Add init_position argument to UniformGenerator #2686
Closed
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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D61622058 |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2686 +/- ##
==========================================
- Coverage 95.29% 95.29% -0.01%
==========================================
Files 495 495
Lines 47832 47829 -3
==========================================
- Hits 45582 45579 -3
Misses 2250 2250 ☔ View full report in Codecov by Sentry. |
|
This pull request was exported from Phabricator. Differential Revision: D61622058 |
saitcakmak
added a commit
to saitcakmak/Ax
that referenced
this pull request
Aug 21, 2024
Summary: Pull Request resolved: facebook#2686 `SobolGenerator` uses `init_position` to ensure that when the model is reconstructed, it resumes candidate generation from where it was left (rather than starting from the beginning of the sequence). Without this, when `deduplicate=False`, the model would generate the same points it has already generated, which would lead to different candidate generation behaviors depending on how often the model was reconstructed. This is undesirable as we want the model to resume generation rather than repeating from scratch. Prior to this diff, `UniformGenerator` did not have `init_position`, so had this exact issue. We fix it here. Reviewed By: Balandat Differential Revision: D61622058
saitcakmak
force-pushed
the
export-D61622058
branch
from
ef379d1
to
353054b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D61622058 |
saitcakmak
force-pushed
the
export-D61622058
branch
from
353054b
to
415da14
Compare
saitcakmak
added a commit
to saitcakmak/Ax
that referenced
this pull request
Aug 21, 2024
Summary: Pull Request resolved: facebook#2686 `SobolGenerator` uses `init_position` to ensure that when the model is reconstructed, it resumes candidate generation from where it was left (rather than starting from the beginning of the sequence). Without this, when `deduplicate=False`, the model would generate the same points it has already generated, which would lead to different candidate generation behaviors depending on how often the model was reconstructed. This is undesirable as we want the model to resume generation rather than repeating from scratch. Prior to this diff, `UniformGenerator` did not have `init_position`, so had this exact issue. We fix it here. Reviewed By: Balandat Differential Revision: D61622058
|
This pull request was exported from Phabricator. Differential Revision: D61622058 |
saitcakmak
added a commit
to saitcakmak/Ax
that referenced
this pull request
Aug 21, 2024
Summary: Pull Request resolved: facebook#2686 `SobolGenerator` uses `init_position` to ensure that when the model is reconstructed, it resumes candidate generation from where it was left (rather than starting from the beginning of the sequence). Without this, when `deduplicate=False`, the model would generate the same points it has already generated, which would lead to different candidate generation behaviors depending on how often the model was reconstructed. This is undesirable as we want the model to resume generation rather than repeating from scratch. Prior to this diff, `UniformGenerator` did not have `init_position`, so had this exact issue. We fix it here. Reviewed By: Balandat Differential Revision: D61622058
saitcakmak
force-pushed
the
export-D61622058
branch
from
415da14
to
0b6a73e
Compare
Summary: Prior to this, if you compared a float with an ndarray, you'd get a boolean ndarray, which would lead to `ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()`. Differential Revision: D61666276
Summary: Pull Request resolved: facebook#2686 `SobolGenerator` uses `init_position` to ensure that when the model is reconstructed, it resumes candidate generation from where it was left (rather than starting from the beginning of the sequence). Without this, when `deduplicate=False`, the model would generate the same points it has already generated, which would lead to different candidate generation behaviors depending on how often the model was reconstructed. This is undesirable as we want the model to resume generation rather than repeating from scratch. Prior to this diff, `UniformGenerator` did not have `init_position`, so had this exact issue. We fix it here. Reviewed By: Balandat Differential Revision: D61622058
|
This pull request was exported from Phabricator. Differential Revision: D61622058 |
saitcakmak
force-pushed
the
export-D61622058
branch
from
0b6a73e
to
c812f1b
Compare
|
This pull request has been merged in e800973. |
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.
Summary:
SobolGeneratorusesinit_positionto ensure that when the model is reconstructed, it resumes candidate generation from where it was left (rather than starting from the beginning of the sequence). Without this, whendeduplicate=False, the model would generate the same points it has already generated, which would lead to different candidate generation behaviors depending on how often the model was reconstructed. This is undesirable as we want the model to resume generation rather than repeating from scratch.Prior to this diff,
UniformGeneratordid not haveinit_position, so had this exact issue. We fix it here.Differential Revision: D61622058