-
#include "surf/random_mgr.h"
#include "xbt/sysdep.h"
#ifdef WIN32
static double drand48(void)
{
- return rand()/(double)RAND_MAX;
+ THROW_UNIMPLEMENTED;
+ return -1;
+}
+
+static double rand_r(unsigned int* seed)
+{
+ THROW_UNIMPLEMENTED;
+ return -1;
}
#endif
-static double custom_random(int generator){
+static double custom_random(Generator generator, long int *seed){
switch(generator) {
-
- case DRAND48:return drand48();
- case RAND: return (double)rand()/RAND_MAX;
- default: return drand48();
+
+ case DRAND48:
+ return drand48();
+ case RAND:
+ return (double)rand_r((unsigned int*)seed)/RAND_MAX;
+ default:
+ return drand48();
}
}
/* Generate numbers between min and max with a given mean and standard deviation */
-float random_generate(random_data_t random){
- float x1, x2, w, y;
-
- if (random == NULL) return 0.0f;
+double random_generate(random_data_t random) {
+ double a, b;
+ double alpha, beta, gamma;
+ double U1, U2, V, W, X;
+
+ if (random == NULL) return 0.0f;
+
+ a = random->mean * ( random->mean * (1 - random->mean) / (random->std*random->std) - 1 );
+ b = (1 - random->mean) * ( random->mean * (1 - random->mean) / (random->std*random->std) - 1 );
+
+ alpha = a + b;
+ if (a <= 1. || b <= 1.)
+ beta = ((1./a)>(1./b))?(1./a):(1./b);
+ else
+ beta = sqrt ((alpha-2.) / (2.*a*b - alpha));
+ gamma = a + 1./beta;
do {
- /* Apply the polar form of the Box-Muller Transform to map the two uniform random numbers to a pair of numbers from a normal distribution.
- It is good for speed because it does not call math functions many times. Another way would be to simply:
- y1 = sqrt( - 2 * log(x1) ) * cos( 2 * pi * x2 )
- */
- do {
- x1 = 2.0 * custom_random(random->generator) - 1.0;
- x2 = 2.0 * custom_random(random->generator) - 1.0;
- w = x1 * x1 + x2 * x2;
- } while ( w >= 1.0 );
-
- w = sqrt( (-2.0 * log( w ) ) / w );
- y = x1 * w;
-
- /* Multiply the Box-Muller value by the standard deviation and add the mean */
- y = y * random->stdDeviation + random->mean;
- } while (!(random->min <= y && y <= random->max));
-
- return y;
+ /* Random generation for the Beta distribution based on
+ * R. C. H. Cheng (1978). Generating beta variates with nonintegral shape parameters. _Communications of the ACM_, *21*, 317-322.
+ * It is good for speed because it does not call math functions many times and respect the 4 given constraints
+ */
+ U1 = custom_random(random->generator,&(random->seed));
+ U2 = custom_random(random->generator,&(random->seed));
+
+ V = beta * log(U1/(1-U1));
+ W = a * exp(V);
+ } while (alpha * log(alpha/(b + W)) + gamma*V - log(4) < log(U1*U1*U2));
+
+ X = W / (b + W);
+
+ return X * (random->max - random->min) + random->min;
}
-random_data_t random_new(int generator, int min, int max, int mean, int stdDeviation){
+random_data_t random_new(Generator generator, long int seed,
+ double min, double max,
+ double mean, double std){
random_data_t random = xbt_new0(s_random_data_t, 1);
+
random->generator = generator;
+ random->seed = seed;
random->min = min;
random->max = max;
- random->mean = mean;
- random->stdDeviation = stdDeviation;
+
+ /* Check user stupidities */
+ if (max < min)
+ THROW2(arg_error,0,"random->max < random->min (%f < %f)",max, min);
+ if (mean < min)
+ THROW2(arg_error,0,"random->mean < random->min (%f < %f)",mean, min);
+ if (mean > max)
+ THROW2(arg_error,0,"random->mean > random->max (%f > %f)",mean, max);
+
+ /* normalize the mean and standard deviation before storing */
+ random->mean = (mean - min) / (max - min);
+ random->std = std / (max - min);
+
+ if (random->mean * (1-random->mean) < random->std*random->std)
+ THROW2(arg_error,0,"Invalid mean and standard deviation (%f and %f)",random->mean, random->std);
+
return random;
}
-