how to use rmvnrnd file to draw multivariate truncated normal distributions?

Hello all,

could you please share an example of how to use the matlab function rmvnrnd?

I am trying to generate a three multivariate truncated normal with data

MU=[5,4,3]; Sigma = [1 .3 .3; .3 1.5 .4; .3 .4 2.5] N=1000;

A=[-10;10;20]; B=[500,600,1000];

However the limits on A and B looks that are not working, I think I did a wrong specification of them. Could you give me a example please?

what I want to do is to have the limits:

X1 belongs to [-10,500] X2 belongs to [10,600] X3 belongs to [20,1000]

Thanks for your comments!

Answer 1

[the cyclist]

Do you mean this function from the FEX?:

http://www.mathworks.com/matlabcentral/fileexchange/34402-truncated-multivariate-normal

If so, then according to the documentation of that function, A and B are not lower and upper limits. Rather, A*x <= B. So, for example,

>> r = rmvnrnd(0,1,10000000,3,2);

will mean that all of the values of r are less than 2/3.

[Mohammad Anas]

How will i sample from a distribution with mean mu= -100, and variance=0.0000001, such that r should be greater than zero. It is giving nan values.

[the cyclist]

I'm confused by what you are asking for.

You want a mean of -100 (i.e. negative 100), with an extremely small variance, and you only want values that are greater than 0? Is that right?

That would be only a tiny, tiny, tiny fraction of the generated values. Like, you'd need to generate a number of values greater than the number of atoms in the universe, before you got a single one greater than 0.

What you are asking for also does not seem like a truncated normal. It sounds like you want the opposite of truncated ... values that are further from the mean than a certain value. You'd need to edit the code to get that.