inst/scripts/FiniteSampleCorrectionFactorScale.R

In RobLox: Optimally Robust Influence Curves and Estimators for Location and Scale

###############################################################################
## Find finite-sample correction factor for asymptotic radius
###############################################################################

library(distr)
library(RobLox)
library(Biobase)

## in combination with sysdata.rda of package RobLox
rowRoblox2 <- function(x, r, mean = 0, k = 1L){
    M <- rowMedians(x, na.rm = TRUE)
    sd <- rowMedians(abs(x-M), na.rm = TRUE)/qnorm(0.75)
    if(r > 10){
        b <- sd/(4*qnorm(0.75)*dnorm(qnorm(0.75)))
        A <- b^2*(1+r^2)
        a <- (qnorm(0.75)^2 - 1)/sd*A
    }else{
        A <- sd^2*RobLox:::.getA.sc(r)
        a <- sd*RobLox:::.geta.sc(r)
        b <- sd*RobLox:::.getb.sc(r)
    }
    robEst <- RobLox:::.kstep.sc.matrix(x = x, initial.est = sd, A = A, a = a, b = b, mean = mean, k = k)
    robEst$est <- as.matrix(robEst$est)
    colnames(robEst$est) <- "sd"
    return(robEst$est)
}

## attaining the maximum finite-sample risk
n <- 10
M <- 1e5
eps <- 0.01
D <- 0.1
fun <- function(r, x, n){
    RadMinmax <- rowRoblox2(x, r = r)
    n*mean(RadMinmax[,1]^2)
}

r <- rbinom(n*M, prob = eps, size = 1)
Mid <- rnorm(n*M)
Mcont <- rep(D, n*M)
Mre <- matrix((1-r)*Mid + r*Mcont, ncol = n)
ind <- rowSums(matrix(r, ncol = n)) >= n/2
while(any(ind)){
    M1 <- sum(ind)
    cat("M1:\t", M1, "\n")
    r <- rbinom(n*M1, prob = eps, size = 1)
    Mid <- rnorm(n*M1)
    Mcont <- r(contD)(n*M1)
    Mre[ind,] <- (1-r)*Mid + r*Mcont
    ind[ind] <- rowSums(matrix(r, ncol = n)) >= n/2
}

fun(r = 1, x = Mre, n = n)

fun1 <- function(D){
    Mcont <- rep(D, n*M)
    Mre <- matrix((1-r)*Mid + r*Mcont, ncol = n)
    fun(r = 1, x = Mre, n = n)
}
sapply(c(seq(0.1, 10, length = 20), 20, 50, 100, 1000, 1e4, 1e6), fun1)


## finite-sample optimal radius
## n at least 3, for n = 2 not possible to have less than 50% contamination
n <- c(3:50, seq(55, 100, by = 5), seq(110, 200, by = 10), seq(250, 500, by = 50))
eps <- c(seq(0.001, 0.01, by = 0.001), seq(0.02, to = 0.5, by = 0.01))
M <- 1e5
contD <- Dirac(1e6)

r.fi <- matrix(NA, nrow = length(eps), ncol = length(n))
colnames(r.fi) <- n
rownames(r.fi) <- eps
#for(j in seq(along = n)){
for(j in 65:74){
    ptm <- proc.time()
    cat("aktuelles n:\t", n[j], "\n")
    i <- 0
    repeat{
        i <- i + 1
        cat("aktuelles eps:\t", eps[i], "\n")
        r <- rbinom(n[j]*M, prob = eps[i], size = 1)
        Mid <- rnorm(n[j]*M)
        Mcont <- r(contD)(n[j]*M)
        Mre <- matrix((1-r)*Mid + r*Mcont, ncol = n[j])
        rm(Mid, Mcont)
        gc()
        ind <- rowSums(matrix(r, ncol = n[j])) >= n[j]/2
        rm(r)
        gc()
        while(any(ind)){
            M1 <- sum(ind)
            cat("M1:\t", M1, "\n")
            r <- rbinom(n[j]*M1, prob = eps[i], size = 1)
            Mid <- rnorm(n[j]*M1)
            Mcont <- r(contD)(n[j]*M1)
            Mre[ind,] <- (1-r)*Mid + r*Mcont
            ind[ind] <- rowSums(matrix(r, ncol = n[j])) >= n[j]/2
            rm(Mid, Mcont, r)
            gc()
        }
        r.fi[i,j] <- optimize(fun, interval = c(eps[i], min(max(2, n[j]*eps[i]*25), 11)), x = Mre, n = n[j])$minimum
        cat("finit:\t", r.fi[i,j], "\t asympt:\t", sqrt(n[j])*eps[i], "\n")
        rm(Mre)
        gc()
        if(round(r.fi[i,j], 2) > 3 | i == length(eps)) break
    }
#    save.image(file = "FiniteSampleScale1.RData")
    cat("Dauer:\t", proc.time() - ptm, "\n")
}

r.as <- outer(eps, sqrt(n))
r.fi[r.fi > 3] <- 3.5
r.fi[is.na(r.fi)] <- 3.5
r.finite <- round(pmax(r.fi, r.as, na.rm = TRUE), 4)