Dear R Users I am maximizing a user defined log likelihood function. It includes variance parameter (sigma). I used R function optim with BFGS maximization method. However, it stops before the solution saying ?sqrt(sigma): NaNs produced? Could anybody know a proper transformation for sigma which can be passed in the function? For the correlation parameter I used Fishers? transformation so it didn?t give any problems. I know we can use ?L-BFGS-B? method instead of ?BFGS?. However, it produces some non sensible estimates. Thanks for taking time -- View this message in context: http://r.789695.n4.nabble.com/Maximization-problem-in-the-optim-function-tp4473408p4473408.html Sent from the R help mailing list archive at Nabble.com.
What distribution's log-likelihood are you using? If sigma is supposed to be square-rooted, you may have to put a constraint or use of abs() might suffice -- though, admittedly I'm not sure what that will do to convergence behavior -- but it's hard to help without seeing the function at hand. Michael On Wed, Mar 14, 2012 at 6:57 PM, Niroshan <wnnperer at ucalgary.ca> wrote:> Dear R Users > > > I am maximizing a user defined log likelihood function. It includes variance > parameter (sigma). I used R function optim with BFGS maximization method. > However, it stops before the solution saying ?sqrt(sigma): NaNs produced? > Could anybody know a proper transformation for sigma which can be passed in > the function? ?For the correlation parameter I used Fishers? transformation > so it didn?t give any problems. I know we can use ?L-BFGS-B? method instead > of ?BFGS?. However, it produces some non sensible estimates. > > Thanks for taking time > > > -- > View this message in context: http://r.789695.n4.nabble.com/Maximization-problem-in-the-optim-function-tp4473408p4473408.html > Sent from the R help mailing list archive at Nabble.com. > > ______________________________________________ > R-help at r-project.org 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.
This is the part of the function in my likelihood function
prob<-function(t1,t2,m1,m2,s_r,s_p,s_e,cor_m){
if ((t1==0) & (t2==0))
log_lik<-log(pmvnorm(lower=rep(-Inf, 2),
upper=c(m1/sqrt(s_r+s_p+s_e+1),m2/sqrt(s_r+s_p+s_e+1)),
mean=rep(0,2), corr=cor_m))
else if ((t1==0) & (t2==1))
log_lik<-log(pmvnorm(lower=c(-Inf,m2/sqrt(s_r+s_p+s_e+1)),
upper=c(m1/sqrt(s_r+s_p+s_e+1),Inf),
mean=rep(0,2), corr=cor_m))
else if ((t1==1) & (t2==0))
log_lik<-log(pmvnorm(lower=c(m1/sqrt(s_r+s_p+s_e+1),-Inf),
upper=c(Inf,m2/sqrt(s_r+s_p+s_e+1)),
mean=rep(0,2), corr=cor_m))
else if ((t1==1) & (t2==1))
log_lik<-log(pmvnorm(lower=c(m1/sqrt(s_r+s_p+s_e+1),m2/sqrt(s_r+s_p+s_e+1)),
upper=c(Inf,Inf),
mean=rep(0,2), corr=cor_m))
return(log_lik)
}
--
View this message in context:
http://r.789695.n4.nabble.com/Maximization-problem-in-the-optim-function-tp4473408p4473602.html
Sent from the R help mailing list archive at Nabble.com.