Glasgow Constraint Solver

[tias]

Matthew's GCS solver interface

Unfinished: multiple constraint tests failing

[tias]

Turns out that the transformations were called with different arguments then what they should be.

I updated the docs to make it clearer what they should be.

All the expression tests now pass, so it seems this would be ready to merge...

I do have one request @mmcilree , can we call it 'gcs' instead of 'glassgowconstraintsolver'?
s.SolverLookup.get("glassgowconstraintsolver", m) is really a lot to type... "gcs" would be nicer... just like gcspy is your lib
(we don't support both a shorthand and longhand, just a single one)

[tias]

Hi Matthew @mmcilree ,

We've refactored our solver interface and solver making guidelines a bit, based also on this effort, and to make it easier to debug.

The two changes are that there should be a transform() function that does the transformations and returns the list of transformed constraints (very useful for debugging). And then the __add__() function can directly loop over the result of transform and post the constraint one by one (so no more need for a _post_constraint(), that gets merged into the add).

Would you be up for making those changes and testing what the current state is? Would be happy to see it progres...

(and we have a summerschool 2nd week of july, could be nice for participants to also have gcs available)

[Wout4]

I can't seem to install gcspy, did anyone else recently try?
Added it to our automated testing, and altough it says gcspy is succesfully installed the tests still give an error, saying we need to install gcspy

[mmcilree]

Hi @Wout4 the gcspy package is very out of date with respect to the Glasgow constraint solver - lots of the solver API has changed since I wrote the python bindings. It's on my todo list to update it and make sure it all works correctly, I'll try to get to it possibly next week.

[tias]

code review, only minor comments

[tias]

then indeed you need the bool() part for Boolean variables

…

On 11/07/2024 16:18, Matthew J McIlree wrote: ***@***.**** commented on this pull request. ------------------------------------------------------------------------ In cpmpy/solvers/gcs.py <#202 (comment)>: > + self.cpm_status.exitstatus = ExitStatus.FEASIBLE + else: + self.cpm_status.exitstatus = ExitStatus.UNSATISFIABLE + + # True/False depending on self.cpm_status + has_sol = self._solve_return(self.cpm_status) + + # translate solution values (of user specified variables only) + self.objective_value_ = None + if has_sol: + # fill in variable values + for cpm_var in self.user_vars: + sol_var = self.solver_var(cpm_var) + if isinstance(cpm_var, _BoolVarImpl): + # Convert back to bool + cpm_var._value = bool(self.gcs.get_solution_value(sol_var)) I'm not sure about this one. What do you mean by well-type? GCS always returns integers for solution values, so I thought I'd need to convert 0-1 back to False/True for boolvars. — Reply to this email directly, view it on GitHub <#202 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABKCNGAFZBSV263CWBRBSDZL2H35AVCNFSM6AAAAABKLFFWK6VHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDCNZSGA2DCNJTGE>. You are receiving this because you authored the thread.Message ID: ***@***.***> [ { ***@***.***": "http://schema.org", ***@***.***": "EmailMessage", "potentialAction": { ***@***.***": "ViewAction", "target": "#202 (comment)", "url": "#202 (comment)", "name": "View Pull Request" }, "description": "View this Pull Request on GitHub", "publisher": { ***@***.***": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

[Wout4]

I'm having some trouble adding an objective to a model:
e.g.

a = boolvar()
m = Model(a == 1)
m.minimize(a).
m.solve('gcs')

This crashes..
Other than that test are successful.

[Wout4]

we should remove this before merging

[Wout4]

so I guess we could add short_table here ?

[Wout4]

cpmpy's ShortTable is SmartTable

[Wout4]

I couldn't find smart table in the gcs python interface, could be nice to add still. Other than that seems ready to merge!

[Wout4]

Some test fail due to pycroptosat not being installed, will merge this anyway as that does not seem to happen on master