Weighted sampling in R

Question

I have a data frame data. At each row i have assigned a weight that is in data$ww. Now I would like to make a sample new_data of data, weighted by df$ww. I have tried with subset but it very slow.

# sample data
data <- data.frame(var1 = log(sample(1:5000)))
ndata <- nrow(data)
maxW <- max(data$var1)

nsample <- 4000
rr <- runif(ndata)
data$ww <- cumsum(exp(data$var1))
new_data <- data[0, ]
i <- 1
while(nrow(new_data) < nsample) {
  new_data[i, ] <- subset(data, data$ww > rr[i] * maxW)[1,]
  i <- i + 1
}

Is there a faster way?

Solution

Use the prob argument of sample():

samp_idx <- sample(seq_len(nrow(data)), nsample, prob=data$ww)
new_data <- data[samp_idx, ]

Something like this. Running time is

# user  system elapsed 
# 0.015   0.000   0.014 

versus your version:

# user  system elapsed 
# 4.278   0.007   4.290 

Comments

[isomorphismes]

Why is this question heavily downvoted?