You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When running benchmarks for the MiniZinc Challenge, there have been some instances where MiniZinc's output processing has become the bottleneck for solvers which output many solutions quickly (essentially penalising you for producing more solutions).
In general, we probably want to ignore solution processing time when comparing solvers, so we could add an output mode where:
Only the objective is output during solving
At the end of solving, the full solutions are output
Outputting objective values would allow us to ensure that the solution was really produced at the time claimed by the solver.
So in this mode, MiniZinc would save the FlatZinc solutions to a buffer, and just print the objective value, then once done would run the solutions through the output model and print the result.
We could also potentially add a new flag for solvers to support this directly - so that they could store the solutions themselves and just output the objective to remove the overhead of printing the full solution (but I don't know if this is ever really a bottleneck anyway, so this might be overkill).
The text was updated successfully, but these errors were encountered:
When running benchmarks for the MiniZinc Challenge, there have been some instances where MiniZinc's output processing has become the bottleneck for solvers which output many solutions quickly (essentially penalising you for producing more solutions).
In general, we probably want to ignore solution processing time when comparing solvers, so we could add an output mode where:
Outputting objective values would allow us to ensure that the solution was really produced at the time claimed by the solver.
So in this mode, MiniZinc would save the FlatZinc solutions to a buffer, and just print the objective value, then once done would run the solutions through the output model and print the result.
We could also potentially add a new flag for solvers to support this directly - so that they could store the solutions themselves and just output the objective to remove the overhead of printing the full solution (but I don't know if this is ever really a bottleneck anyway, so this might be overkill).
The text was updated successfully, but these errors were encountered: