Structure selection

Dependence weights

Structure selection can use absolute Kendall's or Spearman's :

tree_criterion=:tau
# or
tree_criterion=:rho

R-vine selection

Automatic R-vine fitting follows a Dissmann-style sequential procedure: 2. compute pairwise dependence weights;

3. construct a maximum spanning tree;

4. fit the selected pair copulas;

5. generate conditional pseudo-observations;

6. build the next candidate graph under the proximity condition;

7. repeat until the requested truncation level.

tree_algorithm=:kruskal provides deterministic maximum-spanning-tree selection and is the route used by the external parity gate.

C- and D-vines

C-vine order selection favors variables with strong aggregate dependence. D-vine order selection treats the first tree as a path problem and uses an exact route when practical, with a deterministic fallback for larger dimensions.