#define n_MOMENTS (MOMENT_VARIANCE + 1)
+/* Create a new matrix of NEW_SIZE x NEW_SIZE and copy the elements of
+ matrix IN into it. IN must be a square matrix, and in normal usage
+ it will be smaller than NEW_SIZE.
+ IN is destroyed by this function. The return value must be destroyed
+ when no longer required.
+*/
+static gsl_matrix *
+resize_matrix (gsl_matrix *in, size_t new_size)
+{
+ size_t i, j;
+
+ gsl_matrix *out = NULL;
+
+ assert (in->size1 == in->size2);
+
+ if (new_size <= in->size1)
+ return in;
+
+ out = gsl_matrix_calloc (new_size, new_size);
+
+ for (i = 0; i < in->size1; ++i)
+ {
+ for (j = 0; j < in->size2; ++j)
+ {
+ double x = gsl_matrix_get (in, i, j);
+
+ gsl_matrix_set (out, i, j, x);
+ }
+ }
+
+ gsl_matrix_free (in);
+
+ return out;
+}
+
struct covariance
{
/* The variables for which the covariance matrix is to be calculated. */
cov->exclude = exclude;
- cov->n_cm = - 1;
+ cov->n_cm = -1;
cov->cm = NULL;
- cov->categoricals = categoricals_create (catvars, n_catvars, wv);
+ cov->categoricals = categoricals_create (catvars, n_catvars, wv, exclude);
return cov;
}
cm_idx (const struct covariance *cov, int i, int j)
{
int as;
- const int n2j = cov->n_vars - 2 - j;
- const int nj = cov->n_vars - 2 ;
+ const int n2j = cov->dim - 2 - j;
+ const int nj = cov->dim - 2 ;
assert (i >= 0);
- assert (j < cov->n_vars);
+ assert (j < cov->dim);
if ( i == 0)
return -1;
- if (j >= cov->n_vars - 1)
+ if (j >= cov->dim - 1)
return -1;
if ( i <= j)
return i - 1 + as;
}
-static void
+
+/*
+ Returns true iff the variable corresponding to the Ith element of the covariance matrix
+ has a missing value for case C
+*/
+static bool
+is_missing (const struct covariance *cov, int i, const struct ccase *c)
+{
+ const struct variable *var = i < cov->n_vars ?
+ cov->vars[i] :
+ categoricals_get_variable_by_subscript (cov->categoricals, i - cov->n_vars);
+
+ const union value *val = case_data (c, var);
+
+ return var_is_value_missing (var, val, cov->exclude);
+}
+
+
+static double
+get_val (const struct covariance *cov, int i, const struct ccase *c)
+{
+ if ( i < cov->n_vars)
+ {
+ const struct variable *var = cov->vars[i];
+
+ const union value *val = case_data (c, var);
+
+ return val->f;
+ }
+
+ return categoricals_get_binary_by_subscript (cov->categoricals, i - cov->n_vars, c);
+}
+
+void
dump_matrix (const gsl_matrix *m)
{
size_t i, j;
categoricals_update (cov->categoricals, c);
- for (i = 0 ; i < cov->n_vars; ++i)
+ for (i = 0 ; i < cov->dim; ++i)
{
- const union value *val1 = case_data (c, cov->vars[i]);
+ double v1 = get_val (cov, i, c);
- if ( var_is_value_missing (cov->vars[i], val1, cov->exclude))
+ if ( is_missing (cov, i, c))
continue;
- for (j = 0 ; j < cov->n_vars; ++j)
+ for (j = 0 ; j < cov->dim; ++j)
{
double pwr = 1.0;
- const union value *val2 = case_data (c, cov->vars[j]);
- if ( var_is_value_missing (cov->vars[j], val2, cov->exclude))
+ 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 *= val1->f;
+ pwr *= v1;
}
}
}
if (! cov->pass_two_first_case_seen)
{
+ size_t m;
assert (cov->state == 1);
cov->state = 2;
- cov->dim = cov->n_vars + categoricals_total (cov->categoricals);
+ cov->dim = cov->n_vars +
+ categoricals_total (cov->categoricals) - categoricals_get_n_variables (cov->categoricals);
+
cov->n_cm = (cov->dim * (cov->dim - 1) ) / 2;
cov->cm = xcalloc (sizeof *cov->cm, cov->n_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);
+ }
+
+ categoricals_done (cov->categoricals);
+
+ /* 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->n_vars; ++i)
+ for (i = 0; i < cov->dim; ++i)
{
- for (j = 0; j < cov->n_vars; ++j)
+ 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->n_vars; ++i)
+ for (i = 0 ; i < cov->dim; ++i)
{
- const union value *val1 = case_data (c, cov->vars[i]);
+ double v1 = get_val (cov, i, c);
- if ( var_is_value_missing (cov->vars[i], val1, cov->exclude))
+ if ( is_missing (cov, i, c))
continue;
- for (j = 0 ; j < cov->n_vars; ++j)
+ for (j = 0 ; j < cov->dim; ++j)
{
int idx;
double ss ;
- const union value *val2 = case_data (c, cov->vars[j]);
+ double v2 = get_val (cov, j, c);
- const double s = pow2 (val1->f - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) * weight;
+ const double s = pow2 (v1 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j)) * weight;
- if ( var_is_value_missing (cov->vars[j], val2, cov->exclude))
+ if ( is_missing (cov, j, c))
continue;
{
}
ss =
- (val1->f - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
+ (v1 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
*
- (val2->f - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
+ (v2 - gsl_matrix_get (cov->moments[MOMENT_MEAN], i, j))
* weight
;
{
cov->cm [idx] += ss;
}
-
}
}
cov->state = 1;
}
- for (i = 0 ; i < cov->n_vars; ++i)
+ for (i = 0 ; i < cov->dim; ++i)
{
const union value *val1 = case_data (c, cov->vars[i]);
- if ( var_is_value_missing (cov->vars[i], val1, cov->exclude))
+ if ( is_missing (cov, i, c))
continue;
- for (j = 0 ; j < cov->n_vars; ++j)
+ for (j = 0 ; j < cov->dim; ++j)
{
double pwr = 1.0;
int idx;
const union value *val2 = case_data (c, cov->vars[j]);
- if ( var_is_value_missing (cov->vars[j], val2, cov->exclude))
+ if ( is_missing (cov, j, c))
continue;
idx = cm_idx (cov, i, j);
cm_to_gsl (struct covariance *cov)
{
int i, j;
- gsl_matrix *m = gsl_matrix_calloc (cov->n_vars, cov->n_vars);
+ gsl_matrix *m = gsl_matrix_calloc (cov->dim, cov->dim);
/* Copy the non-diagonal elements from cov->cm */
- for ( j = 0 ; j < cov->n_vars - 1; ++j)
+ for ( j = 0 ; j < cov->dim - 1; ++j)
{
- for (i = j+1 ; i < cov->n_vars; ++i)
+ for (i = j+1 ; i < cov->dim; ++i)
{
double x = cov->cm [cm_idx (cov, i, j)];
gsl_matrix_set (m, i, j, x);
}
/* Copy the diagonal elements from cov->moments[2] */
- for (j = 0 ; j < cov->n_vars ; ++j)
+ for (j = 0 ; j < cov->dim ; ++j)
{
double sigma = gsl_matrix_get (cov->moments[2], j, j);
gsl_matrix_set (m, j, j, sigma);
covariance_calculate_double_pass (struct covariance *cov)
{
size_t i, j;
- for (i = 0 ; i < cov->n_vars; ++i)
+ for (i = 0 ; i < cov->dim; ++i)
{
- for (j = 0 ; j < cov->n_vars; ++j)
+ for (j = 0 ; j < cov->dim; ++j)
{
int idx;
double *x = gsl_matrix_ptr (cov->moments[MOMENT_VARIANCE], i, j);
/* Divide the moments by the number of samples */
if ( m > 0)
{
- for (i = 0 ; i < cov->n_vars; ++i)
+ for (i = 0 ; i < cov->dim; ++i)
{
- for (j = 0 ; j < cov->n_vars; ++j)
+ for (j = 0 ; j < cov->dim; ++j)
{
double *x = gsl_matrix_ptr (cov->moments[m], i, j);
*x /= gsl_matrix_get (cov->moments[0], i, j);
}
/* Centre the moments */
- for ( j = 0 ; j < cov->n_vars - 1; ++j)
+ for ( j = 0 ; j < cov->dim - 1; ++j)
{
- for (i = j + 1 ; i < cov->n_vars; ++i)
+ for (i = j + 1 ; i < cov->dim; ++i)
{
double *x = &cov->cm [cm_idx (cov, i, j)];
{
size_t i;
free (cov->vars);
+ categoricals_destroy (cov->categoricals);
for (i = 0; i < n_MOMENTS; ++i)
gsl_matrix_free (cov->moments[i]);
projects / pspp-builds.git / blobdiff