Le 12/10/2010 17:10, LEONID ILYEVSKY a écrit :
> Hi,
Hi Leonid,
>
> I have two questions.
>
> I was going through the SimplexTableau.java code, tried to understand every
> line
> of it. In the dropPhase1Objective() method I was puzzled a little bit, maybe
> you
> can explain.
> You are dropping positive cost non-artificial variables, which means that
> some
> original decision variable may be dropped in the process. However, that
> decision
> variable may be
>
> still there in the main objective Z and the constraints. Or is it now
> substituted by a basic artificial variable?
> Also, according to the getSolution() code, I see that the dropped variable
> will
> be assumed to have zero value. Again, I don't quite understand how that is
> going
> to work.
Do you have some explicit example that show some weird behavior ?
> I would understand if in dropPhase1Objective() you would examine the
> positive
> cost of the variable regarding the main Z objective, but you do it for the
> phase1 W objective (row # 0):
>
>
> // positive cost non-artificial variables
> for (int i = getNumObjectiveFunctions(); i < getArtificialVariableOffset();
> i++)
> {
> if (MathUtils.compareTo(tableau.getEntry(0, i), 0, epsilon) > 0) {
> columnsToDrop.add(i);
> }
> }
>
> Another question. I actually need to solve the quadratic optimization, which
> is
> done by modified simplex method. This is why I was studying the code. Are
> there
> any plans to add it in the apache.commons.math?
This would probably be a good addition to [math]. We don't currently
have plans on this feature, but if you want to contribute it, you're
welcome to do so. The best thing would be to discuss a bit about it on
this list before to make sure everybody agree on the way it would better
fit in the library.
Luc
> I saw a few free implementations, but each of them uses there own libraries
> for
> basic matrix/vector math, and I would really prefer to stick with apache. So
> I
> was thinking of doing it myself based on your current code.
>
> Thanks,
>
> Leonid
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
