Since we have a perfect ANN for path length of 3 & 4 steps, but longer path lengths always topped out in the high 80's or low 90 percentages, my conclusion is that evolution is the bottleneck. It is now proven that the ANN configuration is capable of performing this task. I think it is worth some effort to see if the evolution process can be improved.
Currently, we have no genetic operations that move sequences of floats to alternate locations. We have one and two point crossover, but a sequence of floats can be substituted for another, it cannot be shifted to a different place in the chromosome. Instead of the clumsy phrase "sequence of floats", let me call that a "gene". This is the definition of "gene" used by Richard Dawkins. A biological chromosome is very long, hence a group of 100 or even 1000 units tends to stay together. Even though the chromosome may break in 100 places as a result of mating, since the length of a chromosome is in the millions, the breakpoints are quite far apart. When applies to our software, a "gene" can be defined as any group of consecutive floats which is much shorter than the total chromosome length. In nature, these genes can shift position. This results from several different genetic "operations" that occur. Sometimes a gene is splice erroneously far from where it belongs. Sometime a gene is reversed. Sometimes a gene is removed, and not replaced, thereby shifting the rest of the chromosome. Those operations are much less common than crossover, but they do occur. Hence, I think that it's worth the effort to add one or two additional genetic operators to the code to enable genes to move to other parts of the chromosome. Perhaps I will pick a gene at random and move it to another (random) position. m ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
