Changes to Adapter constructors, require experiment #3415
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: D70103442 |
|
This pull request was exported from Phabricator. Differential Revision: D70103442 |
saitcakmak
added a commit
to saitcakmak/Ax
that referenced
this pull request
Feb 25, 2025
Summary: Pull Request resolved: facebook#3415 The motivation for this change is to bring us closer to relying on the `experiment` as the source of truth about the experiment state & attributes within the modeling layer. We currently support many inputs that are extracted from the `experiment`, just to be passed in alongside it. By making `experiment` a required input, we open the possibility of removing these extra inputs and extracting them directly from `experiment` where they're needed. Makes the following changes to Adapter constructors: - Requires keyword-only arguments. Positional inputs are no-longer supported. - Makes `experiment` required and `search_space` optional. - Re-orders inputs for consistency across sub-classes. In addition: - Removes `model` input to `Adapter._fit`. This is a private method that is only called through `fit_if_implemented` (with `self.model`). Accepting multiple inputs for the same argument only makes the code harder to reason about. - Removes class level attributes, some of which weren't initialized in `__init__`, leading to pyre complaints. All attributes are now initialized in `__init__`. This also eliminates misleading "optional" type hints with `None` default for `model`, which is never `None` in practice. - Removes `Adapter.update`, which has been deprecated for quite some time. - Initializing a Generator from registry with only `search_space` is being deprecated. It is temporarily supported using a dummy experiment for random & discrete adapters, which previously did not require an `experiment`. Differential Revision: D70103442
saitcakmak
force-pushed
the
export-D70103442
branch
from
9d4472e to
adfa3d8
Compare
saitcakmak
added a commit
to saitcakmak/Ax
that referenced
this pull request
Feb 25, 2025
Summary: The motivation for this change is to bring us closer to relying on the `experiment` as the source of truth about the experiment state & attributes within the modeling layer. We currently support many inputs that are extracted from the `experiment`, just to be passed in alongside it. By making `experiment` a required input, we open the possibility of removing these extra inputs and extracting them directly from `experiment` where they're needed. Makes the following changes to Adapter constructors: - Requires keyword-only arguments. Positional inputs are no-longer supported. - Makes `experiment` required and `search_space` optional. - Re-orders inputs for consistency across sub-classes. In addition: - Removes `model` input to `Adapter._fit`. This is a private method that is only called through `fit_if_implemented` (with `self.model`). Accepting multiple inputs for the same argument only makes the code harder to reason about. - Removes class level attributes, some of which weren't initialized in `__init__`, leading to pyre complaints. All attributes are now initialized in `__init__`. This also eliminates misleading "optional" type hints with `None` default for `model`, which is never `None` in practice. - Removes `Adapter.update`, which has been deprecated for quite some time. - Initializing a Generator from registry with only `search_space` is being deprecated. It is temporarily supported using a dummy experiment for random & discrete adapters, which previously did not require an `experiment`. Differential Revision: D70103442
saitcakmak
force-pushed
the
export-D70103442
branch
from
adfa3d8 to
c0c9dba
Compare
|
This pull request was exported from Phabricator. Differential Revision: D70103442 |
Summary: The motivation for this change is to bring us closer to relying on the `experiment` as the source of truth about the experiment state & attributes within the modeling layer. We currently support many inputs that are extracted from the `experiment`, just to be passed in alongside it. By making `experiment` a required input, we open the possibility of removing these extra inputs and extracting them directly from `experiment` where they're needed. Makes the following changes to Adapter constructors: - Requires keyword-only arguments. Positional inputs are no-longer supported. - Makes `experiment` required and `search_space` optional. - Re-orders inputs for consistency across sub-classes. In addition: - Removes `model` input to `Adapter._fit`. This is a private method that is only called through `fit_if_implemented` (with `self.model`). Accepting multiple inputs for the same argument only makes the code harder to reason about. - Removes class level attributes, some of which weren't initialized in `__init__`, leading to pyre complaints. All attributes are now initialized in `__init__`. This also eliminates misleading "optional" type hints with `None` default for `model`, which is never `None` in practice. - Removes `Adapter.update`, which has been deprecated for quite some time. - Initializing a Generator from registry with only `search_space` is being deprecated. It is temporarily supported using a dummy experiment for random & discrete adapters, which previously did not require an `experiment`. Differential Revision: D70103442
saitcakmak
force-pushed
the
export-D70103442
branch
from
c0c9dba to
6445681
Compare
|
This pull request was exported from Phabricator. Differential Revision: D70103442 |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #3415 +/- ##
==========================================
- Coverage 95.99% 95.99% -0.01%
==========================================
Files 539 539
Lines 52794 52800 +6
==========================================
+ Hits 50679 50683 +4
- Misses 2115 2117 +2 ☔ View full report in Codecov by Sentry. |
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:
The motivation for this change is to bring us closer to relying on the
experimentas the source of truth about the experiment state & attributes within the modeling layer. We currently support many inputs that are extracted from theexperiment, just to be passed in alongside it. By makingexperimenta required input, we open the possibility of removing these extra inputs and extracting them directly fromexperimentwhere they're needed.Makes the following changes to Adapter constructors:
experimentrequired andsearch_spaceoptional.In addition:
modelinput toAdapter._fit. This is a private method that is only called throughfit_if_implemented(withself.model). Accepting multiple inputs for the same argument only makes the code harder to reason about.__init__, leading to pyre complaints. All attributes are now initialized in__init__. This also eliminates misleading "optional" type hints withNonedefault formodel, which is neverNonein practice.Adapter.update, which has been deprecated for quite some time.search_spaceis being deprecated. It is temporarily supported using a dummy experiment for random & discrete adapters, which previously did not require anexperiment.Differential Revision: D70103442