Objective Find an algorithm (i.e., solver) that locates the minimum of a ten-parameter quadratic function (9). Terminal set See (8). Function set See (7). Initial population Ramped half-and-half (tree depth of 5) (Koza, 1992). Population size 200. Fitness An average cost value over ten runs, each time from a different initial simplex and with a different displacement vector. Punish lengths of more than 2,000 characters by multiplying the fitness value by 100. Selection The best 90% of the population automatically pass to an intermediate population (elitist selection). Also, 20% of the population is selected using a tournament selection (tournament size of 2) (Poli et al., 2008) to act as parents in crossover operations. Probability for a crossover to occur at a terminal 10%. Tree depth limit No. Mutation No. Decimation In 50th, 150th, and 350th generations. All the individuals whose fitness is less than 0.1% worse than that of the closest better individual are deleted from the population and replaced with new ones. Edit In every 30th generation, all parts that have no effect are removed from solvers in the current population. Termination After 400 generations.
 Objective Find an algorithm (i.e., solver) that locates the minimum of a ten-parameter quadratic function (9). Terminal set See (8). Function set See (7). Initial population Ramped half-and-half (tree depth of 5) (Koza, 1992). Population size 200. Fitness An average cost value over ten runs, each time from a different initial simplex and with a different displacement vector. Punish lengths of more than 2,000 characters by multiplying the fitness value by 100. Selection The best 90% of the population automatically pass to an intermediate population (elitist selection). Also, 20% of the population is selected using a tournament selection (tournament size of 2) (Poli et al., 2008) to act as parents in crossover operations. Probability for a crossover to occur at a terminal 10%. Tree depth limit No. Mutation No. Decimation In 50th, 150th, and 350th generations. All the individuals whose fitness is less than 0.1% worse than that of the closest better individual are deleted from the population and replaced with new ones. Edit In every 30th generation, all parts that have no effect are removed from solvers in the current population. Termination After 400 generations. See text for details.