[Coin-discuss] questions on SMI

[Brady Hunsaker]

Thanks for the quick response, Alan.  There appear to be some good
possibilities.  I'll get back to you on or off list if I can persuade
some students to go for it.  Since I'm also trying to learn the material
myself, I may make a contribution of my own.

Brady

On Wed, 2004-06-23 at 09:02, Alan King wrote:
> 
> Alan King
> Mathematical Sciences, IBM Research
> http://www.research.ibm.com/
> http://www.research.ibm.com/people/k/kingaj/
> 
> Campus Relationship Manager: University of Washington
> http://www.washington.edu
> http://www.ibm.com/university
> 
> coin-discuss-admin at www-124.southbury.usf.ibm.com wrote on 06/22/2004
> 03:21:09 PM:
> 
> > I'm currently sitting in on a course on stochastic programming.
> There
> > are about a dozen Ph.D. students in the class, all of whom must do
> some
> > computational exercises (such as implementing the L-shaped method)
> as
> > well as a final project.  If I act fast I may be able to convince
> one or
> > more of them to do final projects that also contribute to SMI.
> > 
> > However, I'm having a little trouble understanding what SMI can
> already
> > do and what it aspires to.  I've been looking through the
> documentation
> > and code, including the unitTest code.
> > 
> > I'd appreciate it if someone familiar with SMI could answer these
> few
> > questions:
> > 
> > 1.  Current capabilities of SMI:  store scenario-based problems,
> read
> > SMPS format, use OSI to solve deterministic equivalent (extensive
> > form).  Is that correct? 
> 
> Yes: with one addition -- also have ability to store problems with
> independent random vectors.
> 
> > 
> > 2.  Would an L-shaped method algorithm be a good addition?  Is one
> > already in the works?  Are other solution methods particularly
> desirable
> > at this point?
> 
> Yes: an L-shaped would be a good addition.  I am planning to do one
> but won't 
> get around to it for a while.
> Also interesting (and perhaps easier to implement) a progressive
> hedging method.  JJF has code the quadratic already, so should be
> straightforward to implement.
> 
> > 
> > 3.  Is it correct that adding chance constraints would involve
> > developing a new class?  (For starters, I guess it would store the
> > description of the chance constraints and would also provide a
> solution
> > through deterministic equivalents for the basic cases.)  
> > 
> 
> Hmm.  I guess I would subclass SmiScnModel for this (stick to discrete
> distributions) and then have a method "declareChanceConstraint"?
> 
> > 4.  Any other ideas for projects that might be useful at this point?
> 
> Would be nice to have:
>         declare integer variables and propagate through deterministic
> equivalent
>         quadratic objective
>         more classes of stochastic random variables -- especially arma
> type
> 
> > 
> > Students are supposed to decide their final project topics by the
> end of
> > the week, so a response within a day or two would be appreciated. 
> > Thanks,
> > 
> > Brady
> >
> 
> Thanks!
> Alan
>