558 reporting intermediate solutions by IgnaceBleukx · Pull Request #561 · CPMpy/cpmpy

IgnaceBleukx · 2024-12-20T14:38:07Z

Following the discussion in #558.

It adds quite some code in all solver interfaces, but I tried to do it somewhat consistent.
Some solvers are missing still:

minizinc - this needs a lot of engineering with async functions (similar to solveAll)
gcs - there is a callback function but as optimization is not working properly (see Objective in GCS does not have any influence on solution quality #559) I could not test it.

For Exact,it is somewhat hacky as we need to go through "runOnce" instead of the nice "toOptimal" function. So there will be some more ping-pong through the interface.

ThomSerg

Seems a nice quality-of-life improvement. But yeah, the one for exact seems hack-y, especially with the "warning" part. Is there a way to restore the solver afterwards?

cpmpy/solvers/gurobi.py

cpmpy/solvers/minizinc.py

cpmpy/solvers/ortools.py

cpmpy/solvers/TEMPLATE.py

cpmpy/solvers/gurobi.py

ThomSerg · 2025-01-02T15:30:04Z

Besides that it is never mentioned + silently enforced that "display" is only for optimization, it seems good to me.

Wout4 · 2025-01-06T16:03:48Z

we could add the gcs callback also since that bug is fixed or not?

…porting-intermediate-solutions # Conflicts: # cpmpy/solvers/z3.py

IgnaceBleukx · 2025-01-09T09:17:54Z

Made quite some changes to this PR after the last review, so maybe another round of revieuwing from @ThomSerg and/or @Wout4 would be appropriate before merging.

I'm wondering if we should really fill the solver.objective_ argument directly from the underlying solver object.
Alternatively, we can store the original objective expression and compute the objective value based on that. There were quite some tricky things to get right in this PR...

ThomSerg

Went through the entire thing again

ThomSerg · 2025-01-13T12:06:02Z

cpmpy/solvers/choco.py

            self.chc_solver.limit_time(str(time_limit) + "s")

+        if display is not None:
+            raise NotImplementedError("Choco does not support solution callbacks, (TODO?)")


What is on purpose to leave this "TODO" here?

Choco in Java does have callbacking, so just leaving it here for if and when they implement callbacking directly from Python as well

We could also implement it as in Exact, but there will be a lot more overhead in Choco (needs to pass trough C++ and Java...)

cpmpy/solvers/exact.py

ThomSerg · 2025-01-13T13:22:16Z

cpmpy/solvers/minizinc.py

-
+
+        if display is not None:
+            raise NotImplementedError("TODO: implement MiniZinc callbacks -- requires new async methhod (similar to solveAll())")


Meant to leave this TODO?

MiniZinc does support it, but it requires quite a bit of engineering so better for another PR I think

cpmpy/solvers/exact.py

# Conflicts: # cpmpy/solvers/gurobi.py

IgnaceBleukx · 2025-01-17T09:52:44Z

Should be fine now @ThomSerg

IgnaceBleukx · 2025-07-24T08:09:08Z

This PR has been stale for quite some time now, and not sure how to continue on it.
Do we need a more structured approach to support these callbacks? Do we want it at all?
I think it's a nice quality of life improvement, but it also adds a lot of extra code to our solve functions which hinders future maintainability...
Therefore, I'm also fine with abandoning this for now and coming back to it later -- interested to hear others opinions : )

tias · 2026-02-02T09:46:41Z

I propose we do not add it for all solvers, but instead solver interfaces can choose to implement the display= function.

So we would add it to the template and implement it for OrTools. But we would not attempt to make it work for all solvers.

…mediate-solutions # Conflicts: # cpmpy/solvers/TEMPLATE.py # cpmpy/solvers/choco.py # cpmpy/solvers/exact.py # cpmpy/solvers/gcs.py # cpmpy/solvers/gurobi.py # cpmpy/solvers/minizinc.py # cpmpy/solvers/ortools.py # cpmpy/solvers/z3.py

IgnaceBleukx · 2026-04-03T13:48:37Z

Merged master into this one and removed the hacky callbacking for solvers who don't support it natively.

I'm slightly stuck on a couple of things:

Docs: how to describe what the callback is? For now I've copied over the description we use in solveAll
Semantics: solvers have different semantics for when the callback is called... Gurobi and OR-Tools call it for each (improving?) solution, Hexaly just calls the callback periodically during solve
Backwards compatibility: most solvers already had a solution_callback argument that you could give (ortools, cpo and hexaly). Do we want to keep this possibility? For now I've kept it but the argument has been moved to **kwargs. Alternatively, we can simply re-use the solution_callback argument and check if it is already a callback? E.g., an OrtSolutionPrinter for ortools. Otherwise we wrap it in such a callback class ourself?

If we get the callbacking somewhat clean, we can massively cleanup the solveall code for the solvers that support it too I think. But definately for another PR

…mediate-solutions

IgnaceBleukx added 7 commits December 20, 2024 15:02

update ortools with display arg

f5d31c4

update gurobi with display arg

5c95c38

update exact with display arg

dce66af

add for unsupported solvers

d953ab9

add for z3

0251725

add tests

0b5e1f4

update template

96569d5

IgnaceBleukx linked an issue Dec 20, 2024 that may be closed by this pull request

Reporting intermediate solutions #558

Open

IgnaceBleukx requested a review from ThomSerg December 20, 2024 14:38

ThomSerg reviewed Dec 20, 2024

View reviewed changes

IgnaceBleukx added 5 commits December 20, 2024 16:28

remove random import

9ff94c8

fix typo

f08777c

fix typo

47b71df

remove intermediate var

700324f

update description of callback function

0ede47f

IgnaceBleukx added 11 commits January 7, 2025 10:51

callbacks in gcs

7784e93

update docstring to make clear display is only for opt

1ba0125

add gcspy to github workflow

4c65d56

re-update docstring after merge

9dbbd15

remove gcs from gh run

c563223

Merge remote-tracking branch 'refs/remotes/origin/master' into 558-re…

24a43e9

…porting-intermediate-solutions # Conflicts: # cpmpy/solvers/z3.py

ensure exact also works with logging and incrementality

c02601d

update on tests

de39aad

gurobi objective should be taken after optimization call...

d9a9f6c

update tests

370acc9

z3 maximize yields negative version of objective

f36a061

only run until time limit is reached

ab6a556

ThomSerg requested changes Jan 13, 2025

View reviewed changes

IgnaceBleukx added 4 commits January 16, 2025 14:07

fix comment on exact callback

97bd038

also check for timeout explicitly (avoid 1 more iter)

d278ab0

Merge branch 'master' into 558-reporting-intermediate-solutions

9d86789

# Conflicts: # cpmpy/solvers/gurobi.py

fix test to accept solver arg

e28389d

tias marked this pull request as draft April 7, 2025 12:56

ThomSerg mentioned this pull request Dec 23, 2025

solution_callback for Hexaly #787

Merged

IgnaceBleukx added 12 commits April 3, 2026 09:34

update z3 interface

e3c3eb2

update cpo interface

a53b3a5

remove display from choco

0f990dd

update gcs interface

f181095

update ortools interface

40fb4ec

update tests

f57c240

solution callback cpo

7d34ded

refactor callback function gurobi

6812a15

print with argvals

4aba0c6

update hexaly interface

115fcf4

update tests for hexaly

acb1aae

IgnaceBleukx marked this pull request as ready for review April 3, 2026 13:43

IgnaceBleukx added 4 commits April 6, 2026 22:06

Merge remote-tracking branch 'origin/master' into 558-reporting-inter…

020845b

…mediate-solutions

reintroduce solution_callback arg

8f310ba

format docstrings

bfe943a

fix callback printing

a2f1721



		if display is not None:
		raise NotImplementedError("TODO: implement MiniZinc callbacks -- requires new async methhod (similar to solveAll())")

Conversation

IgnaceBleukx commented Dec 20, 2024

Uh oh!

ThomSerg left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ThomSerg commented Jan 2, 2025

Uh oh!

Wout4 commented Jan 6, 2025

Uh oh!

IgnaceBleukx commented Jan 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ThomSerg left a comment

Choose a reason for hiding this comment

Uh oh!

ThomSerg Jan 13, 2025

Choose a reason for hiding this comment

Uh oh!

IgnaceBleukx Jan 13, 2025

Choose a reason for hiding this comment

Uh oh!

IgnaceBleukx Jan 16, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ThomSerg Jan 13, 2025

Choose a reason for hiding this comment

Uh oh!

IgnaceBleukx Jan 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

IgnaceBleukx commented Jan 17, 2025

Uh oh!

IgnaceBleukx commented Jul 24, 2025

Uh oh!

tias commented Feb 2, 2026

Uh oh!

IgnaceBleukx commented Apr 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

IgnaceBleukx commented Jan 9, 2025 •

edited

Loading

IgnaceBleukx Jan 13, 2025 •

edited

Loading