Automated Algorithm Configuration (AAC)

Bioinformatics, and Empirical & Theoretical Algorithmics Laboratory (ß-Lab)
Department of Computer Science
The University of British Columbia


| news |  What is this website about?  |  target algorithms  |  instances


April 20, 2010 First version of this website

What is this website about?

Informally speaking, the automated algorithm configuration (AAC) problem can be stated as follows: given a parameterized target algorithm A, a set of instances P, and a cost metric c, find parameter settings of A that minimize c on P. In practice, we split the benchmark instances into training and test sets (to saveguard against over-tuning for the training instances) and specify a maximal cutoff time (a so-called captime) after which each so-far-unsuccessful run of the target algorithms is terminated.
We refer to procedures for solving the AAC problem as algorithm configuration procedures.

On this website, we provide the building blocks for a collection of AAC problem instances that can be used to benchmark algorithm configuration procedures. The main building blocks for which data is required are (1) target algorithms (including a specification of their parameters as well as wrappers to call them using a unified interface), and (2) benchmark instances of the type of problem the target algorithms solve.

We also provide links to several algorithm configuration procedures.

Target Algorithms and their Parameter Configuration Space

Benchmark Instances

Algorithm configuration procedures

Empirical analysis tools 

Please send any questions, concerns or comments to Frank Hutter