Custom Query (545 matches)

There are some problems with the handling of the GEQ/LEQ supply type in NetworkSimplex.

1. The current version of the optimality conditions for these cases are not sufficient. We should also expect that all node potentials are non-positive (GEQ type) or non-negative (LEQ type).
2. The initialization of the algorithm (artificial arcs and initial basis) is not correct for these inequality cases if we also have negative costs.
3. Using LEQ and GEQ type we have to allow some of the artificial arcs to get into the basis again after they were removed.

mip_test.cc​ fails with the current trunk version of CBC (SVN trunk, rev 1234).

There is also a trivial bug in formatting the corresponding error message. This problem is fixed in the attached patch​, [860cdc983618].

Finally, compiling LEMON with CBC still emits a warning. For me, it looks like a real problem in CBC.

There is an annoying inconsistency in the current releases (1.0, 1.1): we have functions graphCopy() and copyPath(). Moreover the order of their parameters also differ: graphCopy(from, to) and copyPath(to, from).

More than one year ago, we decided to modify the 0.x version of graph copying functions: copyGraph(to, from) --> graphCopy(from, to), see #150. However, we were not careful enoguh, the path copying tools should have been modified, as well.

What could/should we do with this problem after we have released these tools with inconsistent interfaces?