Sat Mar 24 17:16:31 CET 2012
Invertible nodes constraint solver: limitations?
So what are the limitations of an invertible node solver? I'd say: no
dependency loops. If there is a loop, more powerful "primitives" are
necessary. Each loop needs a "global" solver.
Though, note that it's OK to have loops in the (undirected) network,
but not in the derived DAG.
So how would this be called? An N-in, M-out function that's
multi-directional, i.e. any pair of in/out can be swapped.
In case information is lost in one direction this will be related to
bi-directional lenses. See next post. Looks like I need bijective