[Coin-discuss] More details on my user heuristic and cut framework and more questions...

[Fabrizio Ferrandi]

Dear John,

I've appreciated very much you answers. I've solved some of the issue and as 
happen in these case some others arise :-)
After some modifications, mainly following your suggestion, I've fixed my 
heuristic class. Actually the code is not yet complete since
my heuristic take advantage only from columns fixed to one and not from 
columns fixed to zero.
I've also fixed my cut class by introducing the CbcModel and by using 
bestSolution() to obtain the best solution. The aim of that code it to 
further constrain
the solver to avoid that the identified columns become one in the next 
analysis.
In attachment you can find the two mentioned classes and, for the sake of 
completeness, also the class that works as CbcMain: cbc_solver.

Concerning your suggestions about fixing I'll would like to explore the 
nested branch and cut case.
As you say this could help in fathoming the problem.
In fact, at each branch, before to solve it, there are several columns that 
could fixed to zero in case the upper branch is taken.
In particular, I've two types of situations:
- SOS variables
and
- branching variables for which the upper branch implies a more constrained 
ALAP and ASAP of the problem (i.e., fixing some variables to zero).
Probably also the lower branch should further constrain the problem but I've 
to better analyzed this case.

Note that, I've already tried to consider the SOS variables following the 
sos.cpp sample but up to now the best running time is without sos code.

John could you kindly provide me an example/sample where a nested branch and 
cut framework is realized? Does this example require the addition of a new 
virtual
function to fix columns?

Thanks for your support

Fabrizio

_________________________________________________________________
Don't just search. Find. Check out the new MSN Search! 
http://search.msn.click-url.com/go/onm00200636ave/direct/01/