help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Difference between normal_rnd (m,v, n,n) and randn (n)


From: Paul Kienzle
Subject: Re: Difference between normal_rnd (m,v, n,n) and randn (n)
Date: Thu, 4 Mar 2004 23:49:46 -0500


On Mar 4, 2004, at 2:45 PM, Henry F. Mollet wrote:

What's the difference between between using normal_rnd (m,v, n,n) and randn (n) to generate random numbers from a normal distribution. The former takes much longer? I'll probably only need 10+4 random numbers rather than 10+6, so speed may not present an actual problem but for other distributions only
the slower random number generator is available?

These distributions were written for completeness, correctness,
consistency, flexibility and perhaps readability, but not for speed.
In particular, given the m,v parameters for the normal distribution,
the first step is to make m and v both full-sized matrices so that
the code for varying m,v is the same as that for fixed.  This is
expensive.

With a bit of recoding, normal_rnd can be made just as correct,
and almost as efficient as m+sqrt(v)*randn(r,c):

octave:63> t=time; normrnd(3,4,1000,1000); time-t
ans = 0.77674
octave:64> t=time; 3+sqrt(4)*randn(1000,1000); time-t
ans = 0.73710


For compatibility, I've renamed it normrnd.  My version should
support Nd arrays as well as matrices, assuming randn does.

function rnd = normrnd (m, v, dims, c)
  if nargin < 3 || nargin > 4
    usage('normal_rnd(m,v,dims)');
  elseif nargin == 4
    dims = [dims, c];
  endif
  if (!isscalar(m) && any(size(m)!=dims)) ...
        || (!isscalar(v) && any(size(v)!=dims))
    error("dimensions do not match");
  end

  nanidx = isnan(m) | isinf(m) | v<0 | isinf(v);
  if all(nanidx(:))
    rnd = NaN*zeros(dims);
  else
    rnd = m + sqrt(v) .* randn (dims);
    if any(nanidx(:)), rnd(nanidx) = NaN; end
  end



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------



reply via email to

[Prev in Thread] Current Thread [Next in Thread]