Replicable Self-Documenting Experiments with Arbitrary Search Spaces and Algorithms

We introduce moptipy, a toolbox for implementing, experimenting with, and applying optimization algorithms. It features mechanisms for executing fully reproducible experiments. Our seeding procedure for random number generators makes our experiments deterministic. Our system creates self-documenting log files that store the algorithm setup, the system configuration, the random seed, the final solutions, and the progress of the optimization algorithm over time. The parallelization and distribution of experiments works on most operating systems and requires no additional synchronization, inter-process-communication, or libraries. moptipy supports both single- and multi-objective optimization with arbitrary search spaces. Time measurements and computational budgets can be based both on wall clock time and objective function evaluations. moptipy is also an educational platform with comprehensive documentation and an accompanying free electronic book introducing the basic concepts of metaheuristics, discussing the implemented algorithms, and showing their performance on basis of actual experimental results.