Running Experiments with Confidence and Sanity

摘要

Analyzing data from large experimental suites is a daily task for anyone doing experimental algorithmics. In this paper we report on several approaches we tried for this seemingly mundane task in a similarity search setting, reflecting on the challenges it poses. We conclude by proposing a workflow, which can be implemented using several tools, that allows to analyze experimental data with confidence. The extended version of this paper and the support code are provided at https://github.com/Cecca/running-experiments.