[Cbc] CBC

[John Forrest]

Nathan,

For the original question.  If the constraints are added to the problem 
then the LP optimal answer will probably be different - in dual values 
or primal values so the procedure may generate more constraints.

Doing multiple solves at root node (and cuts and heuristics) is 
implemented in Cbc (idea due to  Andrea Lodi , Matteo Fischetti , 
Michele Monaci , Domenico Salvagnini and Andrea Tramontani).  If you set 
multipleRootPasses then your cut generator will be called with different 
starting solutions (if at all degenerate) and different pseudo random seeds.

John Forrest

On 08/09/17 13:11, Nathan Petty wrote:
> As a follow on question, if the answer to the original question is 
> that there is indeed a starting condition that will result in the 
> solution path being the same in multiple runs, is there also then a 
> starting condition seed that will guarantee a different path?  This 
> would also be very helpful because I could start multiple solvers with 
> different seeds in order to generate a family of optimal paths.
>
> On Thu, Sep 7, 2017 at 1:39 PM, Nathan Petty <nathanlpetty at gmail.com 
> <mailto:nathanlpetty at gmail.com>> wrote:
>
>     I am solving a TSP problem using CBC(2.9.9) and I have built a subtour
>     elimination procedure that adds subtour busting constraints and
>     resolves. It keeps adding these subtour elim constrains until the
>     solver finds a feasible optimal solution.
>
>     I save these subtour constraints in my database so that when I
>     resolve, they are loaded into the initial formulation, and I should in
>     theory not run into any more subtours because after all, it just found
>     an optimal feasible using this set of constraints.  But, to my
>     surprise, the solver finds more subtours (albiet many fewer than the
>     first time).  incase it matters, every variable has a unique
>     coefficient (driven my euclidean distance) in my obj function, so
>     there's a nice gradient there to slide on.
>
>     So there is something pseudo non-deterministic about the solution path
>     and my gut feeling is that it might be related to a starting
>     condition/initial solution?   So, is there a way to control that
>     starting condition?  Is it due to something else I am not thinking
>     about?
>
>     Thanks,
>     Nathan Petty
>
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> https://urldefense.proofpoint.com/v2/url?u=https-3A__list.coin-2Dor.org_mailman_listinfo_cbc&d=DwICAg&c=Ngd-ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=js2M0T-3OIMIVDvokcKjokJbk0F8QOCd0mT4FsVFE88&m=zt81WscZmncwk8_M3ot5McLjFc802iGvcGlrBHLyB30&s=A_GJkwvusnj_2LHFyN_xatoN6leMtFJCtKVkrqAX8_c&e=


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20170908/167f3fca/attachment.html>