Start with some initial target.
Build first layer of the classification tree over left-hand side variables.
Build second layer of the classification tree over left-hand side variables.
Build third layer of the classification tree over left-hand side variables.
Relabel the entities in the leaves of the tree.
Use these new labels as the target.
Build first layer of the classification tree over right-hand side variables.
Build second layer of the classification tree over right-hand side variables.
Build third layer of the classification tree over right-hand side variables.
Relabel the entities in the leaves of the tree.
Use these new labels as the target to build a new tree over the left-hand side variables.
Build first layer of a new classification tree over left-hand side variables.
Continue alternating between the two sides...
... until the supports converge.
Take the last tree from each side,
select the branches leading to positive leaves,
and compute the supports
and compute the supports across the two sides.