[R-sig-hpc] suggestion in parallelisation, please.
Rohmatul Fajriyah
rfajriyah at yahoo.com
Fri Oct 3 20:24:02 CEST 2014
Dear All,
I am doing simulation in R and using parallel package in it.As some inputs (M1, M2, and M3) get larger then the simulation getting slow.
I really appreciate if anyone wants to give me suggestion how to make the paralellization better in my program (as below),since I still learn how to do it.
Apologise if this is not an appropriate place to ask some question/suggestion.
Thank you very much in advance.
With kind regards,Rohmatul
===========================================sdata<-function(M3,k,me,sx,sy){ a<-matrix(0,ncol=M3,nrow=2*k) for (i in 1: M3) { a[,i]<-rbind(rnorm(k,mean=me,sd=sx),rnorm(k,mean=me,sd=sy)) } return(a) }Xt<-sdata(10000,25,0,1)X<-as.list(data.frame(Xt))f1<-function(x,M1,M2,n){ X2<-x[1:n] Y2<-x[(n+1):(2*n)] Mx<-mean(X2) Vx<-var(X2) My<-mean(Y2) Vy<-var(Y2) Rac0=seq(0,0,length=M1) for (i in 1:M1) { Z10=seq(0,0,length=M2) Z20=seq(0,0,length=M2) for (j in 1:M2) { x0=rnorm(n,mean=Mx,sd=Sx) y0=rnorm(n,mean=My,sd=Sy) mx0=mean(x0) Vx0=var(x0) my0=mean(y0) Vy0=var(y0) Z10[j]= (Vx0/Vx)+(my0-mx0) Z20[j]= (Vy0/Vy)+(my0-mx0) } Z10<-rank(Z10)/M1 Z20<-rank(Z20)/M1 Rmc0<-summary(lm(Z10~Z20)) Rac0[i]<-Rmc0[[8]] } cat(Rac0,"\n")}
library(parallel)co<-detectCores()
sink("~/Documents/try0.R",append=TRUE)out<-mclapply(X,FUN=f1,M1=10000,M2=1000,n=25,mc.cores=(co-1))sink()
[[alternative HTML version deleted]]
More information about the R-sig-hpc
mailing list