Dynamic Scoring Functions with Variable Expressions: New SLS Methods for Solving SAT (ONLINE APPENDIX)

All of our instances are available at: http://people.cs.ubc.ca/ ̃davet/papers/sat10-dave-instances.zip For our experiments we split the instance sets into halves: a test set, and a training set. To split the instances we used a stratification strategy to ensure that the test and training set were of approximately equal hardness. As an approximate measure of hardness, we measured the file size of the instance. We first sorted the instances by their file size, and then segmented the instances into pairs of consecutive instances. For each consecutive pair, we randomly placed one instance in the test set, and the other in the training set. Only the instances in the training set were used to determine good parameter settings, and only instances in the test set were used to report experimental results.