int pod_experiment(char* observed_data, char* heldout_data,
char* model_root, char* out)
{
corpus *obs, *heldout;
llna_model *model;
llna_var_param *var;
int i;
gsl_vector *log_lhood, *e_theta;
doc obs_doc, heldout_doc;
char string[100];
double total_lhood = 0, total_words = 0, l;
FILE* e_theta_file = fopen("/Users/blei/llna050_e_theta.txt", "w");
// load model and data
obs = read_data(observed_data);
heldout = read_data(heldout_data);
assert(obs->ndocs == heldout->ndocs);
model = read_llna_model(model_root);
// run experiment
init_temp_vectors(model->k-1); // !!! hacky
log_lhood = gsl_vector_alloc(obs->ndocs + 1);
e_theta = gsl_vector_alloc(model->k);
for (i = 0; i < obs->ndocs; i++)
{
// get observed and heldout documents
obs_doc = obs->docs[i];
heldout_doc = heldout->docs[i];
// compute variational distribution
var = new_llna_var_param(obs_doc.nterms, model->k);
init_var_unif(var, &obs_doc, model);
var_inference(var, &obs_doc, model);
expected_theta(var, &obs_doc, model, e_theta);
vfprint(e_theta, e_theta_file);
// approximate inference of held out data
l = log_mult_prob(&heldout_doc, e_theta, model->log_beta);
vset(log_lhood, i, l);
total_words += heldout_doc.total;
total_lhood += l;
printf("hid doc %d log_lhood %5.5f\n", i, vget(log_lhood, i));
// save results?
free_llna_var_param(var);
}
vset(log_lhood, obs->ndocs, exp(-total_lhood/total_words));
printf("perplexity : %5.10f", exp(-total_lhood/total_words));
sprintf(string, "%s-pod-llna.dat", out);
printf_vector(string, log_lhood);
return(0);
}
void inference(char* dataset, char* model_root, char* out)
{
int i;
char fname[100];
// read the data and model
corpus * corpus = read_data(dataset);
llna_model * model = read_llna_model(model_root);
gsl_vector * lhood = gsl_vector_alloc(corpus->ndocs);
gsl_matrix * corpus_nu = gsl_matrix_alloc(corpus->ndocs, model->k);
gsl_matrix * corpus_lambda = gsl_matrix_alloc(corpus->ndocs, model->k);
// gsl_matrix * topic_lhoods = gsl_matrix_alloc(corpus->ndocs, model->k);
gsl_matrix * phi_sums = gsl_matrix_alloc(corpus->ndocs, model->k);
// approximate inference
init_temp_vectors(model->k-1); // !!! hacky
sprintf(fname, "%s-word-assgn.dat", out);
FILE* word_assignment_file = fopen(fname, "w");
for (i = 0; i < corpus->ndocs; i++)
{
doc doc = corpus->docs[i];
llna_var_param * var = new_llna_var_param(doc.nterms, model->k);
init_var_unif(var, &doc, model);
vset(lhood, i, var_inference(var, &doc, model));
gsl_matrix_set_row(corpus_lambda, i, var->lambda);
gsl_matrix_set_row(corpus_nu, i, var->nu);
gsl_vector curr_row = gsl_matrix_row(phi_sums, i).vector;
col_sum(var->phi, &curr_row);
write_word_assignment(word_assignment_file, &doc, var->phi);
printf("document %05d, niter = %05d\n", i, var->niter);
free_llna_var_param(var);
}
// output likelihood and some variational parameters
sprintf(fname, "%s-ctm-lhood.dat", out);
printf_vector(fname, lhood);
sprintf(fname, "%s-lambda.dat", out);
printf_matrix(fname, corpus_lambda);
sprintf(fname, "%s-nu.dat", out);
printf_matrix(fname, corpus_nu);
sprintf(fname, "%s-phi-sum.dat", out);
printf_matrix(fname, phi_sums);
}
int main (int argc, char **argv)
{
int i, j;
int interactive_test = 0;
CA_Arbdata *ca_arbdata = NULL; /* new, link btw acc/syn */
//char *modelmap = NULL;
char *arbfile = NULL;
char* q;
modelID model_sequence[128];
char pronunciation[256];
int pronunciation_len;
int rc;
srec_arbdata *allotree = NULL;
/* initial memory */
CHKLOG(rc, PMemInit());
if(argc<=1){
printf("USAGE: -swiarb <swiarb file> -interactive\n");
exit(1);
}
for(i=1; i<argc; i++) {
if(!strcmp(argv[i],"-swiarb")) {
if(argc==2){
printf("Please specify the swiarb file.\n");
exit(1);
}
arbfile = argv[++i];
printf("using swiarb from file %s\n", arbfile);
} else if(!strcmp(argv[i],"-interactive")) {
interactive_test++;
} else {
printf("error_usage: argument [%s]\n", argv[i]);
exit(1);
}
}
/* get modelID for a triphone */
ca_arbdata = CA_LoadArbdata(arbfile);
for(i=0; i<MAX_INTERACTIVE_NUM; i++){
if(interactive_test){
printf("Type \"quit\" to exit the test.\n");
printf("pronunciation: ");
q = fgets(pronunciation, sizeof(pronunciation), stdin);
if(!strcmp(q,"quit\n")) break;
}
else{
printf("USAGE: -swiarb <swiarb file> -interactive\n");
exit(1);
}
pronunciation_len = strlen(pronunciation)-1;
CA_ArbdataGetModelIdsForPron(ca_arbdata,
pronunciation, pronunciation_len,
&model_sequence[0]);
printf("short pronunciation length is %d.\n", pronunciation_len);
printf("Acoustic model IDs (\"#\" is silence,\"_\" is word boundary):\n");
for (j=0;j<pronunciation_len;j++){
if(j==0){
if(pronunciation_len==1)
printf("triphone:_%c_ -> ModelID:%d\n", pronunciation[j], model_sequence[j]);
else
printf("triphone:_%c%c -> ModelID:%d\n", pronunciation[j], pronunciation[j+1],
model_sequence[j]);
}
else if(j==(pronunciation_len-1)){
printf("triphone:%c%c_ -> ModelID:%d\n", pronunciation[j-1], pronunciation[j], model_sequence[j]);
}
else{
printf("triphone:%c%c%c -> ModelID:%d\n", pronunciation[j-1], pronunciation[j], pronunciation[j+1],
model_sequence[j]);
}
allotree = (srec_arbdata*)ca_arbdata;
printf_vector("pel_ids: ", " %d", allotree->hmm_infos[model_sequence[j]].state_indices,
(unsigned int) allotree->hmm_infos[model_sequence[j]].num_states);
printf("\n");
}
}
CA_FreeArbdata( ca_arbdata);
PMemShutdown();
return 0;
CLEANUP:
return 1;
}
