Exploration of Pareto-preserving search-space transformations in multi-objective test functions - Reproducibility

This repository contains the reproducibility files for the paper "Exploration of Pareto-preserving search-space transformations in multi-objective test functions" It is structured as follows: - basefunctions: File containing the definitions of the transformations, used across the different parts of data collection and analysis. - Visualize_transforms: A notebook which can be used to visualize the transformations, used to generate figures 1 and 2 from the paper - Data_collection: This contains the code used to run the experiments (running the algorithms on the different problem instances). - outrot, outab, outab_obj: These 3 folders contain the full set of trajectory data from the experiments. To use them during the processing and visualization, place them in a single folder, making sure to keep their names. - Data_extraction: This is the file used to process the raw data into the used hypervolume measures - Processed_data: This is the processed version of the trajectory data, used for the visualizations - Visualization: This notebook is used to create the figures as shown in the paper, as well as similar figures for other settings - Figures: This contains alternative versions of the figures shown in the paper (e.g. for other functions) # Reproduction steps ## Data collection To re-run the data collection, you should make use of the code in the corresponding directory. You should install the listed requirements, and then use the 'experiment.py' file to run the scripts. Note: since rpy2 is one of the used packages, this requires R to be installed. Within this R environment, the package 'smoof' needs to be available, which can be installed using (in the R-terminal): `install.packages('smoof')` 'experiment.py' uses the RunExperiment package, so it can be run in parallel or sequentially based on the chosen settings (see the readme in the RunExperiments subfolder for more information). By default, it will run the BetaCDF based settings, change the main-function to run the rotation version instead. Running the data collection will save the results to the folder specified in 'experiment.py', so this should be set to the desired location. ## Data processing Since large parts of our analysis use the hypervolume measure rather than the trajectory itself, we provide a script for the processing of the data files. Note that the used directories are set as global variables at the start of the file, so these should be modified to the respective folders. ## Visualization For the visualization, there are two parts: - For the first 2 figures showing the transformation on a simple 2D grid, you can use the Vizualize_transforms notebook - For the remaining figures (including additional versions), you can use the Visualizations notebook. This notebook is anotated to provide further instructions. ## Questions In case of questions, please dont hesitate to contact us at: Anoynmized during review process