Originally posted by: MachexC

Dear ALEX,

My model is cannot finish when the data is large numbers in Excel file, but if I make data narrow then I get very fast results from the model. However, I don't want to do the data narrow because after the results run, I need to convert the correct percentage (1/1,000,000,000), which is easy to mistake. Thus, do we have anyway so that the model runs fast without making data narrow (keeping original data)?

For example, the objective of the model is minimized = C1 * X2 + C2 * X2

The original data: C1 = 500 USD, C2 = 1000 USD. In this case, the model can not finish the running with engine log as the figure below:

But when data narrow: C1 = 500 USD/1,000,000,000= 0.0000005 , C2 = 1000 USD/1,000,000,000 = 0.000001. Then the model will give fast results.

Thank you,

Mac

I strongly advise against scaling your objective by 1e-9. This creates tiny numbers like 1e-6 and 1e-7 which will probably cause numerical trouble (remember that we use IEEE floating point here, not exact arithmetic).

You can already see there is trouble by looking at your logs: In the log for the unscaled problem you can see that the best feasible solution found so far has objective value 83040.9330 and the lower bound for the optimal objective is 2848.9588. Scaling this by 1e-9 you get

83040.9330 * 1e-9 = 0.000083
 2848.9588 * 1e-9 = 0.000003

In the log for the scaled problem the best feasible solution is 0.0068 and the lower bound is 0.0069. As you can see, scaling the best feasible solution from the first log yields something significantly better than the lower bound in the second log. That is a contradiction.

My guess is that due to your tiny numbers some things are considered equal that actually are not equal.

Can you show the full log for the original formulation (including presolve output and root relaxation time)? What happens if you disable cutting planes?

Originally posted by: MachexC

Hi Daniel,

Please see the first log as the figure below. Yes, I don't want against scaling the objective, but if the data don't make scaling data, the model don't give results. Please help me.

Thank you,

Mac