Posts

Showing posts from October, 2017

The GP.Lab Genetic Programming Kernel (continued)

This second post on our custom Genetic Programming implementation covers selection and genetic operators implemented by the kernel. Selection Strategies The GP kernel implements a generational algorithm and currently supports the three most popular selection modes for choosing parents to produce offspring for the next generation: Tournament Selection , i.e. for a tournament size T randomly pick T programs from the parent population and return the winner of the tournament - the program with the best fitness. The selection method of choice in the GA community as it allows to fine-tune the selection pressure via the tournament size parameter to avoid premature loss of diversity (i.e. fittest programs dominating the population with their copies). Roulette Wheel , a flavor of Fitness Proportionate Selection where the chance of an individual to become a parent is proportional to its fitness, i.e. parents with the highest absolute fitness in the population have a higher chance o