📂Optimization

What is a Tournament in Genetic Algorithms?

Definition ¹

In genetic algorithms, a tournament is a method of natural selection that directly compares the fitness among some individuals, eliminates individuals with lower fitness, and passes the rest on to the next generation.

Explanation

A tournament can be implemented, for example, by randomly pairing up individuals two at a time from the entire solution set and, for each pair, deleting the individual with lower fitness. The selection pressure applied in this way naturally allows individuals with higher fitness to survive, without any complex means or hyperparameter tuning. Of course, various variations are possible, such as letting the inferior individual survive with a certain probability in the fitness comparison, or drawing multiple individuals at once to select several survivors.

Elitism grants individuals with a fitness above a certain level the privilege of always surviving, and they can keep holding on unless there is a change that surpasses them. In contrast, a tournament is a method where upsets can occur, just as they actually do in sports and the like: even a fairly fit individual can be eliminated at any time if it happens to face an individual with higher fitness than itself.