Hey, R users
I am using numerical method for my research paper and the computation burden is
very heavy. first I tried to do it with loops, example code as following, and
it
take hours to converge for only 200 obs. and my real data has 4000 obs. and the
optimization command that I use is:
optim(guess,myfunc1,data=mydata, method="BFGS",hessian=T))
then I tried matrix form computation, it takes only 1/10 of the time the loop
method takes. it may still have room to improve it. at least, the following
part looks ugly.
ccl[,m]<-lia[,1]*lia[,2]*lia[,3]*lia[,4]*lia[,5]
any suggestion are appreciated.
The Loop code:
for(m in 1:ns){
for(i in 1:nt){
vbar2[,i]=a[1]+ eta[m]+acedu[,i]*a[2]+acwrk[,i]*a[3]
vbar3[,i]=b[1]+b[2]*eta[m]+acedu[,i]*b[3]+acwrk[,i]*b[4]
v8[,i]=1+exp(vbar2[,i])+exp(vbar3[,i])
for(j in 1:n){
if (edu[j,i]==1) lia[j,i]=1/v8[j,i]
if (wrk[j,i]==1) lia[j,i]=exp(vbar2[j,i])/v8[j,i]
if (home[j,i]==1) lia[j,i]=exp(vbar3[j,i])/v8[j,i]
}
ccl[,m]<-lia[,i]*ccl[,m]
}
}
The Matrix code:
for(m in 1:ns){
vbar2[,1:nt]=a[1]+ eta[m]+acedu[,1:nt]*a[2]+acwrk[,1:nt]*a[3]
vbar3[,1:nt]=b[1]+b[2]*eta[m]+acedu[,1:nt]*b[3]+acwrk[,1:nt]*b[4]
v8[,1:nt]=1+exp(vbar2[,1:nt])+exp(vbar3[,1:nt])
lia[1:n,]<-ifelse(edu[1:n,]==1,1/v8[1:n,],
ifelse(wrk[1:n,]==1,exp(vbar2[1:n,])/v8[1:n,],
exp(vbar3[1:n,])/v8[1:n,]))
ccl[,m]<-lia[,1]*lia[,2]*lia[,3]*lia[,4]*lia[,5]
}
Nan
from Montreal
______________________________________________
[email protected] 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.
