Fri 10 Feb 2006
I’ve been working on my simple genetic algorithm in LabVIEW and struggling with memory. This was caused by a poor implementation of roulette wheel selection. I essentially created a very large array and randomly picked two parents from this array. The array contained multiple entries of each chromosome that was determined by it’s fitness score. This worked well enough for small populations and small fitness scores, but once things got big…
In this version, I implement a much improved roulette wheel selection algorithm. This is the algorithm documented on Marik Obitkos pages. The improvement was amazing. I found the new implementation gave me a 570x improvement with a population of 1500! And all my memory problems were gone. Developing the new implementation was a little frustrating. LabVIEW likes to make things simple and handle all the details for you such as variable type. This caused me a few headaches… The fitness score was implemented as a 16 bit signed integer. When I was calculating the sum of all fitness scores it was causing an overflow, but not raising any errors or warnings about this. The symptom I was seeing was that only one chromosome would make it through the selection process after a few runs. It wasn’t even the most fit one! After a lot of debugging, I saw this overflow. My next challenge was to convince LabVIEW to use an unsigned 32 bit integer to hold the data. I had to typecast the data going into the calculate sum vi in order to do this.
Here is version 3 with the new implementation of the selection algorithm.
algorithm code genetic labview programming