About jspRewritter.py line 64-68 about neural-rewriter HOT 1 OPEN

Sorry for the delay! About your questions:

1. Suppose J1 and J2 are 2 sets of job nodes, where the schedule time of each job in J2 is the same as the termination time of each job in J1. In our original implementation, we indeed add an edge between each job in J1 and J2, resulting in |J1| x |J2| edges. In our later experiments, we find that with this new implementation, the number of edges is reduced to |J1| + |J2| - 1, while the overall prediction performance does not affect much. Therefore, we released this version, but the original graph construction approach should work slightly better.

2. In our dependency graph representation for job schedules, nodes at the same level do not necessarily have the same termination time. Therefore, when node 5 does not start at the termination time of node 1, there is no edge between them.

from neural-rewriter.