DynamoRep: Trajectory-Based Population Dynamics for Classification of Black-box Optimization Problems

Trajectories of the algorithms Differential Evolution (DE), Evolutionary Strategy (ES), Genetic Algorithm (GA), Covariance Matrix Adaptation Evolution Strategy (CMAES), run on the first 999 problem instances of each 3D problem class in the BBOB benchmark.

We use the pymoo python library to run the DE, GA, CMAES, and ES algorithms on the problem instances from the BBOB benchmark. We run the algorithms in a fixed-budget scenario with a population size of $10d$. Since we are using three-dimensional problems, the population size is 30. 

Each algorithm is run 5 times with different random seeds (2,150,500,750,1000). The algorithm trajectories are saved in the files named according to the following template: "{algorithm_short_name}_dim_{dimension}_seed_{s}.csv". 

We also include 4 files which contain the DynamoRep features calculated for the algorithm trajectories. These are the files "GA_seeds_2-150-500-750-1000", "DE_seeds_2-150-500-750-1000", "CMAES_seeds_2-150-500-750-1000" and "ES_seeds_2-150-500-750-1000"