Open tischi opened 9 months ago
Hi @tischi , I usually keep track of the second part where the relaxed model is solved. The first, presolving, is usually quite fast.
Second part:
...
Nodes | Current Node | Objective Bounds | Work
Expl Unexpl | Obj Depth IntInf | Incumbent BestBd Gap | It/Node Time
0 0 239.35844 0 3341 -399.99616 239.35844 160% - 121s
H 0 0 74.6577562 239.35844 221% - 121s
H 0 0 79.2766681 239.35844 202% - 122s
0 0 233.02530 0 2885 79.27667 233.02530 194% - 124s
H 0 0 93.7168490 233.02530 149% - 124s
H 0 0 110.1277578 233.02530 112% - 131s
...
H 724 774 202.2996744 228.36921 12.9% 1441 1892s
H 728 774 204.2018443 228.36921 11.8% 1440 1892s
820 887 227.32439 22 3407 204.20184 228.36921 11.8% 1375 1941s
The right-most column is the processing time in seconds for this step, which can be limited using tracking_config.time_limit
.
The Objective Bounds
show how close you are to the optimum; they have an upper bound (BestBd
) for our maximization problem, which is the objective at a solution with a relaxed version of the problem, that is, with fewer constraints.
The Incumbent
shows the objective of the current solution -- with all constraints, and the Gap
is the normalized gap between this objective and the bound; more information is provided, here.
It can take forever for large datasets to reach 0 gap, so setting a tolerance is preferred; this is the tracking_config.solution_gap
parameter. We use a gap of 0.1%
by default.
Once the solution gap or the time limit is reached, the optimization/search ends.
Hi @JoOkuma,
Would you mind to explain a bit which of those numbers are important to look at?