fix dumb beta sampling bug

This commit is contained in:
NunoSempere 2023-07-23 09:29:00 +02:00
parent 4dad518d3f
commit 5e39c386f7

View File

@ -74,46 +74,48 @@ float sample_to(float low, float high, uint32_t* seed)
return sample_lognormal(logmean, logsigma, seed); return sample_lognormal(logmean, logsigma, seed);
} }
float sample_gamma(float alpha, uint32_t* seed){ float sample_gamma(float alpha, uint32_t* seed)
{
// A Simple Method for Generating Gamma Variables, Marsaglia and Wan Tsang, 2001 // A Simple Method for Generating Gamma Variables, Marsaglia and Wan Tsang, 2001
// https://dl.acm.org/doi/pdf/10.1145/358407.358414 // https://dl.acm.org/doi/pdf/10.1145/358407.358414
// see also the references/ folder // see also the references/ folder
if(alpha >=1){ if (alpha >= 1) {
float d, c, x, v, u; float d, c, x, v, u;
d = alpha - 1.0/3.0; d = alpha - 1.0 / 3.0;
c = 1.0/sqrt(9.0 * d); c = 1.0 / sqrt(9.0 * d);
while(1){ while (1) {
do { do {
x = sample_unit_normal(seed); x = sample_unit_normal(seed);
v = 1.0 + c * x; v = 1.0 + c * x;
} while(v <= 0.0); } while (v <= 0.0);
v = pow(v, 3); v = pow(v, 3);
u = sample_unit_uniform(seed); u = sample_unit_uniform(seed);
if( u < 1.0 - 0.0331 * pow(x, 4)){ // Condition 1 if (u < 1.0 - 0.0331 * pow(x, 4)) { // Condition 1
// the 0.0331 doesn't inspire much confidence // the 0.0331 doesn't inspire much confidence
// however, this isn't the whole story // however, this isn't the whole story
// by knowing that Condition 1 implies condition 2 // by knowing that Condition 1 implies condition 2
// we realize that this is just a way of making the algorithm faster // we realize that this is just a way of making the algorithm faster
// i.e., of not using the logarithms // i.e., of not using the logarithms
return d*v; return d * v;
} }
if(log(u) < 0.5*pow(x,2) + d*(1.0 - v + log(v))){ // Condition 2 if (log(u) < 0.5 * pow(x, 2) + d * (1.0 - v + log(v))) { // Condition 2
return d*v; return d * v;
} }
} }
}else{ } else {
return sample_gamma(1 + alpha, seed) * pow(sample_unit_uniform(seed), 1/alpha); return sample_gamma(1 + alpha, seed) * pow(sample_unit_uniform(seed), 1 / alpha);
// see note in p. 371 of https://dl.acm.org/doi/pdf/10.1145/358407.358414 // see note in p. 371 of https://dl.acm.org/doi/pdf/10.1145/358407.358414
} }
} }
float sample_beta(float a, float b, uint32_t* seed){ float sample_beta(float a, float b, uint32_t* seed)
float gamma_a = sample_gamma(a, seed); {
float gamma_b = sample_gamma(b, seed); float gamma_a = sample_gamma(a, seed);
return a / (a + b); float gamma_b = sample_gamma(b, seed);
return gamma_a / (gamma_a + gamma_b);
} }
// Array helpers // Array helpers
@ -134,18 +136,20 @@ void array_cumsum(float* array_to_sum, float* array_cumsummed, int length)
} }
} }
float array_mean(float* array, int length){ float array_mean(float* array, int length)
float sum = array_sum(array, length); {
return sum / length; float sum = array_sum(array, length);
return sum / length;
} }
float array_std(float* array, int length){ float array_std(float* array, int length)
float mean = array_mean(array, length); {
float mean = array_mean(array, length);
float std = 0.0; float std = 0.0;
for (int i = 0; i < length; i++) { for (int i = 0; i < length; i++) {
std += pow(array[i] - mean, 2.0); std += pow(array[i] - mean, 2.0);
} }
std=sqrt(std/length); std = sqrt(std / length);
return std; return std;
} }