Quoting MacDonald Phillips <[log in to unmask]>:
> A few years ago I used the Simplex program in the Users directory to do
> some DEA (Data Envelopment Analysis) using Derive 5.06. The contraints
> that DEA uses in the Simplex program are all of the form 'equation >= 0'.
> With 5.06 the constraints were transfered to the Simplex program with no
> problem; they were not simplified in any way. With Derive 6.1 that is not
> the case. If the contraints can be simplified, they are. For instance, a
> contraint of 9.86 x1 + 15.44 x2 - 20.34 x3 >= 0 is automatically simplified
> to 493 x1 + 772 x2 - 1017 x3 >= 0 when it is selected from a matrix of all
> the constraints. Thus, while the answers from the Simplex program
> are "technically" correct, they will not necessarily be scaled correctly in
> terms of the original, unsimplified objective function and constraints.
> And the shadow prices from the simplex table will also be incorrectly
> scaled. Is there any way not to simplfy the constraints within a Derive
> 6.1 program as was the case with Derive 5.06?
>
> Don Phillips
> [log in to unmask]
>
I was also rather unhappy when the Derive team has decided to
simplify automatically all the relations, but Albert Rich has
convinced me that this has some internal advantages.
If you want to use the Simplex program with the constraints
unsimplified, you can invoke directly the function SIMPLEXCALL
which takes as input the coefficient matrix as described in
the documentation. You will have to be careful with the
order of the constraints and with the fact that the constant
terms must be positive.
Best regards,
V. Anisiu
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
|