Our second set of experiments used

Our second set of experiments used a dynamic approach to deciding which chromosomes to evaluate. Instead of a fixed threshold we assumed that the designer would be able to specify a maximum member area for the initial pool. After that, we ran the algorithm in “cycles” with fitness evaluation done opportunistically within each cycle. In one cycle we would apply all of the operators to chromosomes selected by tournament selection, and the fine-tuning operators applied to the best chromosomes. Validity evaluation was deferred for all of these. We ordered the chromosomes by weight, then only evaluated a fixed percentage of those chromosomes starting from the lowest weight. We applied this dynamic approach with selection percentages of 30% and 50%. The idea was that initially we might consider a wider range of weights, but as the population converged we should tend to focus on narrower and narrower ranges of weights. We had 16 trials with 50% selection and 24 trials with 30% selection. The population size was reduced to 40. Due to the time required to complete a set of experiments we were not able to make the number of experimental runs identical for each configuration, preferring instead to have more runs and therefore more precise evaluation of the better algorithms; we report results below as a percentage of runs in order to make the comparison as fair as possible while using as much data as we have available about each configuration.