Get correct primal for MIP problems [needs review] #23
Conversation
cdiener
changed the title
|
This looks good. Nice catch on the optlang bug! |
|
Actually, I guess the |
|
Yes, good catch! I don't really know what do about the duals. There don't seem to be functions to get duals from MIP problems in GLPK.... |
|
From a bit of google searching, I don't think dual values are well defined for MIP problems, or at least not calculated with the normal optimization algorithms. It seems most solvers just return the dual from the LP relaxation instead, so doing that here as well should be fine. |
swiglpk/glpk.i
Outdated
| for(i=1; i<=n; i++){ | ||
| prim = glp_get_col_prim(P, i); | ||
| if (n_int == 0) { |
There was a problem hiding this comment.
Since speed is a concern, wouldn't it be better to perform the conditional check outside the loop? Or is this anyway optimized by the compiler?
There was a problem hiding this comment.
Nice catch! I don't know if it does but better to do it explicitly 👍
|
@KristianJensen ok, added it as suggested. @Midnighter Did make the change but it didn't affect performance at all. I suppose the compiler is good enough in branch optimization. |
Pinging @KristianJensen for review.
Tried to do it like optlang but realized that optlang has a bug when getting primals from MIP problems. If the problem is a MIP problem all primal values have to be requested with
glp_mip_col_valincluding continuous columns. Otherwise you will get incorrect primals for columns that are continuous but constrained by integer variables since it requests them from the "normal" LP solution that ignores integer variables.For now
get_col_primalswill deem a problem a MIP problem if there is at least one binary or integer variable and get all primals from the MIP solution in that case.