0

I am doing a research paper on a specific genetic algorithm and wanted to analyse the influence of using the Gaussian mutation method. However, the only thing I understand is that I have to sample a random Gaussian value and add that to the gene I have read somewhere on the internet that the mean should be 0, which I understand; this gives us negative as well as positive values. However, I have not found a single source that gave an example of what the std. dev should be or how it should be calculated.

Does anyone know how the standard deviation is determined using Gaussian mutation method so I can get a value from it?

I have read the question and answers of this question here on StackOverflow, but it does not provide me with any details regarding my problem.

Community
  • 1
  • 1
Zimano
  • 1,870
  • 2
  • 23
  • 41

1 Answers1

-1

What a reasonable (or even the optimal)mutation strength is, depends on the problem to be solved.
Usually you apply Genetic Algorithms to very hard optimization problems for that usual optimization algorithms fail. You can imagine the possible „solutions“ to such an optimization problem as a fitness landscape with high peaks for good solutions and valleys for bad ones.
So if your problem corresponds to a landscape with many peaks of similar height spread out widely (how do you know?), you should use a broad Gaussian distribution so that your chance to find the maximum peak is higher. If you believe however that you have already a pretty good solution (whatever this is) you could use a smaller distribution to find the maximum faster.
So a reasonable approach is to start with a broad distribution and let the population develop towards a (local) maximum by reducing the distribution width.
Again, the concrete numerical values must be derived from the problem.

EDIT:
If you want to play a little with the effects, you could download my free iPhone/iPad app "Steinertree" that shows the effects of varying mutation strength and population size.

Reinhard Männer
  • 14,022
  • 5
  • 54
  • 116
  • 1
    This does not answer the question.Could you please tell me how to derive concrete numerical values from the problem (Therefore answering my question)? – Zimano Oct 24 '16 at 11:15
  • First, you did not mention any concrete problem. Second, for most concrete problems it is very difficult to "derive" numerical values for mutation strength, crossover strength, and population size. Usually, one has to try out which combination gives good results. Thus my answer is the best I can give. – Reinhard Männer Oct 24 '16 at 14:16
  • I disagree with you and stand by my original comment. I wanted to know *how standard deviation is determined for use with gaussian mutation*. I did not ask about deriving mutation strength, crossover strength nor population size. Please update your answer accordingly as in its current state it does not answer the question and has a chance of being flagged as such by other users. – Zimano Oct 24 '16 at 15:16
  • Then, sorry, I did not understand your question. Maybe you could have explained better, at least for me. – Reinhard Männer Oct 24 '16 at 18:49
  • The question is clearly stated: "Does anyone know how the standard deviation is determined using Gaussian mutation method so I can get a value from it?" – Zimano Oct 25 '16 at 11:43