Hello all,
While browsing the source code I noticed that the flat/uniform
distribution provided in GSL's randist.h provided only a 'half-open'
interval [a,b) akin to what gsl_rng_uniform() provides for [0,1) and not
also an 'open' interval (a,b) akin to what gsl_rng_uniform_pos()
provides for (0,1).
So I wrote a couple of extra functions, which are attached as a git
patch in case they are useful/welcome. (Apologies if instead they are
stupid and I'm missing something; this is a very off-the-cuff
contribution, but it seemed easy enough to deliver and trivial enough to
not mind if it's rejected:-)
More general development question, which is the main reason for joining
this list. A project I'm working on needs to employ random sampling on
several occasions: to be precise, the case of selecting n unique records
from the set {1, ..., N}, as described in these articles:
http://doi.acm.org/10.1145/358105.893
http://doi.acm.org/10.1145/23002.23003
http://doi.acm.org/10.1145/79505.356313
The functionality seems generic and useful enough that I was surprised
not to find a library available. Anyway, it seems like it would be easy
enough to implement as part of GSL. Is there interest in having this?
If so, I'll map out a brief API description and/or sample code.
Best wishes,
-- Joe
From 09e5cdc0dd3991e69982bce6872edf201f25e7ed Mon Sep 17 00:00:00 2001
From: Joseph Rushton Wakeling <[email protected]>
Date: Mon, 10 May 2010 09:13:10 +0200
Subject: [PATCH] Flat (uniform) distribution over open interval.
* Two new functions that provide random variates and the
pdf of a flat (uniform) distribution over the open
interval (a, b).
* Corrected a potentially misleading comment in the
function for a uniform distribution on the half-closed
interval [a, b).
---
randist/flat.c | 33 ++++++++++++++++++++++++++++++++-
randist/gsl_randist.h | 2 ++
2 files changed, 34 insertions(+), 1 deletions(-)
diff --git a/randist/flat.c b/randist/flat.c
index 996366e..0a72a85 100644
--- a/randist/flat.c
+++ b/randist/flat.c
@@ -34,7 +34,7 @@ gsl_ran_flat (const gsl_rng * r, const double a, const double
b)
{
double u = gsl_rng_uniform (r);
- /* A uniform distribution over [a,b] */
+ /* A uniform distribution over the half-open interval [a,b) */
return a * (1 - u) + b * u;
}
@@ -51,3 +51,34 @@ gsl_ran_flat_pdf (double x, const double a, const double b)
return 0;
}
}
+
+
+/* This is the uniform distribution in the range (a, b)
+
+ p(x) dx = 1/(b-a) dx if a < x < b
+ ..... = 0 otherwise
+
+ */
+
+double
+gsl_ran_flat_open (const gsl_rng *r, const double a, const double b)
+{
+ double u = gsl_rng_uniform_pos (r);
+
+ /* A uniform distribution over the open interval (a, b) */
+
+ return a * (1 - u) + b * u;
+}
+
+double
+gsl_ran_flat_open_pdf (double x, const double a, const double b)
+{
+ if (x < b && x > a)
+ {
+ return 1 / (b - a);
+ }
+ else
+ {
+ return 0;
+ }
+}
diff --git a/randist/gsl_randist.h b/randist/gsl_randist.h
index 6f4b0e3..7218c76 100644
--- a/randist/gsl_randist.h
+++ b/randist/gsl_randist.h
@@ -68,6 +68,8 @@ double gsl_ran_fdist_pdf (const double x, const double nu1,
const double nu2);
double gsl_ran_flat (const gsl_rng * r, const double a, const double b);
double gsl_ran_flat_pdf (double x, const double a, const double b);
+double gsl_ran_flat_open (const gsl_rng * r, const double a, const double b);
+double gsl_ran_flat_open_pdf (double x, const double a, const double b);
double gsl_ran_gamma (const gsl_rng * r, const double a, const double b);
double gsl_ran_gamma_int (const gsl_rng * r, const unsigned int a);
--
1.7.0.4
