EDACC Automated Algorithm Configuration
Experiment Design and Administration for Computer Clusters for SAT Solvers. See http://sourceforge.net/projects/edacc/ for the EDACC project.
This tool can be used to optimize the empirical performance of algorithms on problem instances automatically by adjusting free parameters of the algorithms.
Compile by running "ant" in the top level directory containing "build.xml". An executable Java JAR file will be put into the folder "dist/aac/"
How to use
- The tool requires a network connection to an EDACC database and a configuration experiment that was set up using the EDACC GUI. Computation clients should also be started and assigned to the configuration experiment using the GUI.
- Write a configuration file that tells the tool which optimisation procedure to use, the database connection details, optimisation objective etc. See the "contrib/" folder for examples.
- Start the process via "java -jar AAC.jar configfile"
- Java (JRE) 6 or 7
- some procedures need R (tested with version 2.15), JRI and certain R packages (they will tell which at runtime)
The SMBO/Default_SMBO optimization procedure requires a working R installation with the Java-R-interface "JRI" and some R packages.
- JRI for R can be installed in R itself by using the command install.packages("rJava"). See also http://www.rforge.net/JRI/ and the FAQ at the bottom for adjusting environment variables (in particular LD_LIBRARY_PATH) so Java can access JRI.
- The R packages "asbio", "cluster" and "survival" can be installed in an R shell via install.packages("pkgname")