Implementing evolution

New classes

• Each critter has an instance of the Genome class. This contains an array of booleans representing the actual Genome and methods for decoding the Genome, that is, extracting information from it about the value of particular actions in particular states, and for getting the best action for a given state.
• An instance of the class Evolver is reponsible for controlling evolution. It has methods for implementing mating between critters.

Particular methods

• decide() in Brain calls getBestTrait() in the critter's genome. The genome stores a value for each combination of action, touch sensor direction, and texture. Since a state is a combination of several (for example, 6) direction-texture combinations, the value of an action in a state is the sum of the values for the individual positions and textures. It is these sums that are compared in getBestTrait().
  
• When one critter attempts to move to a cell with another critter, World decides whether mating is possible, and if it is, it calls mate() in Evolver with the two parent critters as arguments.
  • mate() creates two new genomes by performing crossover on the genomes of the two parents.
    
  • mate() then mutates the new genomes.
  • mate() finds the two weakest critters in the population, and reincarnates them by replacing their genomes with the two new ones and restoring their strength to the initial value.

© 2004. Indiana University and Michael Gasser. www.indiana.edu/~gasser/Q530/Notes/ga-smarts.html 3 Nov 2004