Latin Hypercube sampling is a form of random sampling except that it uses the stratification strategy to extract the random samples from the entire range, which makes it superior to the MonteCarlo sampling in a way. The things go well with univariate model. But when applying LHS to multivariate modeling, lets say 10 variables, how does LHS deal with the correlation among different variables? If there's no correlation between the variables, can any follow researcher suggest me a better sampling technique which takes account of correlation and performs improved sampling?
Regards,
Dear Mr. Beyk,
To face the mutltivariate challenge with LHS-typed techniques, check for the following links:
To constrain and control the correlations between variables, here are some references:
If needed, to extend the size of LHS while preserving induced correlations:
https://sem.org/wp-content/uploads/2016/03/sem.org-IMAC-XXIII-Conf-s10p04-A-Method-Extending-Size-Latin-Hypercube-Sample.pdf
Finally, I suggest you some overviews comparing different designs of experiments among which Latin Hypercube Sampling:
In situation where there's no correlation between the variables, Orthogonal LHS is convenient.
Best regards
Article An algorithm for fast optimal Latin hypercube design of experiments
Hello,
the correlation should let you decrease the dimension of the space of sampling.
Gabriel
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size