RCC-Bench: Data and Code for Type-I Robustness–Convergence Conflicts in Multi-objective Optimization

Description

This dataset contains the run-level records, final decision sets, numerical reference bundles, sensitivity analyses, RCC construction audits, quality-control records, and public RCC-Bench and Type-I evaluation code supporting the associated article.

Steps to reproduce

The dataset was generated through numerical benchmark experiments implemented in Python 3.11. Each problem--implementation combination was run with 30 independent search seeds and a population size of 100. The primary experiment used five decision variables and a budget of 10,000 complete objective-vector evaluations per run. Additional matrices varied the evaluation budget from 5,000 to 30,000 and the number of decision variables from 5 to 20. For each completed search run, the final decision set was evaluated using the same Type-I post-processing procedure. This procedure uses common scrambled Sobol samples in feasible truncated perturbation neighborhoods, fixed normalization, and boundary-specific reference sets. The deposited records include final decision sets, search-call accounting, Type-I indicator values, sensitivity analyses, numerical reference bundles, and RCC construction audits. The package also provides the RCC-Bench implementation and the common Type-I evaluator, together with the Python dependencies required for these components. Source snapshots for the six comparison algorithms are not included.

Institutions

• Tongji University

  Shanghai, Shanghai

Categories

Funders

• National Natural Science Foundation of China

  Beijing, Beijing

  Grant ID: 62373286

Licence