The recent changes in mces_limit again broke evaluation of empty graphs.
This time the fix sets best_cv and best_ce back to None, and explicitly uses best_cv or pairs or {} as the argument to g.score. Another possibility would be to fix g.score not to fail when pairs=[], but I am too lazy for that.
Also note that all four combinations ({empty,nonempty} vs {empty,nonempty} graphs) work, it is not necessary to check nodes of the system graph for emptiness.
The recent changes in mces_limit again broke evaluation of empty graphs.
This time the fix sets best_cv and best_ce back to None, and explicitly uses
best_cv or pairs or {}
as the argument tog.score
. Another possibility would be to fixg.score
not to fail whenpairs=[]
, but I am too lazy for that.Also note that all four combinations ({empty,nonempty} vs {empty,nonempty} graphs) work, it is not necessary to check nodes of the system graph for emptiness.