Dear Rui,
I thank you for your response but when I run the code with your few
modifications, I still don't get the 8 graphs but I get the following answer :
null device
1
Here below my R code with your modifications. I don't know what I am still
missing ?
##############
set.seed(1)
library(energy)
# Here we define parameters which we use to simulate the data
# The number of null datasets we use to estimate our rejection reject #regions
for an alternative with level 0.05
nsim=50
# Number of alternative datasets we use to estimate our power
nsim2=50
# The number of different noise levels used
num.noise <- 30
# A constant to determine the amount of noise
noise <- 3
# Number of data points per simulation
n=100
# Vectors holding the null "correlations" (for pearson, for spearman, for
kendall and dcor respectively) for each # of the nsim null datasets at a #given noise
level
val.cor=val.cors=val.cork=val.dcor=rep(NA,nsim)
# Vectors holding the alternative "correlations" (for pearson, for #spearman,
for kendall and dcor respectively) #for each of the nsim2 alternative datasets at a given
noise level
val.cor2=val.cors2=val.cork2=val.dcor2= rep(NA,nsim2)
# Arrays holding the estimated power for each of the 4 "correlation" types, for
each data type (linear, #parabolic, etc...) with each noise level
power.cor=power.cors=power.cork=power.dcor= array(NA, c(8,num.noise))
## We loop through the noise level and functional form; each time we #estimate
a null distribution based on #the marginals of the data, and then #use that
null distribution to estimate power
## We use a uniformly distributed x, because in the original paper the #authors
used the same
for(l in 1:num.noise) {
for(typ in 1:8) {
## This next loop simulates data under the null with the correct marginals (x
is uniform, and y is a function of a #uniform with gaussian noise)
for(ii in 1:nsim) {
x=runif(n)
#lin+noise
if(typ==1) {
y=x+ noise *(l/num.noise)* rnorm(n)
}
#parabolic+noise
if(typ==2) {
y=4*(x-.5)^2+ noise * (l/num.noise) * rnorm(n)
}
#cubic+noise
if(typ==3) {
y=128*(x-1/3)^3-48*(x-1/3)^3-12*(x-1/3)+10* noise * (l/num.noise) *rnorm(n)
}
#sin+noise
if(typ==4) {
y=sin(4*pi*x) + 2*noise * (l/num.noise) *rnorm(n)
}
#their sine + noise
if(typ==5) {
y=sin(16*pi*x) + noise * (l/num.noise) *rnorm(n)
}
#x^(1/4) + noise
if(typ==6) {
y=x^(1/4) + noise * (l/num.noise) *rnorm(n)
}
#circle
if(typ==7) {
y=(2*rbinom(n,1,0.5)-1) * (sqrt(1 - (2*x - 1)^2)) + noise/4*l/num.noise
*rnorm(n)
}
#step function
if(typ==8) {
y = (x > 0.5) + noise*5*l/num.noise *rnorm(n)
}
# We resimulate x so that we have the null scenario
x <- runif(n)
# Calculate the 4 correlations
val.cor[ii]=(cor(x,y))
val.cors[ii]=(cor(x,y,method=c("spearman")))
val.cork[ii]=(cor(x,y,method=c("kendal")))
val.dcor[ii]=dcor(x,y)
}
## Next we calculate our 4 rejection cutoffs
cut.cor=quantile(val.cor,.95)
cut.cors=quantile(val.cors,.95)
cut.cork=quantile(val.cork,.95)
cut.dcor=quantile(val.dcor,.95)
## Next we simulate the data again, this time under the alternative
for(ii in 1:nsim2) {
x=runif(n)
#lin+noise
if(typ==1) {
y=x+ noise *(l/num.noise)* rnorm(n)
}
#parabolic+noise
if(typ==2) {
y=4*(x-.5)^2+ noise * (l/num.noise) * rnorm(n)
}
#cubic+noise
if(typ==3) {
y=128*(x-1/3)^3-48*(x-1/3)^3-12*(x-1/3)+10* noise * (l/num.noise) *rnorm(n)
}
#sin+noise
if(typ==4) {
y=sin(4*pi*x) + 2*noise * (l/num.noise) *rnorm(n)
}
#their sine + noise
if(typ==5) {
y=sin(16*pi*x) + noise * (l/num.noise) *rnorm(n)
}
#x^(1/4) + noise
if(typ==6) {
y=x^(1/4) + noise * (l/num.noise) *rnorm(n)
}
#circle
if(typ==7) {
y=(2*rbinom(n,1,0.5)-1) * (sqrt(1 - (2*x - 1)^2)) + noise/4*l/num.noise
*rnorm(n)
}
#step function
if(typ==8) {
y = (x > 0.5) + noise*5*l/num.noise *rnorm(n)
}
## We again calculate our 4 "correlations"
val.cor2[ii]=(cor(x,y))
val.cors2[ii]=(cor(x,y,method=c("spearman")))
val.cork2[ii]=(cor(x,y,method=c("kendal")))
val.dcor2[ii]=dcor(x,y)
}
## Now we estimate the power as the number of alternative statistics #exceeding
our estimated cutoffs
power.cor[typ,l] <- sum(val.cor2 > cut.cor)/nsim2
power.cors[typ,l] <- sum(val.cors2 > cut.cor)/nsim2
power.cork[typ,l] <- sum(val.cork2 > cut.cor)/nsim2
power.dcor[typ,l] <- sum(val.dcor2 > cut.dcor)/nsim2
}
}
save.image()
## The rest of the code is for plotting the image
pdf(file = "power.pdf")
op <- par(mfrow = c(4,2), cex = 0.45)
plot((1:30)/10, power.cor[1,], ylim = c(0,1), main = "Linear", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[1,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[1,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[1,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[2,], ylim = c(0,1), main = "Quadratic", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[2,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[2,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[2,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[3,], ylim = c(0,1), main = "Cubic", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[3,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[3,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[3,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[5,], ylim = c(0,1), main = "Sine: period 1/8", xlab = "Noise Level", ylab
= "Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[5,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[5,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[5,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[4,], ylim = c(0,1), main = "Sine: period 1/2", xlab = "Noise Level", ylab
= "Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[4,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[4,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[4,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[6,], ylim = c(0,1), main = "X^(1/4)", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[6,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[6,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[6,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[7,], ylim = c(0,1), main = "Circle", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[7,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[7,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[7,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[8,], ylim = c(0,1), main = "Step function", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[8,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[8,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[8,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
par(op)
dev.off()
#################
Le dimanche 9 mai 2021 à 22:44:22 UTC+2, Rui Barradas <ruipbarra...@sapo.pt> a
écrit :
Hello,
You are not closing the pdf device.
The only changes I have made to your code are right at the beginning of
the plotting instructions and at the end of the code.
## The rest of the code is for plotting the image
pdf(file = "power.pdf")
op <- par(mfrow = c(4,2), cex = 0.45)
[...]
par(op)
dev.off()
#################
The comments only line is your last code line.
The result is attached.
Hope this helps,
Rui Barradas
Às 19:39 de 09/05/21, varin sacha via R-help escreveu:
Dear R-experts,
I am trying to get the 8 graphs like the ones in this paper :
https://statweb.stanford.edu/~tibs/reshef/comment.pdf
My R code does not show any error message neither warnings but I d'on't get
what I would like to get (I mean the 8 graphs), so I am missing something.
What's it ? Many thanks for your precious help.
#################
set.seed(1)
library(energy)
# Here we define parameters which we use to simulate the data
# The number of null datasets we use to estimate our rejection reject #regions
for an alternative with level 0.05
nsim=50
# Number of alternative datasets we use to estimate our power
nsim2=50
# The number of different noise levels used
num.noise <- 30
# A constant to determine the amount of noise
noise <- 3
# Number of data points per simulation
n=100
# Vectors holding the null "correlations" (for pearson, for spearman, for
kendall and dcor respectively) for each # of the nsim null datasets at a #given noise
level
val.cor=val.cors=val.cork=val.dcor=rep(NA,nsim)
# Vectors holding the alternative "correlations" (for pearson, for #spearman,
for kendall and dcor respectively) #for each of the nsim2 alternative datasets at a given
noise level
val.cor2=val.cors2=val.cork2=val.dcor2= rep(NA,nsim2)
# Arrays holding the estimated power for each of the 4 "correlation" types, for
each data type (linear, #parabolic, etc...) with each noise level
power.cor=power.cors=power.cork=power.dcor= array(NA, c(8,num.noise))
## We loop through the noise level and functional form; each time we #estimate
a null distribution based on #the marginals of the data, and then #use that
null distribution to estimate power
## We use a uniformly distributed x, because in the original paper the #authors
used the same
for(l in 1:num.noise) {
for(typ in 1:8) {
## This next loop simulates data under the null with the correct marginals (x
is uniform, and y is a function of a #uniform with gaussian noise)
for(ii in 1:nsim) {
x=runif(n)
#lin+noise
if(typ==1) {
y=x+ noise *(l/num.noise)* rnorm(n)
}
#parabolic+noise
if(typ==2) {
y=4*(x-.5)^2+ noise * (l/num.noise) * rnorm(n)
}
#cubic+noise
if(typ==3) {
y=128*(x-1/3)^3-48*(x-1/3)^3-12*(x-1/3)+10* noise * (l/num.noise) *rnorm(n)
}
#sin+noise
if(typ==4) {
y=sin(4*pi*x) + 2*noise * (l/num.noise) *rnorm(n)
}
#their sine + noise
if(typ==5) {
y=sin(16*pi*x) + noise * (l/num.noise) *rnorm(n)
}
#x^(1/4) + noise
if(typ==6) {
y=x^(1/4) + noise * (l/num.noise) *rnorm(n)
}
#circle
if(typ==7) {
y=(2*rbinom(n,1,0.5)-1) * (sqrt(1 - (2*x - 1)^2)) + noise/4*l/num.noise
*rnorm(n)
}
#step function
if(typ==8) {
y = (x > 0.5) + noise*5*l/num.noise *rnorm(n)
}
# We resimulate x so that we have the null scenario
x <- runif(n)
# Calculate the 4 correlations
val.cor[ii]=(cor(x,y))
val.cors[ii]=(cor(x,y,method=c("spearman")))
val.cork[ii]=(cor(x,y,method=c("kendal")))
val.dcor[ii]=dcor(x,y)
}
## Next we calculate our 4 rejection cutoffs
cut.cor=quantile(val.cor,.95)
cut.cors=quantile(val.cors,.95)
cut.cork=quantile(val.cork,.95)
cut.dcor=quantile(val.dcor,.95)
## Next we simulate the data again, this time under the alternative
for(ii in 1:nsim2) {
x=runif(n)
#lin+noise
if(typ==1) {
y=x+ noise *(l/num.noise)* rnorm(n)
}
#parabolic+noise
if(typ==2) {
y=4*(x-.5)^2+ noise * (l/num.noise) * rnorm(n)
}
#cubic+noise
if(typ==3) {
y=128*(x-1/3)^3-48*(x-1/3)^3-12*(x-1/3)+10* noise * (l/num.noise) *rnorm(n)
}
#sin+noise
if(typ==4) {
y=sin(4*pi*x) + 2*noise * (l/num.noise) *rnorm(n)
}
#their sine + noise
if(typ==5) {
y=sin(16*pi*x) + noise * (l/num.noise) *rnorm(n)
}
#x^(1/4) + noise
if(typ==6) {
y=x^(1/4) + noise * (l/num.noise) *rnorm(n)
}
#circle
if(typ==7) {
y=(2*rbinom(n,1,0.5)-1) * (sqrt(1 - (2*x - 1)^2)) + noise/4*l/num.noise
*rnorm(n)
}
#step function
if(typ==8) {
y = (x > 0.5) + noise*5*l/num.noise *rnorm(n)
}
## We again calculate our 4 "correlations"
val.cor2[ii]=(cor(x,y))
val.cors2[ii]=(cor(x,y,method=c("spearman")))
val.cork2[ii]=(cor(x,y,method=c("kendal")))
val.dcor2[ii]=dcor(x,y)
}
## Now we estimate the power as the number of alternative statistics #exceeding
our estimated cutoffs
power.cor[typ,l] <- sum(val.cor2 > cut.cor)/nsim2
power.cors[typ,l] <- sum(val.cors2 > cut.cor)/nsim2
power.cork[typ,l] <- sum(val.cork2 > cut.cor)/nsim2
power.dcor[typ,l] <- sum(val.dcor2 > cut.dcor)/nsim2
}
}
save.image()
## The rest of the code is for plotting the image
pdf("power.pdf")
par(mfrow = c(4,2), cex = 0.45)
plot((1:30)/10, power.cor[1,], ylim = c(0,1), main = "Linear", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[1,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[1,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[1,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[2,], ylim = c(0,1), main = "Quadratic", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[2,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[2,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[2,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[3,], ylim = c(0,1), main = "Cubic", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[3,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[3,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[3,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[5,], ylim = c(0,1), main = "Sine: period 1/8", xlab = "Noise Level", ylab
= "Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[5,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[5,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[5,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[4,], ylim = c(0,1), main = "Sine: period 1/2", xlab = "Noise Level", ylab
= "Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[4,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[4,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[4,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[6,], ylim = c(0,1), main = "X^(1/4)", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[6,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[6,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[6,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[7,], ylim = c(0,1), main = "Circle", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[7,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[7,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[7,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
plot((1:30)/10, power.cor[8,], ylim = c(0,1), main = "Step function", xlab = "Noise Level", ylab =
"Power", pch = 1, col = "black", type = 'b')
points((1:30)/10, power.cors[8,], pch = 2, col = "green", type = 'b')
points((1:30)/10, power.cork[8,], pch = 3, col = "blue", type = 'b')
points((1:30)/10, power.dcor[8,], pch = 4, col = "red", type = 'b')
legend("topright",c("cor pearson","cor spearman", "cor kendal","dcor"), pch = c(1,2,3), col =
c("black","green","blue","red"))
#################
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.
