How to duplicate a model?

[peno64]

Hello,

As the title says, how to duplicate a model, only by using the C api calls, to a new instance such that changes in the copy don't affect the original one and if possible can it also copy the solution status such that if I solve it (after making a change) that it iterates further. Also I need to copy the model without integer/binary/semi-cont variables to the copy must become a pure lp model.

Is there somewhere an example of this?

Thank you

[jajhall]

Use https://ergo-code.github.io/HiGHS/stable/interfaces/c_api/#Highs_getLp-NTuple{16,%20Any} to get the LP. If your original problem was an LP and solved to an optimal basis, use https://ergo-code.github.io/HiGHS/stable/interfaces/c_api/#Highs_getBasis-Tuple{Any,%20Any,%20Any} to get the basis.

If you pass the LP data to HiGHS with https://ergo-code.github.io/HiGHS/stable/interfaces/c_api/#Highs_passLp-NTuple{15,%20Any} and then call https://ergo-code.github.io/HiGHS/stable/interfaces/c_api/#Highs_setBasis-Tuple{Any,%20Any,%20Any} to pass the basis, HiGHS will hot start if you modify the LP using HiGHS calls

[jajhall]

And what about calling Highs_clearSolver and then calling Highs_setSolution ? Does this also not help for a warm start?

This is the same as doing nothing - ie leaving the solution of the MIP within HiGHS

[peno64]

And what about calling Highs_clearSolver and then calling Highs_setSolution ? Does this also not help for a warm start?

This is the same as doing nothing - ie leaving the solution of the MIP within HiGHS

Well that is not entirely what I notice. If I don't call Highs_setSolution then HiGHS needs 22 simplex iterations to solve my fixed model. If I do call Highs_setSolution then it only needs 7 iterations. Calling Highs_clearSolver and then Highs_run also results in 22 simplex iterations. So I do have the impression that HiGHS is not using the mip solution as a starting point after I fixed all my integer variables. Or maybe it is using the last solution of the B&B solve? On the other hand it is strange that the same number of iterations are needed when solution is not provided or a clear is done.

[jajhall]

And what about calling Highs_clearSolver and then calling Highs_setSolution ? Does this also not help for a warm start?

This is the same as doing nothing - ie leaving the solution of the MIP within HiGHS

Well that is not entirely what I notice. If I don't call Highs_setSolution then HiGHS needs 22 simplex iterations to solve my fixed model. If I do call Highs_setSolution then it only needs 7 iterations. Calling Highs_clearSolver and then Highs_run also results in 22 simplex iterations. So I do have the impression that HiGHS is not using the mip solution as a starting point after I fixed all my integer variables. Or maybe it is using the last solution of the B&B solve? On the other hand it is strange that the same number of iterations are needed when solution is not provided or a clear is done.

Sorry, I've just checked, and I'd forgotten that when changing the column bounds the incumbent solution is cleared. This is because, in general, changing the column bounds makes the incumbent solution primal infeasible, so it's unlikely to be useful when re-solving.

Yours is a real edge case, in that if an LP had been solved, the basis would be retained, since it's very likely to be useful. Since the incumbent solution would then be ignored when re-solving, clearing it is of no consequence.

[peno64]

This is because, in general, changing the column bounds makes the incumbent solution primal infeasible, so it's unlikely to be useful when re-solving.
Isn't that not only when the new lower bound would get above the current found solution value or the upper bound get below it? So clearing would only be needed in those case?

[jajhall]

Yes, and I considered not clearing the solution in this special case. Now I think about it, there may be cases in future when it's relevant to keep a feasible solution for warm start when there's no basis, so I'll add the special case.