Solving exception of Highs solver

[Zhanwei-Liu]

Hi,

When i use Highs to solve PREP-SHOT, i found two unexpected conditions. I am not sure whether there are some errors for the interface implementation of Highs.

1. Highs solving objective value is zero, which is not align with solving results of Gurobi and Mosek.
2. Highs output model to file is significant slower than Gurobi and Mosek.

Below i show a part of logs for Highs, Gurobi and Mosek, respectively.

• Highs

Coefficient ranges: Matrix [4e-04, 1e+06] Cost [1e+00, 1e+00] Bound [0e+00, 0e+00] RHS [0e+00, 0e+00] Presolving model 83252 rows, 68936 cols, 228318 nonzeros 0s 42692 rows, 28376 cols, 164746 nonzeros 0s 42670 rows, 28339 cols, 146459 nonzeros 0s Presolve : Reductions: rows 42670(-395723); columns 28339(-328753); elements 146459(-819124) Solving the presolved LP Solving the original LP from the solution after postsolve Model status : Optimal Objective value : 0.0000000000e+00 HiGHS run time : 0.39

• Gurobi

Solved with primal simplex Iteration Objective Primal Inf. Dual Inf. Time 582540 1.0538306e+11 0.000000e+00 0.000000e+00 121s Solved in 582540 iterations and 121.15 seconds (286.16 work units) Optimal objective 1.053830552e+11

• Mosek

Primal simplex reoptimization started. ITER DEGITER(%) PFEAS DFEAS POBJ DOBJ TIME 0 0.00 0.00e+00 NA 1.053830552479e+11 NA 0.01 162 99.39 0.00e+00 NA 1.053830552479e+11 NA 0.05 Primal simplex reoptimization terminated. Time: 0.05 Simplex reoptimization terminated. Time: 0.07 Basis identification terminated. Time: 0.69 Optimizer terminated. Time: 9.71 

You can reproduce my results use the following scripts.

git clone -b PyOptInterface https://github.com/PREP-NexT/PREP-SHOT.git cd PREP-SHOT conda env create -f prep-shot.yml conda activate prep-shot python run.py

You can modify the solving configuration in the solve_parameters section of config.json.

BTW, Mosek can quickly solve this example than Gurobi.
 
Do you have any idea for exceptions above?

Thank you very much!

[metab0t]

I change the singapore in config.json to southeast_asia and got 0 objective value for all solvers.

As for the output model to file, we simply call Highs_writeModel API, so it is the problem of HiGHS.

@Zhanwei-Liu

[Zhanwei-Liu]

Sorry for this. I have update config.json. Please use southeast_asia input to run my model.

[Zhanwei-Liu]

Thanks for the explanation!

I set the objective function as follows (model.py):

model.obj = poi.ExprBuilder(model.cost) model.set_objective(model.obj, sense=poi.ObjectiveSense.Minimize) define_constraints(model, para)

I clone my project and re-run my model. Gurobi and Mosek still get correct results instead of zero. You may need to git pull to get latest codes. Thank you!

[Zhanwei-Liu]

If you still cannot reproduce my results, please let me know, Thank you very much!

[metab0t]

Thank you for spotting a bug in HiGHS interface. I ignored the constant term in linear expression when adding linear constraint.
The fix is pushed at 419cecb
I will release a new version soon.

[Zhanwei-Liu]

Thank you very much!

I have tested the latest version, and the bug is fixed!