CpoFzn

CpoFzn is a CP Optimizer solver interface for MiniZinc. Through CpoFzn, models written in MiniZinc can be solved by IBM ILOG CPLEX CP Optimizer.

Getting Started

To get MiniZinc running with CpoFzn, follow these steps.

Prerequisites

Unless you want to use a development version of CpoFzn, we suggest that you use one of versioned tarballs, which can be found here.

You will need to have CPLEX Optimization Studio 20.1 installed. If you don't have it installed already:

• If you are an academic, you can access CPLEX Optimization Studio for free here,
• If you are a commercial user, you can find purchase options here

Download and install MiniZinc from here (release 2.4.2 or later).

You will also need some additional tools.

• A C++ compiler of your choice,
• CMake version 3.5 or later,
• Flex version 2.6.4 or later,
• Bison version 3.0.4 or later.

On UNIX variants most of these can be installed using a package tool like dnf, yum, apt-get or brew.

On Windows, you will most probably need to install a compiler such as Visual Studio. Binaries for Flex and Bison can be found here. Binaries for CMake can be found here.

Building the cpofzn (or cpofzn.exe) executable

We use CMake as a cross-platform build-generator tool. CMake does not build the project, but generates the files needed by your build tool (GNU make, Ninja, Visual Studio, Eclipse, etc). Please refer to CMake documentation for details here.

In most cases CMake detects the right build tool for your platform automatically (assuming that you have some build tool and C++ compiler installed). So the following steps should be enough:

• Open shell in the root folder of this project.
• Type "cmake ." or "cmake -DCPLEX_STUDIO_DIR=<path_to_your_installation> ."
• Type "cmake --build ."

Alternatively you may use some cmake GUI frontend, e.g. cmake-gui. CMake needs to know the installation directory of IBM CPLEX Optimization Studio. If it is not detected automatically, you can specify using -DCPLEX_STUDIO_DIR as mentioned above.

If you are on a Windows system, you may find the following BAT script useful.

   SET        CMAKE_EXE=<Path of the cmake.exe file>
   SET  FLEX_EXECUTABLE=<Path of the win_flex.exe file>
   SET BISON_EXECUTABLE=<Path of the win_bison.exe file>
   SET CPLEX_STUDIO_DIR=<Path of the CPLEX installation directory>
   REM Build code to compile
   "%CMAKE_EXE%" -D "FLEX_EXECUTABLE=%FLEX_EXECUTABLE%" ^
                 -D "BISON_EXECUTABLE=%BISON_EXECUTABLE%" ^
                 -D "CPLEX_STUDIO_DIR=%CPLEX_STUDIO_DIR%" ^
                 -D CMAKE_GENERATOR_PLATFORM=x64 .
   REM Compile 
   "%CMAKE_EXE%" --build . --clean-first --config Release

The resulting cpofzn.exe file can be found in the Release sub-directory. You can safely ignore any warnings issued by your compiler. They are harmless.

Solver registration

Before solving a MiniZinc model using CP Optimizer, it is necessary to first register CpoFzn in your MiniZinc installation. To do that:

1. During the build of the cpofzn executable a file named com.ibm.cpo.msc was generated. This file contains information about locations of files needed by Minizinc.
2. If you open com.ibm.cpo.msc with an editor, you will see the fields executable and mznlib. Should you move the referenced executable or the mznlib directory somewhere else, then you should change the relevant field(s) in the com.ibm.cpo.msc file accordingly. Note that the paths should always be absolute.
3. Copy the file com.ibm.cpo.msc into one of the directories searched by MiniZinc for solver configuration files (see the MiniZinc documentation):
   • $HOME/.minizinc/solvers on Linux and Mac OS
   • the share/minizinc/solvers/ sub-directory of your MiniZinc installation,
   • in any directory listed in the MZN_SOLVER_PATH environment variable.

Usage

Once registered in MiniZinc, CPOptimizer appears as one of the solvers available in MiniZinc. In MiniZinc IDE, CPOptimizer can be chosen from the list of solvers. Additionally, from the command line you can use:

minizinc --solver CPOptimizer

Known issues

Cycles in defines_var

Sometimes MiniZinc generates circular defines_var annotations during flattening. Such circular definitions are refused by cpofzn with errors like this one:

Error: /tmp/mznfileOB3mKO.fzn:23: There is cycle in the model: expression 'X_INTRODUCED_10_' is using itself as a subexpression.
/tmp/mznfileOB3mKO.fzn:988:    .. The cycle contains mutable expression 'X_INTRODUCED_1046_'.
/tmp/mznfileOB3mKO.fzn:43:    .. The cycle contains mutable expression 'X_INTRODUCED_34_'.
/tmp/mznfileOB3mKO.fzn:908:    .. The cycle contains mutable expression 'X_INTRODUCED_913_'.

The problem is not in cpofzn but in MiniZinc itself.

Note about scheduling models

CP Optimizer uses optional interval variables to model tasks in scheduling however MiniZinc uses integer variables (e.g. to represent start times). For this reason, when a MiniZinc variable is used in a scheduling constraint (in particular cumulative) then cpofzn creates an interval variable that is synchronized with the original integer variable. Therefore scheduling models translated from MiniZinc are sub-optimal and it is recommended to use native CP Optimizer model instead. Also optional types (var opt int) are transformed into non-optionals at MiniZinc level. This is a second reason why the generated model is suboptimal.

Contact

You can post your questions on the IBM Optimization forum here under the "Discussion" tab.

Contributors

Some of this code is based on the FlatZinc parser skeleton by Nick Nethercote and Julien Fischer delivered with G12 MiniZinc Distribution version 1.6.

License

Licensed under the Apache License, version 2.0.