GA/GEP Mutation Rates
-
I wasn't sure if this forum was the best one for this question, so I apologize in advance if it isn't. I'm currently learned Gene Expression Programming and I'm a little unclear as to how the mutation rate is applied to a population. I've looked at source code from different implementations and I've seen it used different ways and would like to know if there is actually a proper way to implement it or if it can be implement at the programmers discretion. As an example, say there are 10 chromosomes, each with 2 genes, and each gene has 10 nodes and the mutation rate (mr) is 0.1. I've seen code that mutates nodes implemented in these way:
- Loop through each chromosome and generate a random number (rn) between 0 - 1 on each loop. If the rn <= mr then select the gene for mutation. Then loop through every node of the chromosome, ignoring gene boundaries (so in this example there are 20 nodes = 10 nodes/gene * 2 genes) and generate another rn and if the rn <= mr, mutate the node.
- Loop through each chromosome and generate a random number (rn) between 0 - 1 on each loop. If the rn <= mr then select the gene for mutation. Then loop through each gene, again generating a rn on each loop, and if the rn <= mr select the gene for mutation, otherwise continue the loop. If the gene is selected for mutation, loop through each node of the selected gene (10 nodes since there are 10 nodes/gene), generating a rn on each loop and mutate the selected node if rn <= mr.
- Randomly select 10% of the nodes in the entire populate (in this case 20 nodes since there are 200 nodes = 10 chromosomes * 2 genes/chromosome * 10 nodes/gene).
- Randomly select 10% of the chromosomes (in this case 1 chromosome) and then select the nodes for mutation following the same procedure as Item 1 or Item 2 above, starting from the second loop.
I believe that either Item 1 or Item 2 would be the correct way to apply the mutation operator, but I haven't been able to find a definitive answer either through searches on Google/CP or in the book I have been reading. One other thing that is mentioned in the book I am reading, but isn't explained in detail is the following: there is a population of 10 chromosomes, each with 1 gene, and a total length of 14 nodes. The book states that the mutation rate should be equivalent to two one-point mutations per chromosome, which in this case (because the chromosome's length is 14) should be 0.143. How was a mr of 0.143 derived from the chromosome length of 14? This is
-
I wasn't sure if this forum was the best one for this question, so I apologize in advance if it isn't. I'm currently learned Gene Expression Programming and I'm a little unclear as to how the mutation rate is applied to a population. I've looked at source code from different implementations and I've seen it used different ways and would like to know if there is actually a proper way to implement it or if it can be implement at the programmers discretion. As an example, say there are 10 chromosomes, each with 2 genes, and each gene has 10 nodes and the mutation rate (mr) is 0.1. I've seen code that mutates nodes implemented in these way:
- Loop through each chromosome and generate a random number (rn) between 0 - 1 on each loop. If the rn <= mr then select the gene for mutation. Then loop through every node of the chromosome, ignoring gene boundaries (so in this example there are 20 nodes = 10 nodes/gene * 2 genes) and generate another rn and if the rn <= mr, mutate the node.
- Loop through each chromosome and generate a random number (rn) between 0 - 1 on each loop. If the rn <= mr then select the gene for mutation. Then loop through each gene, again generating a rn on each loop, and if the rn <= mr select the gene for mutation, otherwise continue the loop. If the gene is selected for mutation, loop through each node of the selected gene (10 nodes since there are 10 nodes/gene), generating a rn on each loop and mutate the selected node if rn <= mr.
- Randomly select 10% of the nodes in the entire populate (in this case 20 nodes since there are 200 nodes = 10 chromosomes * 2 genes/chromosome * 10 nodes/gene).
- Randomly select 10% of the chromosomes (in this case 1 chromosome) and then select the nodes for mutation following the same procedure as Item 1 or Item 2 above, starting from the second loop.
I believe that either Item 1 or Item 2 would be the correct way to apply the mutation operator, but I haven't been able to find a definitive answer either through searches on Google/CP or in the book I have been reading. One other thing that is mentioned in the book I am reading, but isn't explained in detail is the following: there is a population of 10 chromosomes, each with 1 gene, and a total length of 14 nodes. The book states that the mutation rate should be equivalent to two one-point mutations per chromosome, which in this case (because the chromosome's length is 14) should be 0.143. How was a mr of 0.143 derived from the chromosome length of 14? This is
IMHO the type of mutation doesn't matter that much. Its purpose is to "jump" to nearby areas in the problem space. Any variation will accomplish that. The best mutation rate can't be stated as a hard and fast rule, as in that book; it will depend on the specific problem you're trying to solve. (BTW, there's no "singularity" at the center of a black hole; the "infinite density" is an artifact of using continuous equations to describe a discrete system. The physicists have mistaken their equations for the reality they're trying to model.)