Solver, if this works, now selects the best possible augmentation path (one that splits an existing path). Does this by having nodes store distance to end and distance to start, to compute distances.
Performance improved, rarely has 1 or 2 steps more. Best case seen so far is 11 steps fewer.
Solver, if this works, now selects the best possible augmentation path (one that splits an existing path). Does this by having nodes store distance to end and distance to start, to compute distances.
Performance improved, rarely has 1 or 2 steps more. Best case seen so far is 11 steps fewer.