Interpreting the Cbc log report

[Azadeh]

Hello Robert,
I want to evaluate the performance of CBC compared to commercial solvers. Initially, I set the  CBC time limit to10-15min (i.e., the time range that commercial solvers needed to converge for the same problem), but CBC was not able to find even a feasible solution.

Hence, I gave it an hour, and it found the optimal solution. Now, I need to find out at what time during this one hour it has found the optimal solution for the very first time (as it seems to me that although it found the optimal solution earlier than an hour but it has continued the search probably to prove optimality.

The report format of Cbc is different from CPLEX or Gurobi's and I have difficulty to interpret Cbc's log report. I was wondering if you could guide me on how to find the very first moment that Cbc has found the optimal solution and could have been stopped and accordingly what options parameters of Cbc I need to change, so in the next test, Cbc stops at the right time. Please kindly find the Cbc log in the attached Word file.

Thank you in advance.
Best regards,
Azadeh

[AMPL Google Group]

The log output for CBC's branch-and-cut procedure is definitely very hard to understand. I suggest leaving slog=0 so that the log does not contain a lot of unimportant information about simplex iterations. Lines that report the current best solution and current lower bound (if minimizing) or upper bound (if maximizing) look like this:

Cbc0010I After 1800 nodes, 21 on tree, 267 best solution, best possible 266 (7.04 seconds)

Apparently these occur only once every 100 nodes, which would normally be sufficient to get a fairly accurate approximation of the time when CBC found the optimal solution. However, in your example, CBC reports

CBC 2.10.3 stopped on time limit, objective 116403562.7
2 nodes, 32301 iterations, 8626.96 seconds

So apparently only two nodes were searched, and it's not clear what else CBC was doing that took so much time. To learn more, you could post a question on the CBC discussion forum.

--
Robert Fourer
am...@googlegroups.com

{#HS:1595852403-105865#}

[Azadeh]

Thank you very much Robert for your guidance. 

Azadeh