Handle equality constraints in UnitX transform (#5179)#5179
Open
sdaulton wants to merge 7 commits intofacebook:mainfrom
Open
Handle equality constraints in UnitX transform (#5179)#5179sdaulton wants to merge 7 commits intofacebook:mainfrom
sdaulton wants to merge 7 commits intofacebook:mainfrom
Conversation
meta-cla
bot
added
the
CLA Signed
|
@sdaulton has exported this pull request. If you are a Meta employee, you can view the originating Diff in D100256484. |
meta-codesync
bot
changed the title
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Apr 17, 2026
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
c9f6890 to
8c73ade
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Apr 17, 2026
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
8c73ade to
af3ce4f
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Apr 17, 2026
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
af3ce4f to
16087b6
Compare
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #5179 +/- ##
==========================================
+ Coverage 96.41% 96.42% +0.01%
==========================================
Files 618 619 +1
Lines 68882 69293 +411
==========================================
+ Hits 66410 66817 +407
- Misses 2472 2476 +4 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Summary: Add support for linear equality constraints (`w^T x == b`) alongside existing inequality constraints (`w^T x <= b`) in Ax's `ParameterConstraint` class. This is the first diff in a stack that threads equality constraints all the way down to BoTorch's `optimize_acqf`. Changes: - Add `extract_coefficient_dict_from_equality` to `ax/utils/common/sympy.py` for parsing `"expr == bound"` strings (SymPy can't parse `==` directly since Python evaluates it as a boolean). - Extend `ParameterConstraint.__init__` to accept `equality=` kwarg alongside existing `inequality=` kwarg. Exactly one must be provided. - Add `is_equality` property. - Update `check()` to use `|w^T x - b| <= tol` for equality constraints. - Update `__repr__`, `clone()`, `clone_with_transformed_parameters()`. - Add comprehensive tests for equality constraints. Differential Revision: D100256486
Summary: Add support for parsing equality constraint strings (e.g. `"x1 + x2 == 3"`) in `constraint_from_str`. This extends the existing `<=`/`>=` parsing to also accept `==` as a comparison operator. - Add `_process_equality_constraint` function (analogous to `_process_linear_constraint`) that constructs `ParameterConstraint(equality=...)`. - Detect `==` in `constraint_from_str` and route to the new function. - Reject equality order constraints (`"x1 == x2"`) with a clear error message. - Update `INVALID_CONSTRAINT_ERROR_MSG` to document `==` support. Differential Revision: D100256487
Summary: Update `SearchSpace.check_membership_df` and `compute_chebyshev_center` to handle equality constraints alongside existing inequality constraints. - `check_membership_df`: branch on `constraint.is_equality` — use `|weighted_sum - bound| <= tol` for equality, `weighted_sum <= bound + tol` for inequality. - `compute_chebyshev_center`: separate equality and inequality constraints into `A_eq/b_eq` and `A_ub/b_ub` for `scipy.optimize.linprog`. Equality constraints don't get the `r * ||a_i||` augmentation since the inscribed ball center must lie on the hyperplane. - `check_membership` already works via the updated `constraint.check()`. Differential Revision: D100256478
Summary: Add equality constraint extraction and propagation through the adapter layer to TorchOptConfig, enabling downstream generators to receive equality constraints. - Add `extract_equality_constraints` in `adapter_utils.py` (filters for `is_equality=True` constraints, returns `(A, b)` matrices). - Update `extract_parameter_constraints` to filter out equality constraints. - Add `equality_constraints` parameter to `validate_and_apply_final_transform` (now returns a 7-tuple). - Add `equality_constraints: tuple[Tensor, Tensor] | None` field to `TorchOptConfig`. - Update `TorchAdapter._get_transformed_model_gen_args` to extract equality constraints and pass them through to `TorchOptConfig`. Differential Revision: D100256480
Summary: Thread equality constraints from TorchOptConfig through to BoTorch's optimize_acqf and related optimizers. This is the key diff that connects Ax's equality constraint representation to BoTorch's SLSQP-based optimizer. - Add `_to_equality_constraints` in `torch/utils.py` — converts (A, b) tensor format to BoTorch's `(indices, coefficients, rhs)` format. No sign negation needed (equality is symmetric). - Update `BoTorchGenerator.gen()` to pass equality constraints to `acqf.optimize()`. - Add `equality_constraints` parameter to `Acquisition.optimize()` and forward to `optimize_acqf`, `optimize_acqf_mixed`, `optimize_acqf_mixed_alternating`. - Raise `ValueError` for discrete optimizers and NSGA-II (unsupported). - Update `validate_candidates` to check equality constraints. - Update `_prune_irrelevant_parameters` and `_remove_infeasible_candidates` to handle equality constraints. - Update `Surrogate.best_point` to pass equality constraints. Differential Revision: D100256482
Summary: Thread equality constraints from the search space through the random adapter to the `HitAndRunPolytopeSampler`. When rejection sampling falls back to polytope sampling, equality constraints from both `fixed_features` and parameter constraints are combined into a single `(C, c)` matrix. - Update `RandomAdapter._gen` to extract equality constraints via `extract_equality_constraints`. - Add `equality_constraints` parameter to `RandomGenerator.gen()`. - Add `_combine_equality_constraints` method that merges fixed-feature-based and parameter-based equality constraints for the polytope sampler. Differential Revision: D100256488
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Apr 20, 2026
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
16087b6 to
83d9a16
Compare
sdaulton
added a commit
to sdaulton/Ax-1
that referenced
this pull request
Apr 20, 2026
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
83d9a16 to
56a7509
Compare
Summary: Pull Request resolved: facebook#5179 Update the UnitX transform to preserve equality constraint type when transforming constraints to unit space. When constructing new constraints, check `c.is_equality` and use `ParameterConstraint(equality=...)` for equality constraints. The math is identical for both types (rescale coefficients by `(u - l)` and adjust bound by `w * l`). Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use `pc.clone()` which automatically preserves `is_equality` — no changes needed. Reviewed By: esantorella Differential Revision: D100256484
sdaulton
force-pushed
the
export-D100256484
branch
from
56a7509 to
6712134
Compare
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
2 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.
Summary:
Update the UnitX transform to preserve equality constraint type when
transforming constraints to unit space. When constructing new constraints,
check
c.is_equalityand useParameterConstraint(equality=...)forequality constraints. The math is identical for both types (rescale
coefficients by
(u - l)and adjust bound byw * l).Other transforms (OneHot, IntToFloat, RemoveFixed, etc.) use
pc.clone()which automatically preserves
is_equality— no changes needed.Reviewed By: esantorella
Differential Revision: D100256484