Forum Discussion
Why does Xpress ignore my warm start lp solution?
I am trying to use a warm start of the newton barrier method using the example provided here.
However, instead of using my supplied solution the solver starts with all values set to 0. I have tried both writeslxsol and loadlpsol. I can see my warm start solution, when I call getlpsol before calling the optimizer, however, calling lpoptimize will set it to 0.
The used Xpress version is 8.13 via Python
4 Replies
Can you share a minimal code example and MPS file to reproduce the problem?
Also, did you try setting `barstartweight` to different values?
I attached a code example and a small lp file. For the small example a value of barstartweight around 5.1 seems to give a start close to the given solution. However, for my larger lp (~350000 variables) the result stays at ||x1|| = 0.0 for barstartweight of 1, 1e4, 1e8 and 1e12.
Apparently, not setting the parameter presolve=0 allows the value to change, so maybe that was the problem? However, I still do not get the point that I gave the problem as a starting point.
How does barstartweight work? I want the solver to start at the feasible point that I provide. Do I just set the value very high? Or is there a specific value I need to set it to?
I am afraid your testprob.lp is too simple to tell anything. Without presolve it works as expected. With presolve enabled the problem is solved in presolve, i.e., in presolve all columns are fixed and there is nothing left to do for barrier. In other words, barrier starts with a model that has 0 rows and cols. In this case it will of course not use your starting solution but just stop. The optimal solution is already found before barrier even starts, that is why x0 and x1 differ.
In any case, when using the starting the solution, you should see a line like
Using primal warmstartin the output. Do you see this line?
Could you share your bigger LP? If you don't want to share it here, you may send to danieljunglas(at)fico(dot)com.
