Further work

Several things could be done to speed up the performance of the simulation:

• Better implement the garbage collection and the memory allocation routines. One does not need to constantly free and malloc memory that probably will be recycled again. An intelligent stack approach to keeping already malloc'ed memory will speed up performance, but will not be so ``nice'' to the OS.
• Use the nodecount to reduce the amount of redundant computation. Currently, each member of the population has its nodecount recomputed every generation. If this information were only computed in the first generation, then any additions and subtractions due to crossover can be taken care of intelligently by only computing the size of the subtree that is taken away/added to the member. This would require some rewriting of the current algorithm.