experiments_ya/sim_regmix_cov_size.R

In kshimotsu/subgroupLRT:

# Computes the rejection frequencies of the LRT

rm(list=ls())

library(parallel)
library(normalregMix)
library(subgroupLRT)

setwd("~/Dropbox/yoichi/ProR/subgroupLRT/experiments")

outfilename <- "sim_regmix_cov_size.RData"

ncores <- detectCores()
cl <- makePSOCKcluster(rep('localhost',ncores-1),master='localhost')
clusterEvalQ(cl, library(normalregMix))
clusterEvalQ(cl, library(subgroupLRT))

sink("sim_regmix_cov_size.out", append=T)

nrep <- 10
# nobset <- c(100,200)
nobset <- c(100)
nnob <- length(nobset)
DGPset <- c(1:1)

m <- 2

#alpha <- c(0.5,0.5) #Y alpha will depend on covariates

mubeta <- matrix(c(1,2,1,2), nrow=2, ncol=2) # one-component model in effect
gammavec <- c(0.3, -0.7)
sigma <- c(0.5, 0.5)

# rejfreq5all <- matrix(0, nrow=length(DGPset), ncol=3*nnob)
# rejfreq1all <- matrix(0, nrow=length(DGPset), ncol=3*nnob)

for (DGP in DGPset)
{
	for (inob in 1:nnob)
	{
		time_start <- proc.time()
		nob <- nobset[inob]
		set.seed(123456)
		# clusterEvalQ(cl,set.seed(123456))

		x.all <- matrix(rnorm(nob*nrep), nrow = nob) - 1 # corresponds to Z in the paper
		w.all <- matrix(rnorm(nob*nrep), nrow = nob) + 1 # corresponds to X in the paper
		y.all <- matrix(double(nob*nrep), nrow = nob)
		for (j in 1:nrep) {
		  w1 <- cbind(1, w.all[, j])
		  w1gamma <- w1 %*% gammavec
		  alpha1 <- exp(w1gamma) / (1 + exp(w1gamma))
		  alpha <- cbind(alpha1, 1 - alpha1)
		  ii <- apply(alpha, 1, function(x) {sample(m, 1, replace = TRUE, x)})
		  x1 <- cbind(1, x.all[, j])
		  x1mubeta <- rowSums(x1 * t(mubeta[, ii]))
		  y.all[, j] <- rnorm(n = nob, mean = mubeta, sd=sigma[ii])
		}

		clusterExport(cl,varlist=c("y.all", "m", "x.all"))
    clusterSetRNGStream(cl, 123456)
#     lrtout <- parLapply(cl,1:nrep, function(j) regmixLRT_homo(y=y.all[,j],
#                               x=x.all[,j], m=m, crit.method="boot", parallel=0) )
# 		pvalsum <- sapply(lrtout,"[[","pvals")
# 		# print(pvalsum)
# 		rejfreq10 <- 100*mean(pvalsum < 0.10)
# 		rejfreq5 <- 100*mean(pvalsum < 0.05)
# 		rejfreq1 <- 100*mean(pvalsum < 0.01)
# 		# rejfreq5all[DGP,(3*inob-2):(3*inob)] <- rejfreq5
# 		# rejfreq1all[DGP,(3*inob-2):(3*inob)] <- rejfreq1
#
# 		time_end <- proc.time()
# 		runtime  <- time_end - time_start
#
# 		print("DGP, nob, nrep")
# 		print(c(DGP, nob, nrep))
# 		print(runtime)
# 		print("rejfreq10, rejfreq5, rejfreq1")
# 		print(c(rejfreq10, rejfreq5, rejfreq1))
# 		rm(list=c("y.all", "x.all"))
# 		save.image(file = outfilename)
	} # end of inob loop

} # end of DGP loop

stopCluster(cl)

# return(list(outall=outall,runtime=runtime))

sink()