Crossover operators

These pages compare the crossover operators of moocore with the equivalent operators of pymoo, DEAP, jMetalPy and DESDEO, when they exist. Each page compares the offspring that the packages produce and the time they take.

The KS test results show whether two packages produce offspring with the same distribution. Each package crosses over the same parents 100 000 times, with bounds wide enough that they never bind, so the results do not depend on how each package handles bounds. For every pair of packages and every decision variable, a two-sample Kolmogorov-Smirnov test compares their offspring. Each table shows the smallest p-value over the variables. A p-value below 0.001 means that the distributions differ.

The performance plots time each operator for an increasing number of parents with 30 variables, and for an increasing number of variables with 100 parents. Each package receives the parents in its own data structure, which is created before the timing starts. When a package expects the caller to choose the other parents of DE or PCX, choosing them is part of the time.

• Uniform crossover
• Single-point crossover
• Blend crossover (BLX-α)
• Single arithmetic crossover
• Local crossover
• Bounded exponential crossover (BEX)
• Differential evolution: DE/rand/1/bin
• Differential evolution: DE/current/1
• Parent-centric crossover (PCX)