[Depends on #3991] Adding method for multistart initialization of nonlinear models. by stephenscini · Pull Request #3994 · Pyomo/pyomo

stephenscini · 2026-07-13T17:12:52Z

Fixes # .

Summary/Motivation:

Additional method in devel/initialization to use multistart optimization as a way for initializing nonliner models.

Changes proposed in this PR:

Updating contrib/multistart solver for reproducibility, and adding support for new solver interface.
Adding support for uniform, lhs, and sobol sampling within the multistart solver
Adding method to devel/initialization

AI-Use Disclosure

AI tools contributed to the development of this PR
- AI tools generated code (apart from tests)
Review process (select ONE):
- Rewritten: All AI-generated content was rewritten by me before being committed.

Notes for reviewers (optional):

Legal Acknowledgement

By contributing to this software project, I have read the contribution guide and agree to the following terms and conditions for my contribution:

I agree my contributions are submitted under the BSD license.
I represent I am authorized to make the contributions and grant the license. If my employer has rights to intellectual property that includes these contributions, I represent that I have received permission to make contributions and grant the required license on behalf of that employer.

stephenscini · 2026-07-13T17:17:22Z

Opening PR, but still in active development. Dependent on #3991 first as it includes those changes.

stephenscini · 2026-07-21T17:35:39Z

Note for 7/21 Dev meeting

Method implemented with support for new solver interface. All original tests passing locally.

Tasks before ready for review:

It looks like there is an issue with changing the SolverFactory version in a different place, which I will address.
Try and simplify interface, check with mentors.
Add tests for new functionality

stephenscini

Went through and added some questions in the code for design meeting.

stephenscini · 2026-07-23T22:13:57Z

 from pyomo.core.expr.compare import assertExpressionsEqual
 from pyomo.core.expr.numeric_expr import LinearExpression

+from pyomo.contrib.multistart.multi import MultiStart


I received an error message when I added the solver to the new SolverFactory that it needed to be in this list. This would mean it is tested with the other solvers from contrib/solver and I am not sure that if that is expected? If so, will all contrib solvers need to eventually be switched over or should the opt/solvers still be the best location.

In this case, I wanted to update it to use these solvers so thought it would also go here. Maybe Legacy?

stephenscini · 2026-07-23T22:18:49Z


-def rand(val, lb, ub):
-    return random.uniform(lb, ub)  # uniform distribution between lb and ub
+def rand(val, lb, ub, rng):


Previously, there was not a set way I could find to make the solver result reproducible, so I made the default random generator np.random with the current RNG interface.

I wanted to otherwise leave the established strategies alone, but could also make all random strategies support the different random generator options? Kept all newer options under rand_vector to use a vectorized approach.

stephenscini · 2026-07-23T22:20:31Z

+
+    if config.strategy == "rand_vector":
+
+        if len(eligible_vars) == 0:


Before, if all the variables are unbounded, it would not reset them but just give the warning. I thought it would be better to error out.

stephenscini · 2026-07-23T22:23:50Z

 logger = logging.getLogger('pyomo.contrib.multistart')


+@document_configdict()


Made changes so it is now loaded as a solver for the new SolverFactory since I wanted to use the new solver interfaces like the other devel/initialization methods, and from discussion with Bethany. Causing testing errors but would like to discuss best way to do this if other method preferred.

stephenscini · 2026-07-23T22:24:40Z

+    def version(self):
+        """Get solver version
+        TODO: This is a solver wrapper, unsure how to define version in this case."""
+


Currently no versions for this. If old was v0.0, would this be v0.1?

stephenscini · 2026-07-23T22:28:44Z

+            """,
+            ),
+        )
+        self.solver = self.declare(


Should I make this have different solver acceptances like MindtPy where there are supported categories, or make it so it accepts the solver object? How consistent should the contrib solvers that accept other solvers be with handling internal solver?

stephenscini · 2026-07-23T22:32:32Z

+            visibility=visibility,
+        )
+
+        self.strategy = self.declare(


Could/should this also support a user supplying a list of initial points? Thinking of obj_at_theta in parmest. Goal in my head is to make as generalized as possible.

stephenscini · 2026-07-23T22:35:03Z

            raise RuntimeError(
                "Multistart solver is unable to handle model with multiple active objectives."
            )
-        if obj is None:


Previously it was restricted from using square models or constant objectives in the multistart solver. Is there a valid reason for this or can it be removed? Is there a case where a model would have no objective or would that cause a different error? Removing would make more sense in my opinion.

stephenscini · 2026-07-23T22:36:26Z

+                description="Tolerance on HCS objective value equality. Defaults to Python float equality precision.",
+            ),
+        )
+        self.break_on_solution = self.declare(


For initialization, I thought this made sense, other thought would be implementing a SolutionLimit argument like Gurobi.

stephenscini · 2026-07-23T22:39:32Z

+                    )
+                    # If we are looking for the first feasible solution, then return immediately
+                    if config.break_on_solution:
+                        return best_result


Would it be a good idea to have this return a dictionary or something instead of just the best result if someone wanted to inspect? Or is this better for a user to set up? Should a user be able to access subsolver information too?

stephenscini · 2026-07-27T18:20:06Z

Had discussion with Bethany and Michael. Main takeaway points:

Invest to make this work for newer SolverFactory. Do not keep in opt.SolverFactory.
Try and resolve all test failures with the new SolverFactory as default. More discussion needed on expectations for contrib solvers as they transition to new SolverFactory.
If all results are infeasible/not optimal, have it throw the 'NoOptimalSolution' error.
-Keep this PR focused on a minimum working version that has key functionality we want. Can always add additional touches later.
LHS makes sense to be default for all methods/strategies, and should be looped in to the old strategies too.
Availability of multistart should depend on availablility of chosen sub-solver (default is ipopt)
Version 0.1.0 makes sense for this new iteration.
Subsolver has its own configuration, but still needs to comply with 'meta-solver' configurations. If subsolver timelimit is 1 minute for 50 runs, but metasolver is 10 minutes, when metasolver time runs out the methods stop.

stephenscini added 23 commits July 2, 2026 13:07

Add time limit warning to global solver

7eb0cbe

Ran black

0f97cb0

Adjusted order to solve unbounded nlp at end of method

352c828

Made naming consistent, commented out pwl while debugging tests

8c0a358

Fixed test, ran black

b6947a0

Forgot comment, ran black

3cb1585

Initial setup progress for multistart initialization

cc40909

Work on lhs, in progress

4e46f90

Semi working version of multistart!

fb66e6c

Changes to resolve testing errors, in progress

86cbd73

Added more loggers to debug stuff

2f17c59

Fix typo

65f7398

Solution loading too strict

55a1e88

Caused other issues, putting this back

c86cbc0

define helper fcn to retry solve, removed from finally block

6aa9ce0

Ran black

3753f4b

Moved string, Ran black

9dd6c48

Merge branch 'initialization-devel' into multistart-init

691b3cd

Made multistart also use nlp solve

2870aff

Adding new features for initialization, in progress

e84f9fb

First draft of additional sampling support and break_on_solution

64ea87a

Ran black

4424e25

Trying to support old and new solver interfaces, in progress.

ead5c80

stephenscini added 6 commits July 13, 2026 11:20

Ran black

1561d17

Ran black on devel/initialization

207bd13

Switching over to new solver factory interface, ran black

9e734b1

Added new initialize bool. Ran black

331d8c8

Merge branch 'main' into multistart-init

e9fa46e

Fixed SolverFactory issue

99d738a

stephenscini changed the title ~~Adding method for multistart initialization of nonlinear models.~~ [Depends on #3991] Adding method for multistart initialization of nonlinear models. Jul 21, 2026

Added vectorized strategy, adjusted defaults, ran black.

db10cfb

stephenscini added 6 commits July 21, 2026 13:35

Changed configs

93e1e7e

Updated solverwrapper to meet new requirements, ran black

e4f564c

Add back suppression warning

4a80136

Ran black in devel

99a51f1

Added multistart to solver list in contrib.solvers

003801f

Ran black

c5e2109

stephenscini commented Jul 23, 2026

View reviewed changes

stephenscini added 6 commits July 23, 2026 17:17

Finished example, ran black

281ff01

Used utility to get vars from shallow clone

127a973

Remove extra print statements

e087e99

Modify config

78f0869

Ran black

45dd8c5

Added break on solution

d58d937

stephenscini added 3 commits July 27, 2026 12:52

Fixing equal bound variables

b7262fb

Starting to fix issues with test_solvers

aa9ba5b

Ran black

a312c49


		if config.strategy == "rand_vector":

		if len(eligible_vars) == 0:

		logger = logging.getLogger('pyomo.contrib.multistart')


		@document_configdict()

Uh oh!

Conversation

stephenscini commented Jul 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Fixes # .

Summary/Motivation:

Changes proposed in this PR:

AI-Use Disclosure

Legal Acknowledgement

Uh oh!

stephenscini commented Jul 13, 2026

Uh oh!

stephenscini commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stephenscini left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

stephenscini commented Jul 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

stephenscini commented Jul 13, 2026 •

edited

Loading

stephenscini commented Jul 21, 2026 •

edited

Loading