Time Spending after Root Relaxation

[01543627906436788]

Hello what does Gurobi do after this line:

Root relaxation: objective 2.961095e+04, 20597 iterations, 109.00 seconds

Then this appears

Total elapsed time = 363.81s

Total elapsed time = 366.95s

Total elapsed time = 374.79s

    Nodes    |    Current Node    |     Objective Bounds      |     Work

 Expl Unexpl |  Obj  Depth IntInf | Incumbent    BestBd   Gap | It/Node Time

     0     0 29610.9503    0 1452 41022.2500 29610.9503  27.8%     -  377s

It spends an immense time in Total elapsed time without producing anything useful.

How can you turn that off and what does it do?

[Greg Glockner]

There are a number of heuristics that are run after solving the root relaxation. Based on our large set of test models, they are beneficial.

[Tú San Phạm]

Could you please kindly let me know what kind of heuristics Gurobi uses here? (or direct me to somewhere that I can find the answer). 

I want to focus on branch-and-cut algorithm so I want to turn those heuristics off. I have set 'Heuristics' param to 0 but it's still the same. Shed some light for me please?

[Simon Felix]

I run into the same issue, only that it is a bit worse in my instance:

Root relaxation: objective 0.000000e+00, 150523 iterations, 2079.08 seconds
Total elapsed time = 2335.49s
Total elapsed time = 10179.15s

[...and so on...]

I didn't find an attribute that would let me control the heuristics with Gurobi 6.5. Did I miss something?

[Tobias Achterberg]

Heuristics (as other algorithmic components) are not controlled by an attribute,
but instead by a parameter. You need to set the "Heuristics" parameter to 0 to
turn off heuristics.

But it might be that the long delay after the initial root LP solve is caused by
something different. If this is the case, just ask again :)


Regards,

Tobias

[Simon Felix]

I did set the Heuristics parameter to 0, but the solver still seems to be doing a lot of work before solving the root relaxation.

The problem in question needs better modelling anyway, so it's probably not that important. I was just curious.

Cheers,
Simon

[01543627906436788]

How can i turn this total elapsed time off?

I have sometimes the cases that it doesn't find a feasible solution and doesn't improve the bound.

I tried setting the heuristic to 0, but the log still appears.

[Tobias Achterberg]

This is not necessarily output from a heuristic. This can also appear from other
parts of the algorithm, e.g., when we try to find an alternative optimal LP
basis that has better properties for MIP solving.

Tobias

[Rolando Bettancourt]

Hi Tobias,

I'm also having this issue. The root relaxation is solved via Barrier in only 17 seconds. However, turning heuristics off, I still get the next "Total Elapsed Time" after 360 seconds. Can you please explain me what could be happening, how to get the appropriate logs, and perhaps how to troubleshoot this?

Thanks again,

-Rolando

[Tobias Achterberg]

What might be the issue is that the search for an alternative optimal LP basis takes so
long. You should try turning off the "DegenMoves" parameter.

Regards,

Tobias

[Rolando Bettancourt]

Hi Tobias, 

Thank you. This seems to be working. I'll check again tomorrow though.

Why does this long time happen? what is the trade-off of turning off the "DegenMoves"? Perhaps this can give me additional insight into my particular problem, not sure.

Thanks again,

-Rolando  

[Tobias Achterberg]

Given an optimal solution to the LP relaxation, the "DegenMoves" procedure is trying to
find a different optimal LP solution with fewer integer variables with fractional solution
value. Such a "more integral" solution is often useful as starting point for heuristics,
and it also seems to help cutting planes. But for some models, this search can be expensive.

You should check the difference in the first node log line in the "IntInf" column. If
"DegenMoves" is enabled, the "IntInf" value (i.e., the number of integer variables with
fractional value) should be smaller.


Regards,

Tobias

[Rolando Bettancourt]

Thank you Tobias! Always helpful!

-Rolando