On the Structural (Dis)Agreement of Landscape Representations in Black-Box Optimization

Repository Structure Overview This repository contains datasets, and experimental results for analyzing algorithm behavior using similarity measures, coverage matrices, and dimensionality reduction techniques. The structure is organized to clearly separate input data, clustering algorithms, and analysis outputs. clustering_ds/ This folder contains the datasets prepared specifically for clustering-based analyses, where algorithm behaviors or feature representations are grouped and compared. Subfolders dataset_DE/Contains the algorithms that were being used for the problems. Differential Evolution (DE) algorithms. These datasets are used to analyze similarity, coverage, and clustering behavior specific to DE. dataset_PSO/Contains the algorithms that were being used for the problems. Particle Swarm Optimization (PSO) algorithms.Structured analogously to the DE datasets to allow direct comparison. These datasets are typically used as inputs for: similarity computations (e.g., cosine similarity), clustering evaluation, dimensionality reduction (PCA, t-SNE, UMAP). datasets/ This folder contains the feature representations of optimization problems used throughout all experiments. Each representation captures different structural and landscape properties of the problems and serves as the input for clustering, similarity analysis, coverage matrix computation, and visualization. The datasets include the following representations: ELA (Exploratory Landscape Analysis) DeepELA TransOptAS DOE2Vec results/ This folder contains all experimental outputs, organized by analysis type, algorithm inclusion, and parameter settings. 1. Cosine Similarity Results cosine_similarity_ALG_results_ds1_ds2_best_silhouette_5_500/Results of cosine similarity analysis between dataset pairs (ds1, ds2), using configurations selected by the best silhouette score (from the grid search). cosine_similarity_DE/ – Results specific to DE algorithms cosine_similarity_PSO/ – Results specific to PSO algorithms cosine_similarity_results_ds1_ds2_best_silhouette_5_500/Aggregated cosine similarity results independent of algorithm type. 2. Coverage Matrix Grid Search Results grid_search_coverage_matrix_-1_1_5_500_no_percent_all_pairs/Grid search results for coverage matrices over the range [-1, 1], computed for all algorithm pairs. DE_included/ coverage_matrix_DE/ – Coverage matrices including DE algorithms PSO_included/Coverage matrices including PSO algorithms grid_search_coverage_matrix_-1_1_5_500_without_percent_whole_functions_DE_algorithms/Coverage matrices computed on whole functions, focusing on DE algorithms. 4. Dimensionality Reduction Visualizations gridsearch_results_scaling_-1_1_5_500_merged_visualizations_git/Combined and cleaned visual outputs from grid search experiments These folders contain low-dimensional projections of high-dimensional feature representations. PCA/ PCA_2D/ – 2D PCA projections PCA_3D/ – 3D PCA projections t-SNE/ t-SNE_2D/ – 2D t-SNE visualizations t-SNE_3D/ – 3D t-SNE visualizations uMAP/UMAP-based visualizations. 5. Additional Result Sets gridsearch_results_scaling_-1_1_5_500_step5_last_one/Results and metrics computed during the gridsearch for choosing the best method. hcv_metrics_based_on_best_grid_5_500/Homogeneity, completeness , and v-measure (HCV) metrics computed using the best grid configuration. overlap_results/Results analyzing overlap between pairs between different feature representations or algorithms results_coverage_matrices_DE/Final and aggregated coverage matrix results specifically for DE algorithms.