不同的 set.seed 每次运行在 R

Question

I want to "measure" which Regression Method is more robust to the outliers.

For this, I sum the variances of model coefficients. Each run, I generate data from t-distribution. I set.seed Ten times to have Ten specific data.

However, I also want to have Ten different seed each run. So, in total, I will have 10 sums of the variances. The code below is giving me one sum of the first (Ten different seed).

How can I do this?

#######################################
p <- 5
n <- 50
#######################################
FX <- function(seed, data) {
#for loops over a seed #
for (i in seed) {
set.seed(seed)  
# generating data from t-distribution #
x<- matrix(rt(n*p,1), ncol = p)
y<-rt(n,1)
dat=cbind(x,y)
data<-as.data.frame(dat)
# performing a regression model on the data #
lm1 <- lm(y ~ ., data=data)
lm.coefs <- coef(lm1)
            
lad1 <- lad(y ~ ., data=data, method="BR")
lad.coefs <- coef(lad1)
          }
# calculate variance of the coefficients # 
return(`attr<-`(cbind(lmm=var(lm.coefs), lad=var(lad.coefs)), "seed", seed))
}
#######################################
seeds <- 1:10  ## 10 set seed to have diffrent data set from t-distribution #
res <- lapply(seeds, FX, data=data) # 10 diffrent variance of 10 data/model
sov <- t(sapply(res, colSums)) # put them in matrix
colSums(sov) # sum of 10 varainnces for each model.

   

Answer 1

Here is something closer to your intended results. The code below fixes a key issues from your original code. It was not clear on what data was intended to be returned from the function.

1. This creates a vector of seeds numbers inside the function

2. This also creates a vector to inside the function to store the value of the variance of coefficients for each iteration of the loop. (not sure if is what you want).

3. I needed to comment out the lad function since I do not know which package this is from. (you would need to follow 2 from above to add this back in.

4. Some general clean of the code

   p <- 5 n <- 50 FX <- function(seed, data) { #for loops over a seed # #Fixes the starting seed issue startingSeed <- (seed-1)*10 +1 seeds <- seq( startingSeed, startingSeed+9) #create vector to store results from loop iteration lm.coefs <- vector(mode="numeric", length=10) index <- 1 for (i in seeds) { set.seed(i) # generating data from t-distribution # x<- matrix(rt(n*p,1), ncol = p) y<-rt(n,1) data<-data.frame(x, y) # performing a regression model on the data # lm1 <- lm(y ~., data=data) lm.coefs[index] <- var(coef(lm1)) # lad1 <- lad(y ~., data=data, method="BR") # lad.coefs <- coef(lad1) index <- index +1 } # calculate variance of the coefficients # return(`attr<-`(cbind(lmm=lm.coefs), "seed", seed)) } seeds <- 1:10 ## 10 set seed to have diffrent data set from t-distribution # res <- lapply(seeds, FX, data=data) # 10 diffrent variance of 10 data/model sov <- t(sapply(res, colSums)) # put them in matrix colSums(sov) # sum of 10 varainnces for each model.

Hope this provides the answer or at least guidance to solve your problem.