MathOpt SCIP tests failing

[pet-mit]

Hello, I'm trying to run the MathOpt-GScip tests (math_opt_solvers_gscip_solver_test).
I am getting this error:

math_opt_solvers_gscip_solver_test: (...)/or-tools/cmake-build-debug/_deps/scip-src/src/scip/prop_symmetry.c:6972: SCIPincludePropSymmetry: Assertion `strlen(SYMsymmetryGetName()) >= 5' failed.

(debugging it, I saw that SYMsymmetryGetName() returns "none").
Is something wrong with my configuration?

Thank you for your help

[lperron]

Which version ? Which build system ? Which platform ?

[lperron]

Bliss is disabled on windows

[pet-mit]

Using cmake, on the main branch (actually my branch here that is up to date with your main branch), under Ubuntu 22.
I was not aware we need third party software. Should I install bliss or something else?

[pet-mit]

changing "SYM" from "none" to "bliss" when building SCIP here seems to fix the issue for me

[lperron]

Yes, but I do not think we pull bliss in our cmakefiles. So it may work for you as bliss is installed, but this is not robust. Laurent Perron | Operations Research | ***@***.*** | (33) 1 42 68 53 00 Le mer. 15 janv. 2025 à 15:46, Peter Mitri ***@***.***> a écrit :

…

changing "SYM" from "none" to "bliss" when building SCIP here <https://github.com/google/or-tools/blob/82914bc4f9490722acdce4e82adac0f8e11ff967/cmake/dependencies/CMakeLists.txt#L277> seems to fix the issue for me — Reply to this email directly, view it on GitHub <#4497 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3NLCORJVZB3GNVG5E32KZYFPAVCNFSM6AAAAABVFN4GD6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOBUGM4TQMY> . You are receiving this because you commented.Message ID: ***@***.***>

[pet-mit]

SCIP v920 automatically downloads Bliss (https://github.com/scipopt/scip/blob/74cea9222e98f7351dc4540f562a6dfa4c1d7595/CMakeLists.txt#L150). But it seems that it doesn't anymore, in newer version.
However, current version of SCIP comes with Nauty and Sassy. Maybe they're worth a try?

I'm insisting because in the current state of the build, SCIP does not seem to support quadratic objective problems through MathOpt (under Ubuntu at least).

Here is an example if you want to reproduce it (or you can just try the failing unit tests of gscip):

  auto model = std::make_unique<Model>("quadratic objective");

  std::vector<Variable> x;
  x.push_back(model->AddVariable(0.0, 200.0, false, "x_0"));
  x.push_back(model->AddVariable(0.0, 299.0, false, "x_1"));
  x.push_back(model->AddVariable(0.0, 1e20, false, "x_2"));
  x.push_back(model->AddVariable(0.0, 1e20, false, "x_3"));
  x.push_back(model->AddVariable(-101, 100.0, false, "x_4"));
  x.push_back(model->AddVariable(0.0, 100.0, false, "x_5"));
  x.push_back(model->AddVariable(0.0, 101.0, false, "x_6"));

  model->AddLinearConstraint(x[4] - x[5] + x[6] == 0, "c_0");
  model->AddLinearConstraint(x[0] - x[4] - x[2] == 101, "c_1");
  model->AddLinearConstraint(x[1] + x[4] - x[3] == 299, "c_2");
  model->AddLinearConstraint(0.7 * x[4] <= 89.1, "c_3");
  model->AddLinearConstraint(0.7 * x[4] >= -1100, "c_4");

  model->Minimize(0.005 * x[0] * x[0] + 0.0033444 * x[1] * x[1]);

  SolveArguments args;
  args.parameters.enable_output = true;

  const absl::StatusOr<SolveResult> result =
      Solve(*model, SolverType::kGscip, args);

[lperron]

No it does not, the team is working on it. I would love that when I fetch SCIP, I get a fully functional version. But the reality is that it does not work. Getting papilo to work is a dependency nightmare, automatically configuring everything is complex. Long story short, we are a bit short handed to make it work now. Laurent Perron | Operations Research | ***@***.*** | (33) 1 42 68 53 00 Le mer. 15 janv. 2025 à 16:33, Peter Mitri ***@***.***> a écrit :

…

SCIP v920 automatically downloads Bliss ( https://github.com/scipopt/scip/blob/74cea9222e98f7351dc4540f562a6dfa4c1d7595/CMakeLists.txt#L150). But it seems that it doesn't anymore, in newer version. However, current version of SCIP comes with Nauty and Sassy <https://github.com/scipopt/scip/blob/master/INSTALL.md#building-scip-using-cmake-cmake>. Maybe they're worth a try? I'm insisting because in the current state of the build, SCIP does not seem to support quadratic objective problems through MathOpt (under Ubuntu at least). Here is an example if you want to reproduce it (or you can just try the failing unit tests of gscip): auto model = std::make_unique<Model>("quadratic objective"); std::vector<Variable> x; x.push_back(model->AddVariable(0.0, 200.0, false, "x_0")); x.push_back(model->AddVariable(0.0, 299.0, false, "x_1")); x.push_back(model->AddVariable(0.0, 1e20, false, "x_2")); x.push_back(model->AddVariable(0.0, 1e20, false, "x_3")); x.push_back(model->AddVariable(-101, 100.0, false, "x_4")); x.push_back(model->AddVariable(0.0, 100.0, false, "x_5")); x.push_back(model->AddVariable(0.0, 101.0, false, "x_6")); model->AddLinearConstraint(x[4] - x[5] + x[6] == 0, "c_0"); model->AddLinearConstraint(x[0] - x[4] - x[2] == 101, "c_1"); model->AddLinearConstraint(x[1] + x[4] - x[3] == 299, "c_2"); model->AddLinearConstraint(0.7 * x[4] <= 89.1, "c_3"); model->AddLinearConstraint(0.7 * x[4] >= -1100, "c_4"); model->Minimize(0.005 * x[0] * x[0] + 0.0033444 * x[1] * x[1]); SolveArguments args; args.parameters.enable_output = true; const absl::StatusOr<SolveResult> result = Solve(*model, SolverType::kGscip, args); — Reply to this email directly, view it on GitHub <#4497 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3JOIEOJQUB46AIQZ3T2KZ5U7AVCNFSM6AAAAABVFN4GD6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOBUGQ2DQMA> . You are receiving this because you commented.Message ID: ***@***.***>