Fix target_train_data_fraction overriding pretrain_data_fraction#1070
Merged
Fix target_train_data_fraction overriding pretrain_data_fraction#1070
Conversation
pyeres
changed the title
pyeres
requested review from
W4ngatang,
iftenney,
pruksmhc and
sleepinyourhat
as code owners
Closed
sleepinyourhat
approved these changes
Apr 22, 2020
Contributor
sleepinyourhat
left a comment
sleepinyourhat
left a comment
There was a problem hiding this comment.
I see some naming issues that I'd recommend fixing, but the logic all looks reasonable. Merge when ready.
This was referenced Sep 17, 2020
leo-liuzy
pushed a commit
to leo-liuzy/dynamic_jiant
that referenced
this pull request
Nov 11, 2020
…-mll#1070) * add private instance generator field to Task w/ getter and setter * update build_tasks to use new instance generator Task field setter * update trainer to get phase-appropriate instance generators * update evaluate to use new Task instance generator getter * update pred writing task to use new instance generator getter/setter * add tests for new Task instance generator functionality
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR addresses the bug described in issue #1066 ("Results with pretrain_data_fraction args").
This PR...
Taskobject (_instance_generators) to store instance generators with the appropriate phase-specific data fraction settings.get_instance_generatorand setterset_instance_generator, toTaskto access the_instance_generators.Task's that an exception is raised when training instance generator is requested without specifying the phase (because phase determines data fraction).Validation: the config provided in issue #1066 was run with these code changes. Results logged for the task (MNLI) using a pretraining data fraction were more in line with expectation: after 25k steps training accuracy was ~100%, while val and test accuracy were ~75%.