As a quickfix(TM) to running into local maxima, genetic diversity might be useful. A algorithm needs to be implemented that takes a look at a list of genomes (sorted by fitness value) and puts a penalty on genomes which are closely related to other genomes. This metric shouldn't put a penalty on the best genome, and should assign a penalty on genomes only for being closely related to better genomes.
Possible Problem: Performance difference with regard to fitness function gets exagerated as all genomes are closely related in absolute terms in later generations.
As a quickfix(TM) to running into local maxima, genetic diversity might be useful. A algorithm needs to be implemented that takes a look at a list of genomes (sorted by fitness value) and puts a penalty on genomes which are closely related to other genomes. This metric shouldn't put a penalty on the best genome, and should assign a penalty on genomes only for being closely related to better genomes. Possible Problem: Performance difference with regard to fitness function gets exagerated as all genomes are closely related in absolute terms in later generations.