Evaluating random forest models for irace

Automatic algorithm configurators can greatly improve the performance of algorithms by effectively searching the parameter space. As algorithm con.guration tasks can have large parameter spaces and the execution of candidate algorithm configurations is o.en very costly in terms of computation time, further improvements in the search techniques used by automatic con.gurators are important and increase the applicability of available con.guration methods. One common technique to improve the behavior of search methods when evaluations are computationally expensive are surrogate model techniques. .ese models are able to exploit the scarce available data and help to direct the search towards evaluating the most promising candidate con.gurations. In this paper, we study the use of random forests models as surrogate models in irace, a .exible automatic con.guration tool based on iterated racing that has been successfully applied in the literature. We evaluate the performance of the random forest model using di.erent se.ings when trained with data obtained from the irace con.guration process and we evaluate their performance under similar conditions as in the con.guration process. .is preliminary work aims at providing guidelines for the incorporation of random forest to the con.guration process of irace.