Writing solutions with pySCIPopt

[abfariah]

Hi everyone,

To other users of the pySCIPopt package : the solution provided in the examples/python_example/run.py file to write the best primal solution in a __.sol file is not suited for series with a large number of variables. For rhs obj series 1(>90e3 variables), it takes 15 to 25 min to write the problem solution.
Using the writeBestSol(filename : str, write_zeros : bool) method from pyscipopt.Model, this time is reduced to 0.013s.

To the organizers : is it ok to change the approach for writing down solutions? The .sol format is very close to the one you describe in RULES.md.

Here is a sample on rhs_obj_1 with write_zeros = False:

To be compared to the examples/python_example/run.py version:

[ambros-gleixner]

Hi Paul, thank you very much for noting this! I just pushed an update to the Python example that should also print solutions much faster. Can you check? Is there still a huge difference to the writeBestSol() method?

[abfariah]

Hi Ambros, the writing time on rhs obj 1 is now 0.31s, which is longer but totally fine. My main worry was to exceed the strict 1.1 * t global time limit of the competition, this is not a concern anymore.

Thank your for such a swift response !

[ambros-gleixner]

Okay, happy to hear. Using writeBestSol() would also have been okay, but then I would have suggested a quick sed command to fix the formatting.

[abfariah]

A simple detail, I compared the .sol files produced by both methods and I noticed that with the "manual" solution, there were entries in the solution files with 1e-14 values. These entries were not present in the file produced by SCIP's writeBestSol() methods, meaning they were interpreted as 0.

To rectify this, it may be best to write for example :

if abs(solval) > 1e-10:

instead of :

if solval != 0.0:

in the loop on variables.

[ambros-gleixner]

True, thanks!