A1728
Title: Monte Carlo sampling benchmark suite
Authors: Zeyu Ding - TU Dortmund (Germany) [presenting]
Abstract: A benchmark suite is introduced to assess the quality of Monte Carlo (MC) samples. Our suite enables quantitative comparisons using metrics like the sliced Wasserstein distance and other statistical measures. These are applied to both independent and identically distributed (iid) and correlated samples generated by MC methods, such as Markov Chain Monte Carlo or Nested Sampling. By gathering test statistics from repeated comparisons, we evaluate MC sampling quality. The suite includes diverse target functions of varying complexity and dimensionality, providing a flexible platform for testing sampling algorithms. Implemented as a Julia package, it allows users to select and extend test cases and metrics as needed. Users can run external sampling algorithms on these functions, input their samples, and obtain metrics that compare their quality to iid samples from our package. This standardized approach offers clear, quantitative measures of sampling quality, aiding researchers in validating and improving sampling methods.
