PD-SOS: the importance of being constrained

This release is a permanent version of the current SOS platform containing all classes for reproducing the results described in section 4 of the article "The importance of being constrained: dealing with infeasible solutions in Differential Evolution" located as indicated below and in the original ECJ submission according to the reproducibility guidelines of this CFP for the special issue on Reproducibility in Evolutionary Computation.

Pre-processing code for section 5:

The entire experimental setup of this section is coded the class 'ECJ2022TIOBR' in the 'mains.AlgorithmicBehaviour' package. This class contains the main method and can be run in SOS for generating the raw data on f0, thus reproducing the results analysed in section 4 of the main article.

NB: we make available all data generated for this study in Zenodo. These are subsequently processed with the scripts in this GitHub repo.

Objective function code for section 5:

The f0 function is implemented in the "Infeasibility and Structural Bias" (ISB) suite of SOS, whose code is available in the class 'ISBSuite' of the 'benchmarks' package.

This suite and other methods (mentioned in the next block) were added to extend SOS - see "Population Dynamics" PD-SOS Caraffini, F. 2021 - to investigate specific algorithmic behaviours.

Algorithm code for section 5:

The code for the algorithms under investigation in this section is available in the 'DE_TIOBR' class inside the package 'algorithms.AlgorithmicBehaviour'.

This class implements a modular DE similar to the one already present in the algorithms folder of SOS, but equipped with several methods from PD-SOS Caraffini, F. 2021 and further novel ones for e.g. recording POIS with a given SDIS and measuring the cosine similarity between search directions - for information on acronyms please see the original manuscript and relevant previous articles as e.g. [Kononova, A. V. et al. 2021, von Stein, B. et al. 2021, Vermetten, D. et al. 2021, Caraffini, F. et al. 2019]

References to the main article and relevant repos

• Further processing code https://github.com/Dvermetten/DE_TIOBR
• Dataset https://doi.org/10.5281/zenodo.5900706
• Figures www.doi.org/10.6084/m9.figshare.18319394.v1

PD-SOS

The Stochastic Optimisation Software (SOS) is a research-oriented software platform for Metaheuristic Optimisation (Stochastic Optimisation).

This release of SOS contains new features for the analysis of heuristic for optimisation in terms of structural bias and tendency in generating infeasible solutions. These include a new benchmark suite and dedicated packages already equipped with mechanisms allowing for tracking the internal dynamics of the candidate solutions in population-based algorithms. These Population Dynamics (PD) packages will be further extended in the future to achieve a better understanding of the working mechanism of widely used heuristics for optimisation.

For further details please visit www.tinyurl.com\FabioCaraffini-SOS

some relevant articles are listed below:
-Structural bias in differential evolution: A preliminary study (2019) AIP Conference Proceedings, DOI: 10.1063/1.5089972 [link]
-Infeasibility and structural bias in differential evolution (2019), Information Sciences, DOI: 10.1016/j.ins.2019.05.019 [link]
-Can Single Solution Optimisation Methods Be Structurally Biased? (2020), IEEE WCCI, DOI: 10.1109/CEC48606.2020.9185494 [link]
-Can Compact Optimisation Algorithms Be Structurally Biased? (2020), PPSN 2020, DOI: 10.1007/978-3-030-58112-1_16 [link]

The Stochastic Optimisation Software (SOS) platform

The Stochastic Optimisation Software (SOS) is a research-oriented software platform for Metaheuristic Optimisation (Stochastic Optimisation). For further details please visit www.tinyurl.com\FabioCaraffini-SOS

Stochastic Optimisation Software (SOS) platform

The SOS platform facilitates the design of optimisation algorithms such as (both stochastic and deterministic) metaheuristics for (but not limited to) real-valued single objective problems thanks to:

-the possibility of easily combining together already implemented algorithmic components, such as several variation operators (e.g. crossover, mutation, etc. ) and selection mechanisms from Evolutionary Computation, Memetic Algorithms/Computing and Hyper-Heuristics ;
-the availability of several ancillary methods for manipulating matrices, performing mathematical operations, handling the computational budget, parameter tuning, executing and comparing between algorithms and versions of the same algorithm;

and it helps to produce and interpreting results thanks to:

• the presence of several ready-to-use popular benchmarks suites (e.g. BBOB2010--2019, CEC2015--2017, popular functions), examples of published real-world applications, and benchmark real-world problems;
  -the presence of ancillary modules executing algorithms, over the aforementioned (or newly implemented) problems, also in Multi-thread to accelerate the production of numerical results:
  -the availability of methods collecting results and creating PDF and LaTeX tables with the outcome of several statistic tests and classic AVG=- STD comparison (see http://www.cse.dmu.ac.uk/~fcaraf00/NumericalResults/)

SOS is mean for stochastic optimisation but it is not limited to it: deterministic metaheuristic algorithms can be implemented and compared against a large number of algorithms already present in this repository.

Examples of studies performed via SOS are:

· Infeasibility and structural bias in differential evolution (2019), Information Sciences, DOI: 10.1016/j.ins.2019.05.019
· HyperSPAM: A study on hyper-heuristic coordination strategies in the continuous domain (2019), Information Sciences, DOI: 10.1016/j.ins.2018.10.033
· Compact Optimization Algorithms with Re-Sampled Inheritance (2019), in LNCS, DOI: 10.1007/978-3-030-16692-2_35
· Improving (1+1) covariance matrix adaptation evolution strategy: A simple yet efficient approach (2019), AIP Conference Proceedings, DOI: 10.1063/1.5089971
· Structural bias in differential evolution: A preliminary study (2019) AIP Conference Proceedings, DOI: 10.1063/1.5089972
· A study on rotation invariance in differential evolution (2018), Swarm and Evolutionary Computation, DOI: 10.1016/j.swevo.2018.08.013
· Rotation Invariance and Rotated Problems: An Experimental Study on Differential Evolution (2018), LNCS including LNAI, DOI: 10.1007/978-3-319-77538-8_41
· Large scale problems in practice: The effect of dimensionality on the interaction among variables (2017), LNCS including LNAI, DOI: 10.1007/978-3-319-55849-3_41
· Cluster-Based Population Initialization for differential evolution frameworks (2015), Information Sciences, DOI: 10.1016/j.ins.2014.11.026
· Continuous parameter pools in ensemble differential evolution (2015), IEEE SSCI'15, DOI: 10.1109/SSCI.2015.216
· Multicriteria adaptive differential evolution for global numerical optimization (2015), Integrated Computer-Aided Engineering, DOI: 10.3233/ICA-150481
· Structural bias in population-based algorithms (2015), Information Sciences, DOI: 10.1016/j.ins.2014.11.035