[R] using solve.qp without a quadratic term
maechler at stat.math.ethz.ch
Mon Dec 24 15:02:17 CET 2007
>>>>> "ML" == Mark Leeds <markleeds at verizon.net>
>>>>> on Fri, 21 Dec 2007 19:38:19 -0600 (CST) writes:
ML> I was playing around with a simple example using solve.qp ( function is in the quadprog package ) and the code is below. ( I'm not even sure there if there is a reasonable solution because I made the problem up ).
ML> But, when I try to use solve.QP to solve it, I get the error that D in the quadratic function is not positive
ML> definite. This is because Dmat is zero
ML> because I don't have a quadratic term in my
ML> objective function. So, I was wondering if
ML> it was possible to use solve.QP when there isn't
ML> a quadratic term in the objective function.
ML> I imagine that there are other functions in R that can be used but I would like to use solve.QP because, in my real problem,
ML> I will have a lot of fairly complex constraints
ML> and solve.QP provides a very nice way for implementing
ML> them. Maybe there is another linear solver that allows you to implement hundreds of constraints just solve.QP that I am unaware of ? Thanks for any suggestions.
ML> # IN THE CODE BELOW, WE MINIMIZE
ML> # -3*b1 + 4*b2 + 6*b3
ML> # SUBJECT TO
ML> # b1 + b2 + b3 >=0
ML> # -(b1 b2 + b3) >= 0
ML> # IE : b1 + b2 + b3 = 0.
So you want to solve a *linear* programming problem,
not a quadratic. Linear is typically considerably easier.
The recommended (and hence always installed) package 'boot'
has function simplex() to do this
and I see two other CRAN packages 'linprog' and 'lpSolve' also
for the same problem; since ?simplex says that it may not be
very efficient for large problems, you would e.g. lpSolve
ML> Dmat <- matrix(0,3,3) # QUADRATIC TERM
ML> dvec <- c(-3,4,6) # LINEAR TERM
ML> Amat <- matrix(c(1,-1,0,1,-1,0,1,-1,0),3,3)
ML> bvec = c(0,0,0) # THIRD ZERO IS SAME AS NO CONSTRAINT
ML> result <- solve.QP(Dmat, dvec, Amat)
ML> R-help at r-project.org mailing list
ML> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
ML> and provide commented, minimal, self-contained, reproducible code.
More information about the R-help