[R] help using R's linprog for LP

[Hans W Borchers]

Medha Atre <medha.atre <at> gmail.com> writes:

> 
> Hi,
> 
> I found the reason. By default it puts a condition for x >= 0. Is
> there a way to get rid of this condition?


The constraints x >= 0 are used in most linear programming realizations.
Some bounds from below are needed. The trick to circumvent the restriction
is as follows:

Assume you know  x >= d  where some or all of the d_i can be negative.
Replace x with  y = x - d >= 0  and minimize c'y with  Ay <= b - A d !
Your solution is then  x = y + d , that is

    solveLP(cvec, bvec - Amat %*% dvec, Amat)$solution + dvec

Of course this works with all linear programming packages in R.


                [Good theorists] always make an even number of sign errors,
                and the bad ones always make an odd number of sign errors.
                                      --  Anthony Zee in: QFT in a Nutshell


> Medha
> 
> On Fri, Oct 23, 2009 at 2:34 PM, Medha Atre <medha.atre <at> gmail.com> wrote:
> > Hi,
> >
> > I am using R in one of my courses. I am trying to use R's linprog
> > package to solve to formulate 2-class classification problem as Linear
> > programming problem.
> >
> > For my formulation, I need to set to "cvec" to all 0s.
> > I know the points are linearly separable so an optimal solution "x"
> > does exist, which satisfies all the constraints.
> >
> > But given the constraints and setting "cvec" to all 0s it simply gives
> > me an "x" of all 0s.
> >
> > How can I fix this?
> >
> > Medha
> >
> 
>