Page 122 - 49A Field Guide to Genetic Programming

108 11 GP Theory and its Applications

Figure 11.2: Plots of the evolution average size over 500 generations for
multiple runs of the 6-MUX problem with various forms of covariant parsi-
mony pressure. The “Constant” runs had a constant target size of 150. In
the “Sin” runs the target size was sin((generation + 1)/50.0) × 50.0 + 150.
For the “Linear” runs the target size was 150 + generation. The “Limited”
runs used no size control until the size reached 250, then the target was held
at 250. Finally, the “Local” runs used c = Cov(`, f)/ Var(`), which allowed
a certain amount of drift but still avoided runaway bloat (see text).