arguments, and those who useseed_everything=Truein LightningCLI. Those uses will now have to select the number randomly first, and then pass it in toseed_everything, unless they are happy with the new default of 0.Fixes #18638
(to be confirmed)
📚 Documentation preview 📚: https://pytorch-lightning--18846.org.readthedocs.build/en/18846/
cc @Borda @carmocca @justusschock @awaelchli