The general challenges of experimenting with simulation algorithms have already been discussed in chapter 3 (p. 93). This chapter builds up on that by briefly outlining how these challenges are currently met by the experimentation layer of JAMES II (sec. 7.1). Section 7.2 introduces a simple yet powerful algorithm selection technique that can be plugged into the experimentation layer. It both explores and exploits algorithm performance at the processing time of a simulation problem, but without relying on the performance database or the data analysis methods presented in chapter 5 (p. 153) and chapter 6 (p. 177). Then, section 7.3 shows how combining this technique with additional components facilitates the execution of large-scale experiments to investigate the runtime performance of simulation algorithms. A calibration method that yields significant speed-up for such experiments is also introduced (sec. 7.3.2).
KeywordsAlgorithm Selection Portfolio Selection Simulation Algorithm Experimentation Methodology Benchmark Model
Unable to display preview. Download preview PDF.