/* Call this function for every case in the data set.
After all cases have been passed, call covariance_calculate
*/
void
covariance_accumulate (struct covariance *cov, const struct ccase *c)
{
size_t i, j, m;
const double weight = cov->wv ? case_data (c, cov->wv)->f : 1.0;
assert (cov->passes == 1);
if ( !cov->pass_one_first_case_seen)
{
assert ( cov->state == 0);
cov->state = 1;
}
for (i = 0 ; i < cov->dim; ++i)
{
const union value *val1 = case_data (c, cov->vars[i]);
if ( is_missing (cov, i, c))
continue;
for (j = 0 ; j < cov->dim; ++j)
{
double pwr = 1.0;
int idx;
const union value *val2 = case_data (c, cov->vars[j]);
if ( is_missing (cov, j, c))
continue;
idx = cm_idx (cov, i, j);
if (idx >= 0)
{
cov->cm [idx] += val1->f * val2->f * weight;
}
for (m = 0 ; m < n_MOMENTS; ++m)
{
double *x = gsl_matrix_ptr (cov->moments[m], i, j);
*x += pwr * weight;
pwr *= val1->f;
}
}
}
cov->pass_one_first_case_seen = true;
}
// true if all values in ['.', '?']
bool cif_array::is_missing_all() const {
int n = get_nrows();
for (int i = 0; i < n; ++i) {
if (!is_missing(i))
return false;
}
return true;
}
/* Call this function for every case in the data set */
void
covariance_accumulate_pass1 (struct covariance *cov, const struct ccase *c)
{
size_t i, j, m;
const double weight = cov->wv ? case_data (c, cov->wv)->f : 1.0;
assert (cov->passes == 2);
if (!cov->pass_one_first_case_seen)
{
assert (cov->state == 0);
cov->state = 1;
}
if (cov->categoricals)
categoricals_update (cov->categoricals, c);
for (i = 0 ; i < cov->dim; ++i)
{
double v1 = get_val (cov, i, c);
if ( is_missing (cov, i, c))
continue;
for (j = 0 ; j < cov->dim; ++j)
{
double pwr = 1.0;
if ( is_missing (cov, j, c))
continue;
for (m = 0 ; m <= MOMENT_MEAN; ++m)
{
double *x = gsl_matrix_ptr (cov->moments[m], i, j);
*x += pwr * weight;
pwr *= v1;
}
}
}
cov->pass_one_first_case_seen = true;
}
int create_decoding_matrix(int *gen_matrix, int *dec_matrix, int *missing_idxs, int k, int m)
{
int i, j;
int n = k+m;
for (i = 0, j = 0; i < n && j < k; i++) {
if (!is_missing(missing_idxs, i)) {
copy_row(gen_matrix, dec_matrix, i, j, k);
j++;
}
}
return j == k;
}
GLint shader::operator[](const std::string& variable)
{
if (is_missing(variable))
return -1;
auto it = std::lower_bound(m_vars.begin(), m_vars.end(),
std::make_pair(variable, 0));
if ((it == m_vars.end()) || (it->first != variable))
{
GLint loc = glGetUniformLocation(m_prog, variable.c_str());
if (loc == -1) {
LOG(WARNING) << "Uniform '" << variable << "' not found in <"
<< m_vert_name << ", " << m_frag_name << ">.";
mark_missing(variable); /* Will not be reported again. */
} else m_vars.insert(it, std::make_pair(variable, loc));
return loc;
}
else return it->second;
}
/* Call this function for every case in the data set */
void
covariance_accumulate_pass2 (struct covariance *cov, const struct ccase *c)
{
size_t i, j;
const double weight = cov->wv ? case_data (c, cov->wv)->f : 1.0;
assert (cov->passes == 2);
assert (cov->state >= 1);
if (! cov->pass_two_first_case_seen)
{
size_t m;
assert (cov->state == 1);
cov->state = 2;
if (cov->categoricals)
categoricals_done (cov->categoricals);
cov->dim = cov->n_vars;
if (cov->categoricals)
cov->dim += categoricals_df_total (cov->categoricals);
cov->n_cm = (cov->dim * (cov->dim - 1) ) / 2;
cov->cm = xcalloc (cov->n_cm, sizeof *cov->cm);
/* Grow the moment matrices so that they're large enough to accommodate the
categorical elements */
for (i = 0; i < n_MOMENTS; ++i)
{
cov->moments[i] = resize_matrix (cov->moments[i], cov->dim);
}
/* Populate the moments matrices with the categorical value elements */
for (i = cov->n_vars; i < cov->dim; ++i)
{
for (j = 0 ; j < cov->dim ; ++j) /* FIXME: This is WRONG !!! */
{
double w = categoricals_get_weight_by_subscript (cov->categoricals, i - cov->n_vars);
gsl_matrix_set (cov->moments[MOMENT_NONE], i, j, w);
w = categoricals_get_sum_by_subscript (cov->categoricals, i - cov->n_vars);
gsl_matrix_set (cov->moments[MOMENT_MEAN], i, j, w);
}
}
/* FIXME: This is WRONG!! It must be fixed to properly handle missing values. For
now it assumes there are none */
for (m = 0 ; m < n_MOMENTS; ++m)
{
for (i = 0 ; i < cov->dim ; ++i)
{
double x = gsl_matrix_get (cov->moments[m], i, cov->n_vars -1);
for (j = cov->n_vars; j < cov->dim; ++j)
{
gsl_matrix_set (cov->moments[m], i, j, x);
}
}
}
/* Divide the means by the number of samples */
for (i = 0; i < cov->dim; ++i)
{
for (j = 0; j < cov->dim; ++j)
{
double *x = gsl_matrix_ptr (cov->moments[MOMENT_MEAN], i, j);
*x /= gsl_matrix_get (cov->moments[MOMENT_NONE], i, j);
}
}
}
for (i = 0 ; i < cov->dim; ++i)
{
double v1 = get_val (cov, i, c);
if ( is_missing (cov, i, c))
continue;
for (j = 0 ; j < cov->dim; ++j)
{
int idx;
double ss ;
double v2 = get_val (cov, j, c);
const double s = pow2 (v1 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) * weight;
if ( is_missing (cov, j, c))
continue;
{
double *x = gsl_matrix_ptr (cov->moments[MOMENT_VARIANCE], i, j);
*x += s;
}
ss =
(v1 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
*
(v2 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
* weight
;
idx = cm_idx (cov, i, j);
if (idx >= 0)
{
cov->cm [idx] += ss;
}
}
}
cov->pass_two_first_case_seen = true;
}
