AAS 390 Chapter Notes - Chapter 1: Cedalion, Parallel Algorithm, Performance Metric
Document Summary
To avoid duplication of content we only give a high-level overview of our algorithm here and refer to a technical report for details on our methodology (seipp, sievers, and hutter. An extended version of that report is forthcoming. The paper at hand focuses mostly on the con guration setup we used for generating portfolios for the ipc 2014 learning track. Cedalion is our algorithm for automatically con guring se- quential planning portfolios. Given a parametrized planner and a set of training instances, it iteratively selects the pair of planner con guration and time slice that improves the cur- rent portfolio the most per time spent. At the end of each it- eration all instances for which the current portfolio nds the best solution are removed from the training set. The algo- rithm stops when the the total runtime of the added con gu- rations reaches the portfolio time limit (usually 30 minutes) or if the training set becomes empty.