math - how to generate pseudo-random positive definite matrix with constraints on the off-diagonal elements? -


User binds a unique, non-trivial, upper / lower, in which the connection between the variables of each pair is related Var / covar matrix

For example: I want a variant matrix in which all the variables are 0. Rho (x_i, x_j) | > 0.6, RO (x_i, x_j) variable is going to be relating between x_i and x_j.

Thank you.


Okay, something has been found in a quick and dirty solution, even if anyone knows another exact way to go there, its will welcome.


I have lost my original login, so I am posting this question again, a new login semi is the correct term for random - Robert Gold < / P>

* It is a good issue, but I think that means that it means pseudo-eclipse while talking about semi-pseudo-random (Computer Randomization: -P) - Fortran

* With "correlation", do you mean "co-representative"? - Svente

* No, I really mean correlation. I want to generate a positive definite matrix such that the comparison of all relations is more than a trivial boundary. - vak

* See my answer Do you insist on whether the sample correlations are within specific boundaries, or only the population's correlation that produces samples? I suggest a suggestion that if your problem is preceded, Woodchips can work.

* woodship: No, I'm afraid your solution will not work, please be in the original danger (above link) go over the answers. Thank you.

Here is my answer to my answer in the original answer:

"People Come on, something should be simple "

I'm sorry, but it's not like that. The lottery is not enough to win, it is demanded that the cub is not enough to win the series. Nor can you ask for a solution to a mathematical problem and suddenly it can be detected that it is easy.

The problem of generating pseudo-random deviation with sample parameters in a particular category is invalid, at least if straying should in any way actually be pseudo-random, depending on the category, a lucky Maybe I suggested a rejection plan, but also said that it was not likely to be a good solution. If there are many dimensions and tight categories on correlations, then the potential for success is poor. Also the sample size is also important, because the sample variation of the resulting correlations will be driven.

If you really want a solution, then you have to sit down and clarify your goal clearly and completely. Do you want a random sample with a nominal specific correlation structure, but there are strict limitations on the correlations? Is there any sample correlation matrix which binds to the objective would be satisfactory? What variations have been given too?


Comments

Popular posts from this blog

c++ - Linux and clipboard -

What is expire header and how to achive them in ASP.NET and PHP? -

sql server - How can I determine which of my SQL 2005 statistics are unused? -