Skip to content

alexzucca90/feasibilityjump

 
 

Repository files navigation

Feasibility Jump

This repository contains the reference C++ implementation of the Feasibility Jump, which is a primal heuristic for mixed-integer linear programming.

See the open access paper for details about the algorithm.

Usage

To build the Feasibility Jump reference implementation with FICO Xpress integration, Xpress must be installed and the $XPRESSDIR environment variable must be set to the directory where Xpress is installed. To build, run:

make

This should produce the executable xpress_fj, which solves MIP problems given in .mps format. Its command line interface is:

Usage: xpress_fj [--timeout|t TIMEOUT] [--save-solutions|-s OUTDIR] [--verbose|-v] [--heuristic-only|-h] [--exponential-decay|-e] [--relax-continuous|-r] INFILE

For example, to use the FJ heuristic alone, and save solutions to the current directory, run:

./xpress_fj -h benchmark/instances/miplib2017benchmark/academictimetablesmall.mps --save-solutions .

To run the integrated solver where FJ solutions are given to Xpress, run:

./xpress_fj --timeout 60 benchmark/instances/miplib2017benchmark/academictimetablesmall.mps --save-solutions .

Benchmark

The run_benchmark.sh script will reproduce results used in the paper.

The script assumes that you are using Linux and that the $XPRESSDIR environment variable is set. It then does the following:

  • Builds both xpress_fj and a baseline solver xpress_baseline that only calls the Xpress solver with standard parameters.
  • Runs each of the two solvers on 240 MPS instances from the MIPLIB 2017 benchmark set, which needs to be placed (as .mps files) in the directory benchmark/instances/miplib2017benchmark. Timeouts of both 60 and 600 seconds are used.
  • Produces an HTML file that presents the results in benchmark/report.html.

About

Feasibility Jump

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 79.5%
  • Python 13.6%
  • HTML 5.6%
  • Other 1.3%