//------------------------------------------------------------
double MLB::logp(const ChoiceData & dp)const{
// for right now... assumes all choices are available to everyone
// uint n = dp->n_avail();
wsp.resize(nch_);
fill_eta(dp, wsp);
uint y = dp.value();
double ans = wsp[y] - lse(wsp);
return ans;
}
double DPMM::logp(const Vector &x) const {
int number_of_components = mixture_components_.size();
double ans = 0;
if (number_of_components == 1) {
ans += mixture_components_[0]->logp(x);
return ans;
}
Vector counts = allocation_counts();
// The Dirichlet process is the limit of finite mixture models
// with symmetric Dirichlet priors (with total mass alpha) on the
// mixing weights.
Vector probs(number_of_components, alpha() / number_of_components);
probs += counts;
probs /= sum(probs); // Posterior mode of mixing weights.
Vector log_probs = log(probs);
Vector wsp = log_probs;
for (int i = 0; i < number_of_components; ++i) {
wsp[i] += mixture_components_[i]->logp(x);
}
ans += lse(wsp);
return ans;
}
void MlvsDataImputer::impute_latent_data_point(const ChoiceData &dp,
SufficientStatistics *suf,
RNG &rng) {
model_->fill_eta(dp, eta); // eta+= downsampling_logprob
if (downsampling_) eta += log_sampling_probs_; //
uint M = model_->Nchoices();
uint y = dp.value();
assert(y < M);
double loglam = lse(eta);
double logzmin = rlexp_mt(rng, loglam);
u[y] = -logzmin;
for (uint m = 0; m < M; ++m) {
if (m != y) {
double tmp = rlexp_mt(rng, eta[m]);
double logz = lse2(logzmin, tmp);
u[m] = -logz;
}
uint k = unmix(rng, u[m] - eta[m]);
u[m] -= mu_[k];
wgts[m] = sigsq_inv_[k];
}
suf->update(dp, wgts, u);
}
Vec & MLB::predict(Ptr<ChoiceData> dp, Vec &ans)const{
fill_eta(*dp, ans);
ans = exp(ans-lse(ans));
return ans;
}
double PCR::response_prob(uint r, const Vector & Theta, bool logsc)const{
fill_eta(Theta);
double lognc = lse(eta_);
double ans = eta_[r] - lognc;
return logsc ? ans : exp(ans);
}
