I am solving multi IPs in a sequence, each of them differing only in a few constraints and objective coefficients. Technically it's a mixed-IP, but all of the important decision variables are all Boolean, while the couple of nonBoolean ones are tied to the Boolean ones (e.g. nonBoolean is the weighted sum of some Boolean variables) and serve to encodes a few constraints.

What could one do to speed up the process, given that the IPs are somewhat similar? I thought about using warm starts, although c++ concert technology for cplex does not allow warm starts for Boolean variables (I could change my Booleans to be IloNums and constraining the value between 0 and 1, though).

asked 28 Nov '16, 04:19

Emir's gravatar image

accept rate: 0%

edited 28 Nov '16, 04:20

One possibility would be to take the solution to problem N and try to tweak it to get a feasible solution to problem N+1, then feed that as a MIP start. This assumes that the problems are ordered in a way that makes it plausible to get a feasible solution to the next problem in "reasonable" time. Another possibility is to feed the previous optimum as a MIP start to the next problem and set the MIP start effort and number of repair tries high enough that CPLEX might be able to do the tweaking for you. There is, of course, absolutely no guarantee that either approach pays dividends.


answered 28 Nov '16, 19:30

Paul%20Rubin's gravatar image

Paul Rubin ♦♦
accept rate: 19%

Thank you for your input. I think I will do something similar to that, but I was hoping that there is something else one could take advantage of. For example, some information obtained during the solution process, like some common cuts.

(30 Nov '16, 08:16) Emir

Depending on the specific structure of the models, and what changes from one problem to the next, it's conceivable that some cuts might remain valid (and possibly even relevant) in the next problem. That would have to be determined from the model specifics, though (meaning there's no generic answer). As a trivial example, if you changed only objective coefficients, any cuts that defined the IP hull or at least trimmed infeasible portions of the LP hull would be valid in the next problem.

(30 Nov '16, 11:43) Paul Rubin ♦♦
Your answer
toggle preview

Follow this question

By Email:

Once you sign in you will be able to subscribe for any updates here



Answers and Comments

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text](http://url.com/ "Title")
  • image?![alt text](/path/img.jpg "Title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported



Asked: 28 Nov '16, 04:19

Seen: 1,452 times

Last updated: 30 Nov '16, 11:43

OR-Exchange! Your site for questions, answers, and announcements about operations research.