SCIP++ refs/heads/main
 
Loading...
Searching...
No Matches
SCIP++: A C++ wrapper for SCIP

CI Status Coverage Doxygen Conan Center

SCIP++ is a C++ wrapper for SCIP's C interface. It automatically manages the memory, and provides a simple interface to create linear expressions and inequalities.

Usage

The documentation can be found at https://scipopt.github.io/SCIPpp/

Here is a simple example where we create a new model, add two variables, add a linear inequality as constraint, and ask SCIP to solve the maximization problem.

#include <scippp/model.hpp>
using namespace scippp;
int main() {
Model model("Simple");
auto x1 = model.addVar("x_1", 1);
auto x2 = model.addVar("x_2", 1);
model.addConstr(3 * x1 + 2 * x2 <= 1, "capacity");
model.setObjsense(Sense::MAXIMIZE);
model.solve();
}
A SCIP optimization model.
Definition: model.hpp:39
C++ wrapper for SCIP.

Model Creation

A model can be created

  • without an existing SCIP environment,
  • with an existing SCIP environment where all default plugins should be added to, and
  • with an existing SCIP environment where no additional plugins should be added to.
Model m1("ModelWithoutExistingSCIPEnvironment");
SCIP* scip2;
SCIPcreate(&scip2);
Model m2("ModelWithExistingSCIPEnvironmentWhereDefaultPluginsWillBeAdded", scip2);
SCIP* scip3;
SCIPcreate(&scip3);
SCIPincludeDefaultPlugins(scip3);
Model m3("ModelWithExistingSCIPEnvironmentWhereNoPluginsWillBeAdded", scip3, false);

Adding Variables

Variables can be added

  • one at a time,
  • multiple in a vector, and
  • multiple when the number is known at compile time.
Model model("Example");
auto x = model.addVar("x");
auto vec = model.addVars("x_", 42);
const auto& [x0, x1] = model.addVars<2>("x_");

When adding multiple variables simultaneously to the model, they all have a coefficient of zero in the objective function by default. This can be changed to one by scippp::COEFF_ONE:

const auto& [x1, x2] = model.addVars<2>("x_", COEFF_ONE);

For the coefficient, any object providing an index operator can be used:

double operator[](std::size_t index) const

Adding Constraints

Linear inequalities can be added to the model. They can be built from linear expressions:

const auto& [x0, x1, x2, x3] = model.addVars<4>("x_");
LinExpr sum1;
sum1 += 42 * x0;
sum1 += x1;
model.addConstr(sum1 <= 0.5, "constraint1");
LinExpr sum2 = x1 + x2;
model.addConstr(sum2 == 1.25, "constraint2");
model.addConstr(1 <= x2 + 2 * x3, "constraint3");
Represents a linear combination of variables plus a constant term.
Definition: lin_expr.hpp:13

Setting Parameters

The namespace scippp::params contains all parameters shown https://www.scipopt.org/doc/html/PARAMETERS.php in the header parameters.hpp. They are strongly typed, so that no string can be set as the value for a parameter expecting an integer. Instead of using the predefined parameters, one can also use scippp::params::Param<T> directly.

model.setParam(params::LIMITS::MAXSOL, 1);
model.setParam(params::DISPLAY::VERBLEVEL, 0);
model.setParam(params::Param<bool>("write/printzeros"), true);
Stores the argument type and string representation of a parameter.
Definition: param.hpp:14

The optimization goal can be changed by:

model.setObjsense(Sense::MAXIMIZE);

Accessing a Solution

A model can be asked for the status and the number of solutions. A variable can be asked for its value in a given solution as

  • floating point number via getSolVal(sol).
  • integer via getSolValAsInt(sol).
  • long integer via getSolValAsLongInt(sol), and
  • it can be checked for zero via isZero(sol).
const auto& [x0, x1] = model.addVars<2>("x_");
model.solve();
if (model.getNSols() > 0 && model.getStatus() == SCIP_STATUS_OPTIMAL) {
Solution sol { model.getBestSol() };
cout << "x0 + x1 =" << x0.getSolVal(sol) + x1.getSolVal(sol) << endl;
}
Wrapper for a SCIP solution.
Definition: solution.hpp:13

IO

A model can be written to file via Model::writeOrigProblem if a std::filesystem::directory_entry is given as argument. If it is just a string representing a file extension, it is written to standard output.

Numerics

The model exposes

  • SCIPepsilon via epsilon(),
  • SCIPround via round(double), and
  • SCIPisZero via isZero(double)

Access Solving Statistics

Use the Statistics<T> objects from the header solving_statistics.hpp to access solving statistics in a type-safe way:

...
model.solve();
auto pb { model.getSolvingStatistic(statistics::PRIMALBOUND) };

Features Not Yet Supported

For features not yet supported by SCIP++, one can access the underlying raw SCIP object via

SCIP* scip = model.scip();

Build

Without Conan

We use Conan as package manager. That is not required! As long as find_package(scip CONFIG REQUIRED) (and find_package(Boost CONFIG REQUIRED) for the tests) work(s), any kind of dependency management system can be used.

Build and install:

cmake .
make ScipPP
make install

Build and run tests:

cmake -DBUILD_TESTS=ON .
make tests
./test/tests

With Conan v2 and CMake v3.19 or later

Build and install:

conan install .
cmake --preset conan-release .
cmake --build build/Release --target ScipPP
cmake --install build/Release

Build and run tests:

conan install -o with_tests=True .
cmake --preset conan-release .
cmake --build build/Release --target tests
build/Release/test/tests

If your setting of OS, compiler, C++ or stdlib version is one where conan-center does not host pre-compiled binaries, add --build=missing when you run conan install. The dependencies will then be built from source (don't worry, they will be available only for projects using conan, they do not interfere with versions you might already have installed on the system). So, when you see an error message like

ERROR: Missing prebuilt package for 'bliss/0.77', 'boost/1.81.0', 'bzip2/1.0.8', 'gmp/6.2.1', 'libbacktrace/cci.20210118', 'scip/8.0.3', 'soplex/6.0.3', 'zlib/1.2.13'
Check the available packages using 'conan list bliss/0.77:* -r=remote'
or try to build locally from sources using the '--build=missing' argument

change the install-command to

conan install --build=missing .

With Conan v2 and CMake v3.18 or earlier

When CMake presets are not support, use the toolchain file that conan generates.

Build and install:

conan install .
cmake . -G "Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=./build/Release/generators/conan_toolchain.cmake -DCMAKE_POLICY_DEFAULT_CMP0091=NEW -DCMAKE_BUILD_TYPE=Release
make ScipPP
make install

Build and run tests:

conan install -o with_tests=True .
cmake . -G "Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=./build/Release/generators/conan_toolchain.cmake -DCMAKE_POLICY_DEFAULT_CMP0091=NEW -DCMAKE_BUILD_TYPE=Release
make tests
./build/Release/test/tests

Utils

Use gen_constexpr_parameters to transform all SCIP parameters into constexpr scippp::params::Param<T> objects which can be added to the parameters.hpp header.

Use extract_solvingstats to transform all SCIP methods that access solving statistics into static const scippp::statistics::Statistic<T> objects which can be added to the solving_statistics.hpp header.

Maintainer

This project is maintained by Ivo Hedtke ivo (dot) hedtke (at) dbschenker (dot) com.

Code of Conduct

SCIP++ follows the Contributor Covenant Code of Conduct v2, see code_of_conduct.md.

Contributor License Agreement

This project does not use a CLA.

License

SCIP++ is licensed under the Apache-2 license, see [LICENSE](LICENSE).