A more complex issue than first thought - undersampling according to the Gaussian distribution

Apologies for asking something very similar to last time but I was wondering if you could help me. I have points on a 17x17x17 cubic lattice and I have stored the x y and z corrdinates of these points in a 4913x3 matrix. I would like to undersample these points and only take measurements from a subset. I want to undersample according to the Gaussian distribution but initially I tired to simply undersample the indices according to the Gaussian, however, of course this will not work as it is necessary for me to also take into account the position of the points on the lattice. Any idea how I would do this I have thought to find the mean and variance of the 4913x3 matrix and then use normrnd (mu, sigma) to subsample the points

[Image Analyst]

I don't understand. WHERE you take the points from shouldn't depend on what the mean and variance of the array elements is, should it? I mean, you have elements going from x=1 to x=17. Now you could take 1000 points with x values in the range 7+/-3 (of course you'll have to interpolate if you want the array at non-integer x locations), but I don't see how 7+/-3 has anything to do with whether the mean of the entire array is 128 or 200 or 25000 or whatever. Please explain that to me.

[Bran]

Hi there! So sorry to take a while to respond. Well my matrix is 3 x 4913 and represent x, y and z points for points in my cubic lattice. I want to subsample the points according to the normal distribution (hence the inner 3 x 3 x 3 would be more densely sample) but I haven't stored them in any particular fashion in my matrix. So I thought of using something like Y = datasample(x,k,1,'replace' false) but I know this is based on the uniform distribution. I was wondering if there is a normal distribution equivalent?