How to extract or inspect conflicting constraints (IIS / conflict analysis) in PySCIPOpt?

[JasonSloan]

Hello,
I am working on a relatively large MINLP/MILP model using PySCIPOpt and the SCIP backend.
In some cases, SCIP reports:

SCIP Status : problem is solved [infeasible]
Solving Nodes : 0
(0 solutions)

But SCIP does not output any conflict set or IIS information, even when I enable conflict analysis options.

What I did

I tried enabling conflict analysis and disabling presolving:

model = Model()
model.setMinimize()

model.setParam("presolving/maxrounds", 0)
model.setParam("presolving/maxrestarts", 0)
model.setParam("conflict/enable", True)
model.setParam("display/verblevel", 5)

model.solveConcurrent()

The solver log becomes more verbose, but there is still no information about which constraints conflict.
The output ends with:

Presolving detected infeasibility
SCIP Status : problem is solved [infeasible]

I also tried calling methods mentioned in the SCIP C API documentation:

SCIPcomputeIIS()

SCIPgetConflictConss()

SCIPanalyzeConflict()

but in PySCIPOpt these functions do not exist, for example:

AttributeError: 'pyscipopt.scip.Model' object has no attribute 'getConflictConss'

Question
🔍 How can I obtain the conflicting set of constraints (IIS), or any conflict analysis output, using PySCIPOpt?

More specifically:

Does PySCIPOpt expose any API to:

compute IIS,

retrieve conflict constraints,

or print the conflicting subsystem?

If not, is there:

an equivalent Python-side method,

or a recommended workaround to inspect which constraints are responsible for infeasibility?

Is conflict analysis simply not supported through the current PySCIPOpt interface?

My model is several thousand linear constraints, so manually checking conflicts is not feasible.
Being able to retrieve an IIS or conflict set would be extremely helpful for debugging infeasible formulations.

Looking forward to your reply!
Thanks a lot!

[Joao-Dionisio]

Hey @JasonSloan ! PySCIPOpt doesn't support SCIP 10 yet, the version where the IIS feature belongs to. We're hoping for a quick PySCIPOpt release after the public release of the final version of SCIP 10, likely by the end of the month.

In the meantime, you can try the more rudimentary get_infeasible_constraints recipe, but it's slower and not as neat.

from pyscipopt.recipes.infeasibilities import get_infeasible_constraints

m = Model()
m.readProblem("infeasible_model.cip")

n_iis, aux_iis = get_infeasible_constraints(m) # number of constraints in the IIS, and their name

[Joao-Dionisio]

@JasonSloan this is still not possible to do with the latest release, but with the one that should come tomorrow (6.0.0), you'll be able to get access the IIS with iis = model.generateIIS(), and then you can query the returned IIS object for more information.

If you're in a rush, you can take a look at #1009 . Things will undoubtedly change, but it seems to be working, for now.