RSG datasets

Each dataset was parameterized by the number of clusters k ∈ {2,10,50}, number of features F ∈ {10,50,200}, and number of samples per cluster Ne ∈ {5,50,100,500,5000}. The generator samples cluster-specific covariance matrices that are symmetric and positive semi-definite, ensuring realistic correlation structures between features. A mixing parameter α was tuned to ensure a range of cluster overlaps and complexity levels.