gsl_vector_view xty;
gsl_vector_view xi;
gsl_vector_view xj;
+ gsl_vector *param_estimates;
size_t i;
size_t j;
/*
Use QR decomposition via GSL.
*/
+
+ param_estimates = gsl_vector_alloc (1 + X->size2);
design = gsl_matrix_alloc (X->size1, 1 + X->size2);
for (j = 0; j < X->size1; j++)
}
gsl_multifit_linear_workspace *wk =
gsl_multifit_linear_alloc (design->size1, design->size2);
- rc = gsl_multifit_linear (design, Y, cache->param_estimates,
+ rc = gsl_multifit_linear (design, Y, param_estimates,
cache->cov, &(cache->sse), wk);
+ for (i = 0; i < cache->n_coeffs; i++)
+ {
+ cache->coeff[i].estimate = gsl_vector_get (param_estimates, i);
+ }
if (rc == GSL_SUCCESS)
{
gsl_multifit_linear_free (wk);
+ gsl_vector_free (param_estimates);
}
else
{