Currently the crossing reduction is an extremely simple layer sweep barycenter. This doesn't deal well with many situations, in particular when nodes are only connected in one direction.
I could imagine that iterative improvement steps (trying to re-route individual long) would go a long way. We could also play with the solver that's already embedded anyway.
Currently the crossing reduction is an extremely simple layer sweep barycenter. This doesn't deal well with many situations, in particular when nodes are only connected in one direction.
I could imagine that iterative improvement steps (trying to re-route individual long) would go a long way. We could also play with the solver that's already embedded anyway.