int main(int argc, char* argv[]) {
corpus* corpus;
long t1;
(void) time(&t1);
seedMT(t1);
// seedMT(4357U);
if (argc > 1)
{
if (strcmp(argv[1], "est")==0)
{
INITIAL_ALPHA = atof(argv[2]);
NTOPICS = atoi(argv[3]);
read_settings(argv[4]);
corpus = read_data(argv[5]);
make_directory(argv[7]);
run_em(argv[6], argv[7], corpus);
}
if (strcmp(argv[1], "inf")==0)
{
read_settings(argv[2]);
corpus = read_data(argv[4]);
infer(argv[3], argv[5], corpus);
}
}
else
{
printf("usage : lda est [initial alpha] [k] [settings] [data] [random/seeded/*] [directory]\n");
printf(" lda inf [settings] [model] [data] [name]\n");
}
return(0);
}
int pc_prism_em_6(void)
{
struct EM_Engine em_eng;
RET_ON_ERR(check_smooth(&em_eng.smooth));
RET_ON_ERR(run_em(&em_eng));
release_num_sw_vals();
return
bpx_unify(bpx_get_call_arg(1,6), bpx_build_integer(em_eng.iterate )) &&
bpx_unify(bpx_get_call_arg(2,6), bpx_build_float (em_eng.lambda )) &&
bpx_unify(bpx_get_call_arg(3,6), bpx_build_float (em_eng.likelihood)) &&
bpx_unify(bpx_get_call_arg(4,6), bpx_build_float (em_eng.bic )) &&
bpx_unify(bpx_get_call_arg(5,6), bpx_build_float (em_eng.cs )) &&
bpx_unify(bpx_get_call_arg(6,6), bpx_build_integer(em_eng.smooth )) ;
}
bool CvEM::train( const CvMat* _samples, const CvMat* _sample_idx,
CvEMParams _params, CvMat* labels )
{
bool result = false;
CvVectors train_data;
CvMat* sample_idx = 0;
train_data.data.fl = 0;
train_data.count = 0;
CV_FUNCNAME("cvEM");
__BEGIN__;
int i, nsamples, nclusters, dims;
clear();
CV_CALL( cvPrepareTrainData( "cvEM",
_samples, CV_ROW_SAMPLE, 0, CV_VAR_CATEGORICAL,
0, _sample_idx, false, (const float***)&train_data.data.fl,
&train_data.count, &train_data.dims, &train_data.dims,
0, 0, 0, &sample_idx ));
CV_CALL( set_params( _params, train_data ));
nsamples = train_data.count;
nclusters = params.nclusters;
dims = train_data.dims;
if( labels && (!CV_IS_MAT(labels) || CV_MAT_TYPE(labels->type) != CV_32SC1 ||
(labels->cols != 1 && labels->rows != 1) || labels->cols + labels->rows - 1 != nsamples ))
CV_ERROR( CV_StsBadArg,
"labels array (when passed) must be a valid 1d integer vector of <sample_count> elements" );
if( nsamples <= nclusters )
CV_ERROR( CV_StsOutOfRange,
"The number of samples should be greater than the number of clusters" );
CV_CALL( log_weight_div_det = cvCreateMat( 1, nclusters, CV_64FC1 ));
CV_CALL( probs = cvCreateMat( nsamples, nclusters, CV_64FC1 ));
CV_CALL( means = cvCreateMat( nclusters, dims, CV_64FC1 ));
CV_CALL( weights = cvCreateMat( 1, nclusters, CV_64FC1 ));
CV_CALL( inv_eigen_values = cvCreateMat( nclusters,
params.cov_mat_type == COV_MAT_SPHERICAL ? 1 : dims, CV_64FC1 ));
CV_CALL( covs = (CvMat**)cvAlloc( nclusters * sizeof(*covs) ));
CV_CALL( cov_rotate_mats = (CvMat**)cvAlloc( nclusters * sizeof(cov_rotate_mats[0]) ));
for( i = 0; i < nclusters; i++ )
{
CV_CALL( covs[i] = cvCreateMat( dims, dims, CV_64FC1 ));
CV_CALL( cov_rotate_mats[i] = cvCreateMat( dims, dims, CV_64FC1 ));
cvZero( cov_rotate_mats[i] );
}
init_em( train_data );
log_likelihood = run_em( train_data );
if( log_likelihood <= -DBL_MAX/10000. )
EXIT;
if( labels )
{
if( nclusters == 1 )
cvZero( labels );
else
{
CvMat sample = cvMat( 1, dims, CV_32F );
CvMat prob = cvMat( 1, nclusters, CV_64F );
int lstep = CV_IS_MAT_CONT(labels->type) ? 1 : labels->step/sizeof(int);
for( i = 0; i < nsamples; i++ )
{
int idx = sample_idx ? sample_idx->data.i[i] : i;
sample.data.ptr = _samples->data.ptr + _samples->step*idx;
prob.data.ptr = probs->data.ptr + probs->step*i;
labels->data.i[i*lstep] = cvRound(predict(&sample, &prob));
}
}
}
result = true;
__END__;
if( sample_idx != _sample_idx )
cvReleaseMat( &sample_idx );
cvFree( &train_data.data.ptr );
return result;
}
