MathGroup Archive 2007

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Truncated version of lognormal distribution

  • To: mathgroup at smc.vnet.net
  • Subject: [mg82779] Re: Truncated version of lognormal distribution
  • From: Mark Fisher <particlefilter at gmail.com>
  • Date: Wed, 31 Oct 2007 06:02:53 -0500 (EST)
  • References: <fg6qj6$dks$1@smc.vnet.net>

On Oct 30, 4:39 am, P_ter <peter_van_summe... at yahoo.co.uk> wrote:
> Hello,
> did anyone program the truncated version of the lognormal distribution? Or have a reference how to do it? I would like to draw from such distribution.
> For the truncated version of the gaussian distribution I only know about Myron Tribus (Rational Descriptions 1969). It has a nice explanation. It is not difficult to use that, but I need anyway some reference for the results.
> What Tribus did by tens of BASIC lines, can now be done in three lines in Mathematica. That is for me also interesting to experience.
> with friendly greetings,
> P_ter

Hi,

Here is a simple way to draw from the truncated lognormal:

fun[{m_, s_}, {min_, max_}] :=
 Block[{r},
 While[
  r = RandomReal[LogNormalDistribution[m, s]];
  !(min <= r <= max)];
 r]

(If you are using a version of Mathematica prior to 6, change
RandomReal to Random.)

This is an example of "importance sampling", whereby one makes draws
from pdf f by drawing from pdf g and then resampling the draws using
the weights

w = f[#]/g[#]& /@ draws

In this case, g is simply the pdf for the lognormal and f is
proportional to g for draws that are in-bounds and zero for draws that
are out-of-bounds. Thus, all out-of-bounds draws get a weight of zero
(so these draws are discarded) and all in-bounds draws get the same
positive weight (so these draws are all retained).

For more general problems, one could use

RandomChoice[w -> draws, Length[draws]]

to resample the data. Unfortunately, RandomChoice does not allow zero-
valued weights and thus one is forced to write code that traps for
this. (I believe this is a design error. Perhaps the folks at WRI will
rethink this decision in the fullness of time.)

--Mark



  • Prev by Date: Re: "Accumulate" in Mathematica 6
  • Next by Date: Re: wrong answer or no answer?
  • Previous by thread: Truncated version of lognormal distribution
  • Next by thread: Setting Negatives to Zero