Loading ...
Sorry, an error occurred while loading the content.

RE: AI-GEOSTATS: generating 2-dim point pattern

Expand Messages
  • Ted Harding
    ... Hi Peter, The problem with doing it the same way is that the same way only applies in 1 dimension (as a consequence of the result that F(X) has a
    Message 1 of 2 , Mar 10, 2003
      On 10-Mar-03 Peter Bossew wrote:
      > I want to generate a set of points in a plane, {(xi,yi)}, such
      > that the points are distributed according to a chosen probability
      > distribution density f, i.e., the infinitesimal rectangle
      > [x,x+dx] X [y,y+dy] shall contain f(x,y)*dx*dy points.
      > How do I do this ??
      > In other words, I am looking for a generalization of the 1-dim method,
      > where this is done with F*(z), where F=cumul(f) and F* = inverse of F,
      > and z = random out of [0,1] (or any appropriate interval if f is not
      > normalized).
      > I tried to do it the same way, just using complex numbers, but this
      > does not seem to work, and could also not be generalized for n>2
      > dimensions.

      Hi Peter,
      The problem with "doing it the same way" is that "the same way" only
      applies in 1 dimension (as a consequence of the result that F(X) has
      a uniform distribution, so that F*(X) has the distribution of X).
      In more than 1 dimension, you still only have the result that F(X,Y)
      has a uniform distribution, and this is only 1 equation. There is
      no _straightforward_ way of basing it on two equations

      (F1(X,Y), F2(X,Y)) = (Z1,Z1) where (Z1,Z2) is uniform on [0,1]x[0,1]

      However, there are possible approaches. One is to use conditional
      distributions: let F(X) be the marginal distribution for X, and let
      G(Y;x) be the conditional distribution of Y given that X=x. Then
      both U=F(X) and, for each x, V=G(Y;x) have uniform distributions,
      so provided you can invert these as F*(U) and G*(V;x) you can then
      sample U=u,V=v and obtain x=F*(u) and then y=G*(v;x).

      The practical issues here depend a lot on the form of F(X,Y),
      since both obtaining the conditional distribution and inverting
      the functions F(X), G(Y;x) may be difficult in practice.

      If it is straightforward to obtain the conditional distributions
      for Y given X and X given Y, then Gibbs Sampling can enable you to
      sample from (X,Y) without inverting the functions.

      If you can't obtain the conditional distributions then it may be
      impossible to follow such approaches, and then you may need to fall
      back on simulating a random process with a known mechanism which has
      the property of yielding (X,Y) distributed as f(x,y) -- if you know
      of such a process!

      If you could describe the density f(x,y) you are trying to sample
      from to the list, then perhaps someone can suggest an approach.

      > (PS. sorry if it is trivial.)

      No, it isn't trivial!

      Best wishes,

      E-Mail: (Ted Harding) <Ted.Harding@...>
      Fax-to-email: +44 (0)870 167 1972
      Date: 10-Mar-03 Time: 08:29:19
      ------------------------------ XFMail ------------------------------

      * To post a message to the list, send it to ai-geostats@...
      * As a general service to the users, please remember to post a summary of any useful responses to your questions.
      * To unsubscribe, send an email to majordomo@... with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
      * Support to the list is provided at http://www.ai-geostats.org
    Your message has been successfully submitted and would be delivered to recipients shortly.