Mitchell Timin wrote: > I have a 4 neuron ANN that makes only one mistake for the 3-step path. > That is out of 24 bits, 3 bits for each of 8 possible paths. We can use > this one if we don't get a perfect one. An ANN that makes one error is > still interesting. This ANN has 28 weights & biases. Some of those may > turn out to be zero, and can be left out. > > I mentioned earlier that runmany.bat goes very quickly. This is mainly > because none of the runs of waybak take anywhere near their specified > time. They all end quickly due to evolutionary stagnation. So I made > another batch file that calls runmany.bat 6 times, and then merges those > results into a final run. Hence we have 3 levels of funnelling, where > 216 separate populations are merged. This file is attached, as is the > chomosome file. This file was evolved using the 3 to 4 step > parameters. But when I test this chromosome with only 3 steps, then it > makes exactly one bit error. > > Bigrun.bat takes about ten to 20 minutes to run, depending on which > machine I use. (for 4 neurons and 3 to 4 steps) The obvious next step is biggerun.bat, attached. This one calls bigrun.bat six times, each of which calls runmany six times, and each of those runs call wb.exe 36 times. I have not tested this; I just started a run; I figger it will take about 2 hours.
I think I will write a mini-essay about merging populations in evolutionary computation. m m -- I'm proud of http://ANNEvolve.sourceforge.net. If you want to write software, or articles, or do testing or research for ANNEvolve, let me know. ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
