Dear friends.
I use ConstrOptim( ) and got error message "initial value not
available".
My understanding of "initial value not available" is that one of the
following 3 cases happens:
1.The objective function is not well defined at the point of the initial
value.
2. The differentiation of the objective function is not well defined at the
point of the initial value.
3. The initial value violate the constrain of "ui %*% theta - ci >=
0"
But my situation does not belong to any of the above cases.
I have attached my code bellow and could you please help me take a look?
Best Wishes
Yuchen Luo
rm(list = ls())
ui=matrix(c(1,-1,0,0,0,0,0,0,1,-1,0,0,0,0,0,0,1,-1),6,3)
ci=c(0,-0.5,0,-2,0,-0.6)
apple=expression(rr*(1.0-rec)*(1.0-(pnorm(-lambda/2.0+log(((ss+(tot/sh*
1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/lambda)-((ss+(tot/sh*1000.0)*lbar)/(tot/sh*
1000.0)/lbar*exp(lambda*lambda))*pnorm(-lambda/2.0-log(((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/lambda))+(exp(rr*(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*
1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))))*((((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))))))-(((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))))))))/((pnorm(-lambda/2.0+log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/lambda)-((ss+(tot/sh*1000.0)*lbar)/(tot/sh*
1000.0)/lbar*exp(lambda*lambda))*pnorm(-lambda/2.0-log(((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/lambda))-(pnorm(-sqrt((sigmae*ss/(ss+lbar*(tot/sh*
1000.0)))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*mat+lambda*lambda)/2.0+log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/sqrt((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))*mat+lambda*lambda))-((ss+(tot/sh*
1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda))*pnorm(-sqrt((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*mat+lambda*lambda)/2.0-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/sqrt((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*mat+lambda*lambda)))*exp(-rr*mat)-(exp(rr*(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*
1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))))*((((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt(mat+(lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))))))-(((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+((ss+(tot/sh*1000.0
)*lbar)/(tot/sh*1000.0)/lbar*exp(lambda*lambda))^(-sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))+0.5)*pnorm(-log(((ss+(tot/sh*1000.0)*lbar)/(tot/sh*1000.0
)/lbar*exp(lambda*lambda)))/((sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0))))))+sqrt(0.25+2.0*rr/
(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
))))*(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))*sqrt((lambda*lambda/(sigmae*ss/(ss+lbar*(tot/sh*1000.0
)))/(sigmae*ss/(ss+lbar*(tot/sh*1000.0)))))))))))
fit.error=function(rec,lambda,lbar)
{sum((eval(apple)*1000-orange)^2/(orange^2))
}
fit.error.grr=function(rec,lambda, lbar)
{drec=sum(eval(D(apple,'rec'))*(eval(apple)*1000-orange)/(orange^2))
dlambda=sum(eval(D(apple,'lambda'))*(eval(apple)*1000-orange)/(orange^2))
dlbar=sum(eval(D(apple,'lbar'))*(eval(apple)*1000-orange)/(orange^2))
c(drec,dlambda,dlbar)
}
rr=c(4.33,4.22,4.27,4.43,4.43,4.44,4.45,4.65,4.77,4.77)
tot=rep(13319.17,10)
sh=rep(1553656,10)
sigmae=c(0.172239074,0.188209271,0.193703774,0.172659891,0.164427247,
0.24602361,0.173555309,0.186701165,0.193150456)
ss=c(56.49,56.39,56.55,57.49,57.37,55.02,56.02,54.35,54.09)
orange=rep(21.25,10)
constrOptim(c(0.5,0.3,0.5), fit.error, fit.error.grr, ui=ui,ci=ci)
[[alternative HTML version deleted]]
Duncan Murdoch
2007-Sep-09 12:36 UTC
[R] What does it mean by "initial value not available"?
On 09/09/2007 7:01 AM, Yuchen Luo wrote:> Dear friends. > I use ConstrOptim( ) and got error message "initial value not available". > My understanding of "initial value not available" is that one of the > following 3 cases happens: > > 1.The objective function is not well defined at the point of the initial > value. > 2. The differentiation of the objective function is not well defined at the > point of the initial value. > 3. The initial value violate the constrain of "ui %*% theta - ci >= 0" > > But my situation does not belong to any of the above cases. > > I have attached my code bellow and could you please help me take a look?I haven't tried your code, but there's one obvious error: ...> fit.error=function(rec,lambda,lbar) > {sum((eval(apple)*1000-orange)^2/(orange^2)) > }The function optimizes over the elements of the first parameter. You've got two other parameters there, and I think you're trying to optimize over them as well. Put them all into one vector. The documentation for constrOptim doesn't make this as clear as it should; I'll clarify (by copying the docs from ?optim). Duncan Murdoch> > > fit.error.grr=function(rec,lambda, lbar) > {drec=sum(eval(D(apple,'rec'))*(eval(apple)*1000-orange)/(orange^2)) > dlambda=sum(eval(D(apple,'lambda'))*(eval(apple)*1000-orange)/(orange^2)) > dlbar=sum(eval(D(apple,'lbar'))*(eval(apple)*1000-orange)/(orange^2)) > c(drec,dlambda,dlbar) > } > > > rr=c(4.33,4.22,4.27,4.43,4.43,4.44,4.45,4.65,4.77,4.77) > tot=rep(13319.17,10) > sh=rep(1553656,10) > sigmae=c(0.172239074,0.188209271,0.193703774,0.172659891,0.164427247, > 0.24602361,0.173555309,0.186701165,0.193150456) > ss=c(56.49,56.39,56.55,57.49,57.37,55.02,56.02,54.35,54.09) > orange=rep(21.25,10) > > constrOptim(c(0.5,0.3,0.5), fit.error, fit.error.grr, ui=ui,ci=ci) > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.
Dear Professor Mordoch. Thank you very much for your help! Your time is highly appreciated! I do intend to optimize over 3 parameters and the way I did it is constrOptim(c(0.5,0.3,0.5), fit.error, fit.error.grr, ui=ui,ci=ci). Also, There is a missing value for both sigmae and ss. (The last 3rd and 4th line) Please correct them as: sigmae=c(0.172239074,0.188209271,0.193703774,0.172659891,0.164427247,0.24602361,0.173555309,0.186701165,0.193150456, 0.1857315601) ss=c(56.49,56.39,56.55,57.49 ,57.37,55.02,56.02,54.35,54.09,54.67) Best Wishes Yuchen Luo> On 9/9/07, Duncan Murdoch <murdoch at stats.uwo.ca> wrote: > > On 09/09/2007 7:01 AM, Yuchen Luo wrote: > > > Dear friends. > > > I use ConstrOptim( ) and got error message "initial value not > available". > > > My understanding of "initial value not available" is that one of the > > > following 3 cases happens: > > > > > > 1.The objective function is not well defined at the point of the initial > > > value. > > > 2. The differentiation of the objective function is not well defined at > > the > > > point of the initial value. > > > 3. The initial value violate the constrain of "ui %*% theta - ci >= 0" > > > > > > But my situation does not belong to any of the above cases. > > > > > > I have attached my code bellow and could you please help me take a look? > > > > I haven't tried your code, but there's one obvious error: > > ... > > > fit.error=function(rec,lambda,lbar) > > > {sum((eval(apple)*1000-orange)^2/(orange^2)) > > > } > > > > The function optimizes over the elements of the first parameter. You've > > got two other parameters there, and I think you're trying to optimize > > over them as well. Put them all into one vector. > > > > The documentation for constrOptim doesn't make this as clear as it > > should; I'll clarify (by copying the docs from ?optim). > > > > Duncan Murdoch > > > > > > > > > > > fit.error.grr=function(rec,lambda, lbar) > > > {drec=sum(eval(D(apple,'rec'))*(eval(apple)*1000-orange)/(orange^2)) > > > > dlambda=sum(eval(D(apple,'lambda'))*(eval(apple)*1000-orange)/(orange^2)) > > > dlbar=sum(eval(D(apple,'lbar'))*(eval(apple)*1000-orange)/(orange^2)) > > > c(drec,dlambda,dlbar) > > > } > > > > > > > > > rr=c(4.33,4.22,4.27,4.43,4.43,4.44,4.45,4.65,4.77,4.77) > > > tot=rep(13319.17,10) > > > sh=rep(1553656,10) > > > sigmae=c(0.172239074,0.188209271,0.193703774,0.172659891,0.164427247, > > > 0.24602361,0.173555309,0.186701165,0.193150456) > > > ss=c(56.49,56.39,56.55,57.49,57.37,55.02,56.02,54.35,54.09) > > > orange=rep(21.25,10) > > > > > > constrOptim(c(0.5,0.3,0.5), fit.error, fit.error.grr, ui=ui,ci=ci) > > > > > > [[alternative HTML version deleted]] > > > > > > ______________________________________________ > > > R-help at stat.math.ethz.ch mailing list > > > https://stat.ethz.ch/mailman/listinfo/r-help > > > PLEASE do read the posting guide > > http://www.R-project.org/posting-guide.html > > > and provide commented, minimal, self-contained, reproducible code. > > > > >