I have to run computer simulations to find the optimal parameters for a structural problem. I have 8 factors and each of 3 levels. I can not run a full factorial design as it will take a lot of time to run 3^8 simulations. I want to reduce the number of simulations. I am confused between the Fractional Factorial method and Latin Hyper-cube method. Both can reduce the no. of experiments (simulation runs in this case), but what is fundamental difference between the two method? Which one is appropriate for this problem ?
Suggestions regarding other methods which can reduce the no. of simulation run are also welcome.
If you already have your levels, then a fractional factorial design is appropriate.
As you have noticed, a Latin Hypercube is a similar kind of design, except that it is used when the variables are continuous, not categorical or discrete, as yours are. It can divide a large space into an orthogonal design in a random, non-collapsible, and potentially sequentially-sampled way.
Herman Carstens thanks for your answer. Can you explain what do you mean by "variables are continuous"?
Latin Cubes and Hypercubes have same optimal properties. In particular, they are orthogonal. It means that the estimation of factors influence that you are getting are non-dependent statistically.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics