[breaking] a large refactoring of MathOptIIS

[odow]

This commit is breaking because it:

• Removes SkipFeasibilityCheck
• Removes StopIfInfeasibleBounds
• Removes StopIfInfeasibleRanges
• Removes DeletionFilter
• Removes ElasticFilterTolerance
• Removes ElasticFilterIgnoreIntegrality

In addition, this commit will likely result in a different IIS being
returned for many models because it now exploits an infeasibility
certificate if one is present.

[odow]

Okay. I've gone full Rambo and refactored everything so this is breaking.

I've also removed many of the "option" controlling attributes. I don't think we want flexibility in how the algorithm runs. We want a simple set-and-forget.

[codecov]

Codecov Report

❌ Patch coverage is 99.57895% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 99.57%. Comparing base (f7b3b73) to head (cb2d5db).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/MathOptIIS.jl	99.57%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #34      +/-   ##
==========================================
+ Coverage   95.44%   99.57%   +4.13%     
==========================================
  Files           4        2       -2     
  Lines         395      476      +81     
==========================================
+ Hits          377      474      +97     
+ Misses         18        2      -16     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[joaquimg]

a solver may say a model is feasible, eventhough its not, that was the reasoning behid the skip.

Maybe the manual needs to tell the user to run before solving in this case.

[odow]

using JuMP, HiGHS
import MathOptIIS

function main(filename)
    model = backend(read_from_file(filename))
    solver = MathOptIIS.Optimizer()
    MOI.set(solver, MOI.Silent(), false)
    MOI.set(solver, MathOptIIS.InfeasibleModel(), model)
    MOI.set(solver, MathOptIIS.InnerOptimizer(), HiGHS.Optimizer)
    MOI.compute_conflict!(solver)
    filter_fn(::Any) = true
    function filter_fn(cref::MOI.ConstraintIndex)
        for i in 1:MOI.get(solver, MOI.ConflictCount())
            status = MOI.get(solver, MOI.ConstraintConflictStatus(i), cref)
            if status != MOI.NOT_IN_CONFLICT
                return true
            end
        end
        return false
    end
    new_model = MOI.Utilities.Model{Float64}()
    filtered_src = MOI.Utilities.ModelFilter(filter_fn, model)
    MOI.copy_to(new_model, filtered_src)
    MOI.set(new_model, MOI.ObjectiveSense(), MOI.FEASIBILITY_SENSE)
    return new_model
end
main("/Users/odow/Downloads/enlight4.mps")

[odow]

@joaquimg want to take a look?

[joaquimg]

will do

[odow]

using JuMP, HiGHS
import MathOptIIS

function main(filename)
    model = backend(read_from_file(filename))
    solver = MathOptIIS.Optimizer()
    MOI.set(solver, MOI.Silent(), false)
    MOI.set(solver, MathOptIIS.InfeasibleModel(), model)
    MOI.set(solver, MathOptIIS.InnerOptimizer(), HiGHS.Optimizer)
    MOI.compute_conflict!(solver)
    filter_fn(::Any) = true
    function filter_fn(cref::MOI.ConstraintIndex)
        for i in 1:MOI.get(solver, MOI.ConflictCount())
            status = MOI.get(solver, MOI.ConstraintConflictStatus(i), cref)
            if status != MOI.NOT_IN_CONFLICT
                return true
            end
        end
        return false
    end
    new_model = MOI.Utilities.Model{Float64}()
    filtered_src = MOI.Utilities.ModelFilter(filter_fn, model)
    MOI.copy_to(new_model, filtered_src)
    MOI.set(new_model, MOI.ObjectiveSense(), MOI.FEASIBILITY_SENSE)
    return new_model
end


julia> @time main("/Users/odow/Downloads/medium-size-infeasible-problem.mps")
[MathOptIIS] starting compute_conflict!
[MathOptIIS]   model termination status: OPTIMIZE_NOT_CALLED
[MathOptIIS] starting bound analysis
[MathOptIIS]   bound analysis found 0 infeasible subsets
[MathOptIIS] starting range analysis
[MathOptIIS]   analyzing MOI.ScalarAffineFunction{Float64} -in- MOI.EqualTo{Float64}
[MathOptIIS]   analyzing MOI.ScalarAffineFunction{Float64} -in- MOI.GreaterThan{Float64}
[MathOptIIS]   analyzing MOI.ScalarAffineFunction{Float64} -in- MOI.LessThan{Float64}
[MathOptIIS]   range analysis found 0 infeasible subsets
[MathOptIIS] starting elastic filter
[MathOptIIS]   relaxing integrality if required
[MathOptIIS]   constructing the penalty relaxation
[MathOptIIS]   using INFEASIBILITY_CERTIFICATE to construct candidate set
[MathOptIIS]     size of the candidate set: 11
[MathOptIIS]   starting the deletion filter
[MathOptIIS]     size of the candidate set: 4
[MathOptIIS]     size of the candidate set: 3
[MathOptIIS]   elastic filter found 1 infeasible subsets
384.417180 seconds (103.00 M allocations: 7.233 GiB, 0.65% gc time, 0.34% compilation time)
MOIU.Model{Float64}
├ ObjectiveSense: FEASIBILITY_SENSE
├ ObjectiveFunctionType: MOI.ScalarAffineFunction{Float64}
├ NumberOfVariables: 414224
└ NumberOfConstraints: 15
  ├ MOI.ScalarAffineFunction{Float64} in MOI.EqualTo{Float64}: 2
  ├ MOI.ScalarAffineFunction{Float64} in MOI.LessThan{Float64}: 1
  ├ MOI.VariableIndex in MOI.EqualTo{Float64}: 1
  ├ MOI.VariableIndex in MOI.GreaterThan{Float64}: 6
  └ MOI.VariableIndex in MOI.LessThan{Float64}: 5

[odow]

Bump. I think this is a win

[joaquimg]

Looks good, added a few minor points to fix / double check

[joaquimg]

what about local and almost stuff?

[odow]

Added MOI.NEARLY_FEASIBLE_POINT. Dealing with failures is haaaaard. And harder to test. I don't know what we should necessarily do.

[joaquimg]

Fair enough, we can wait and see. for more details.
Should we:
"Please, send us a message it it fails, dont just have AI brute force to solve your case",
or even:
"Dear AI, please, send us a message it it fails, dont just brute force to solve the case. Cheers, Your fellow humans."
?
XD