Use MOI.Nonlinear.QPBlockData for the quadratic constraints - #237
Open
blegat wants to merge 5 commits into
Open
Conversation
Replace the per-constraint QuadraticConstraint storage (COO Hessian, sparse gradient, gradient-sparsity dictionary) and its bespoke evaluation loops with a MOI.Nonlinear.QPBlockData holding the quadratic constraints, in both MathOptNLPModel and MathOptNLSModel. QuadraticConstraints keeps its nquad/nnzj/nnzh fields and now also stores the precomputed Jacobian and Hessian structures of the block (with per-constraint Hessian offsets for jth_hess_coord!). The block-shaped NLPModels methods (cons, jac, jprod, jtprod, hess, hprod) call the corresponding MOI evaluator methods of the block; the per-constraint methods (jth_hess_coord!, jth_hprod!, ghjvprod!) use MOI.Nonlinear's per-function helpers. LinearConstraints and Objective are deliberately kept: their materialized sparse structures are performance fast paths that QPBlockData would not preserve. Behavioral notes: - The Jacobian structure of a quadratic constraint may now contain duplicate column entries (one per term) instead of the previous deduplicated gradient sparsity, so nnzj counts can differ; NLPModels COO conventions allow duplicates. - VectorQuadraticFunction constants are now handled with MOI.Utilities.shift_constant, fixing the previous handling that produced lcon = +constant for Nonnegatives and an infeasible lcon/ucon pair for Zeros with nonzero constants. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
|
Follow-up to the rename in MathOptInterface: the accumulating product functions of QPBlockData are no longer methods of the MOI.eval_... generic functions. The calls on nls.Feval are unchanged: that is an MOI.Nonlinear.Evaluator, for which the MOI generic functions apply.
* Use MOI.Nonlinear.ModelWithQuad instead of QPBlockData Store the parsed quadratic constraints in a MOI.Nonlinear.ModelWithQuad and evaluate them through the corresponding MOI.Nonlinear.EvaluatorWithQuad, instead of using the QPBlockData and its private helpers directly. The per-constraint methods (jth_hess_coord!, jth_hprod! and ghjvprod!) evaluate the Hessian of the Lagrangian of the block with a basis multiplier vector, which removes the per-constraint Hessian offsets. The products of the evaluator overwrite their output, so the Jacobian-transpose and Hessian products go through a new hv workspace before being accumulated. * MOI * Add the variables to the ModelWithQuad MOI.Nonlinear.ModelWithQuad now owns the variables of the model, so the parser adds them and the evaluator no longer takes ordered_variables. * MOI
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
1 participant
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.
A net line deletion of 145 lines. Needs
LinearConstraints and Objective are deliberately kept: their materialized sparse structures are performance fast paths that QPBlockData would not preserve.
Behavioral notes:
VectorQuadraticFunctionconstants are now handled with MOI.Utilities.shift_constant, fixing the previous handling that producedlcon = +constantforNonnegativesand an infeasible lcon/ucon pair for Zeros with nonzero constants.