Warm starting in integer programming is very tricky. It is not at all like the LP case and certainly not always guaranteed to help. It tends to work best when the problem modifications are slight, but if the tree that would have generated from scratch for two different instances is much different, then it may not very well. Basically, we are trying to re-use the same tree and continue the solution process from there after modifying the LP relaxations in the leaf nodes. If you warm start with a whole sequence of problems, the warm start tree may get larger and larger. At some point, you have to either start again from scratch or prune back the tree because the advantages of the warm start will be overcome. <div>
<br></div><div>There are some parameters you can play with that try different warm-starting strategies---it is certainly not guaranteed that the default is going to perform well. For example, in some cases, you may want to prune the tree back a little to generate the warm start each time, but this is problem-dependent. As with any effort to improve the efficiency of an algorithm for solving an NP-hard problem, the tuning process is both an art and a science.</div>
<div><br></div><div>Cheers,</div><div><br></div><div>Ted<br><br><div class="gmail_quote">On Fri, Sep 30, 2011 at 7:56 AM, Alexander Sviridenko <span dir="ltr"><<a href="mailto:oleks.sviridenko@gmail.com">oleks.sviridenko@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">It seems like the problem was solved. Thank you for support.<br>
<br>
However, since I was able to use warm starting, I have a question. I'm<br>
taking a problem and solve it. Then I'm changing rhs bounds and<br>
resolve it and so a lot of times. But the profit from warm starting in<br>
this case in comparison with a simple use of branchAndBound() for each<br>
resolving is very small. And sometimes even leads to negative result<br>
when the warm starting case works longer. Maybe you have some thoughts<br>
about this?<br>
<br>
Thank you.<br>
<font color="#888888"><br>
Alexander<br>
</font></blockquote></div><br><br clear="all"><div><br></div>-- <br>Dr. Ted Ralphs<br>Associate Professor, Lehigh University<br>(610) 628-1280<br>ted 'at' lehigh 'dot' edu<br><a href="http://coral.ie.lehigh.edu/~ted" target="_blank">coral.ie.lehigh.edu/~ted</a><br>
<br>
</div>