double lda_m_step(lda* model, lda_suff_stats* ss) {
int k, w;
double lhood = 0;
for (k = 0; k < model->ntopics; k++)
{
gsl_vector ss_k = gsl_matrix_column(ss->topics_ss, k).vector;
gsl_vector log_p = gsl_matrix_column(model->topics, k).vector;
if (LDA_USE_VAR_BAYES == 0)
{
gsl_blas_dcopy(&ss_k, &log_p);
normalize(&log_p);
vct_log(&log_p);
}
else
{
double digsum = sum(&ss_k)+model->nterms*LDA_TOPIC_DIR_PARAM;
digsum = gsl_sf_psi(digsum);
double param_sum = 0;
for (w = 0; w < model->nterms; w++)
{
double param = vget(&ss_k, w) + LDA_TOPIC_DIR_PARAM;
param_sum += param;
double elogprob = gsl_sf_psi(param) - digsum;
vset(&log_p, w, elogprob);
lhood += (LDA_TOPIC_DIR_PARAM - param) * elogprob + gsl_sf_lngamma(param);
}
lhood -= gsl_sf_lngamma(param_sum);
}
}
return(lhood);
}
double c_ctr::doc_inference(const c_document* doc, const gsl_vector* theta_v,
const gsl_matrix* log_beta, gsl_matrix* phi,
gsl_vector* gamma, gsl_matrix* word_ss,
bool update_word_ss) {
double pseudo_count = 1.0;
double likelihood = 0;
gsl_vector* log_theta_v = gsl_vector_alloc(theta_v->size);
gsl_vector_memcpy(log_theta_v, theta_v);
vct_log(log_theta_v);
int n, k, w;
double x;
for (n = 0; n < doc->m_length; n ++) {
w = doc->m_words[n];
for (k = 0; k < m_num_factors; k ++)
mset(phi, n, k, vget(theta_v, k) * mget(m_beta, k, w));
gsl_vector_view row = gsl_matrix_row(phi, n);
vnormalize(&row.vector);
for (k = 0; k < m_num_factors; k ++) {
x = mget(phi, n, k);
if (x > 0)
likelihood += x*(vget(log_theta_v, k) + mget(log_beta, k, w) - log(x));
}
}
if (pseudo_count > 0) {
likelihood += pseudo_count * vsum(log_theta_v);
}
gsl_vector_set_all(gamma, pseudo_count); // smoothing with small pseudo counts
for (n = 0; n < doc->m_length; n ++) {
for (k = 0; k < m_num_factors; k ++) {
x = doc->m_counts[n] * mget(phi, n, k);
vinc(gamma, k, x);
if (update_word_ss) minc(word_ss, k, doc->m_words[n], x);
}
}
gsl_vector_free(log_theta_v);
return likelihood;
}