ParamILS

ParamILS is a state-of-the-art method for parameter tuning and algorithm configuration.
It has been used in dozens of academic applications to improve state-of-the-art solvers for more then ten hard computational problems (see its 130+ Google scholar citations). ParamILS has also had an impact in several industrial applications. In all these applications, ParamILS has yielded substantial speedups of state-of-the-art solvers for hard combinatorial problems, such as propositional satisfiability (SAT), mixed integer programming (MIP), AI planning, answer set programing (ASP), and timetabling. By doing so, ParamILS has helped several systems win solver competitions.

Latest News

October 10, 2012: For the fifth term running, the exams of UBC's 40.000 students have been scheduled with a timetabler optimized with ParamILS. An early version of the solver was described by Fawcett et al., [2009].

View previous news here.

People

• Frank Hutter, Postdoctoral research fellow (UBC)
• Holger Hoos, Professor (UBC)
• Thomas Stützle, Research Associate (Université Libre de Bruxelles, ULB)
• Kevin Leyton-Brown, Associate Professor (UBC)
• Chris Fawcett, Ph.D. student (UBC): Responsible for maintenance and future feature development for ParamILS.

Papers

Below we list the papers directly pertaining to the ParamILS method. Dozens of papers applying ParamILS to optimize various solvers are listed on the Applications page.

• ParamILS: An Automatic Algorithm Configuration Framework
  Frank Hutter, Holger H. Hoos, Kevin Leyton-Brown, and Thomas Stützle. Journal of Artificial Research 2009, volume 36, pages 267-306.
  [The main ParamILS paper. This paper describes ParamILS and then introduces adaptive capping, the key concept to achieving large speedups for runtime optimization. It then demonstrates that algorithm configuration is practical for a wide range of application domains. It also discusses pitfalls and remedies, as well as advanced concepts, such as meta-configuration (using ParamILS to optimize its own parameters).
  Note: the experiments in this paper for adaptive capping only show a moderate effect because short cutoff times of seconds were used - in applications requiring larger cutoff times (e.g., hours) to solve harder instances, adaptive capping often reduces the required configuration budget by several orders of magnitude. ]
  [pdf] [bibtex]
  


• Automatic Algorithm Configuration based on Local Search
  Frank Hutter, Holger Hoos, and Thomas Stützle - Proc. of AAAI 2007.
  [The conference paper that introduced a first version of ParamILS, demonstrating preliminary promising results.]
  [pdf] [bibtex]


• Parallel algorithm configuration
  Frank Hutter, Holger H. Hoos, and Kevin Leyton-Brown. Proc. of LION 2012.
  [Demonstrated that ParamILS is extremely well suited for parallelization by multiple independent runs: due to its high variance, using 4 parallel ParamILS runs in parallel yielded more than 4-fold speedups, and using 16 cores still yielded close to 16-fold speedups in several cases.
  Note: performing 10 or 25 multiple independent runs had been our practice since as early as 2007, but this paper was the first to formally study the effects of this policy.]
  [pdf] [bibtex]

Data

• Benchmark instances used in our AAAI-07 paper, and parameter files for SAT4J and GLS+
• Algorithms used in our JAIR paper, their parameters and the values we considered for each parameter
• Benchmark instances used in our JAIR paper, and parameter settings found with FocusedILS for each of them

Software

• ParamILS version 2.3.5 [source, executables and examples] [quick start guide]. This is the newest version, and the one you should be using to tune your algorithms. Minor bugfix extending the 2.3.2 bugfix to nondeterministic cases when the user has specified an instance-seed file but failed to specify N correctly. A warning is now printed, and the appropriate values are clamped.
• Old versions can be downloaded here

Licensing

Please send any questions, concerns or comments to Frank Hutter or Chris Fawcett