double GenericGaussianVarianceSampler::draw(
RNG &rng,
double data_df,
double data_ss) const {
double DF = data_df + 2 * prior_->alpha();
double SS = data_ss + 2 * prior_->beta();
if (sigma_max_ == 0.0) {
return 0.0;
} else if(sigma_max_ == infinity()){
return 1.0 / rgamma_mt(rng, DF/2, SS/2);
} else {
return 1.0 / rtrun_gamma_mt(rng, DF/2, SS/2, 1.0/square(sigma_max_));
}
}
void MVT::Impute(bool sample, RNG &rng) {
std::vector<Ptr<WVD> > &V(mvn->dat());
for (uint i = 0; i < V.size(); ++i) {
Ptr<WVD> d = V[i];
const Vector &y(d->value());
double delta = siginv().Mdist(y, mu());
double a = (nu() + y.length()) / 2.0;
double b = (nu() + delta) / 2.0;
double w = sample ? rgamma_mt(rng, a, b) : a / b;
d->set_weight(w);
}
mvn->refresh_suf();
wgt->refresh_suf();
}
// Draws sigma given phi and observed data.
void ArPosteriorSampler::draw_sigma(){
// ss = y - xb y - xb
// = yty - 2 bt xty + bt xtx b
const Vec &phi(model_->phi());
const Vec &xty(model_->suf()->xty());
const Spd &xtx(model_->suf()->xtx());
double ss = 2 * siginv_prior_->beta();
ss += xtx.Mdist(phi) - 2 * phi.dot(xty) + model_->suf()->yty();
double df = 2 * siginv_prior_->alpha();
df += model_->suf()->n();
double siginv;
if (upper_sigma_truncation_point_ == BOOM::infinity()) {
siginv = rgamma_mt(rng(), df/2, ss/2);
} else {
siginv = rtrun_gamma_mt(rng(), df/2, ss/2,
1.0/pow(upper_sigma_truncation_point_, 2));
}
model_->set_sigsq(1.0 / siginv);
}
void ZeroInflatedPoissonSampler::draw() {
double p = model_->zero_probability();
double pbinomial = p;
double ppoisson = (1-p) * dpois(0, model_->lambda());
double nc = pbinomial + ppoisson;
pbinomial /= nc;
int nzero = lround(model_->suf()->number_of_zeros());
double nzero_binomial = rbinom_mt(rng(), nzero, pbinomial);
double nzero_poission = nzero - nzero_binomial;
int counter = 0;
do {
if (++counter > 1000) {
report_error("rbeta produced the value 0 over 1000 times.");
}
p = rbeta_mt(rng(),
zero_probability_prior_->a() + nzero_binomial,
zero_probability_prior_->b() + nzero - nzero_binomial +
model_->suf()->number_of_positives());
} while (p <= 0.0 || p >= 1.0);
model_->set_zero_probability(p);
double a = lambda_prior_->alpha() + model_->suf()->sum_of_positives();
double b = lambda_prior_->beta() + model_->suf()->number_of_positives();
b += nzero_poission;
double lambda = -1; // need to declare lambda before the do loop.
do {
if (++counter > 1000) {
report_error("rgamma produced the value 0 over 1000 times.");
}
lambda = rgamma_mt(rng(), a, b);
} while (lambda <= 0.0);
model_->set_lambda(lambda);
}
Vector rmvt_ivar_mt(RNG & rng, const Vector &mu, const SpdMatrix &ivar, double nu){
double w = rgamma_mt(rng, nu/2,nu/2);
return rmvn_ivar_mt(rng, mu, w*ivar);
}
Vector rmvt_mt(RNG & rng, const Vector &mu, const SpdMatrix &Sigma, double nu){
double w = rgamma_mt(rng, nu/2, nu/2);
return rmvn_mt(rng, mu, Sigma/w);
}
double rchisq_mt(BOOM::RNG & rng, double df){
if (!R_FINITE(df) || df <= 0.0) ML_ERR_return_NAN;
return rgamma_mt(rng, df / 2.0, 2.0);
}
