CODEWITHSUNDEEP
r
user33484
user33484

Reputation: 568

removing for loop with sample function

I have the following data structure:

set.seed(100)
x <- data.frame("smp_1"=runif(20)*100,"smp_2"=runif(20)*99)
x["weight_1"] = x$smp_1/sum(x$smp_1)
x["weight_2"] = x$smp_2/sum(x$smp_2)


> head(x)
     smp_1     smp_2   weight_1    weight_2
1 66.61718 68.976341 0.05721288 0.061115678
2 24.65804 77.966842 0.02117709 0.069081607
3 66.10397  1.611913 0.05677212 0.001428216
4 93.95866  1.793973 0.08069459 0.001589529
5 19.96638 31.008240 0.01714774 0.027474488
6 66.35187 97.033923 0.05698502 0.085975770

now I want to create a new data frame which samples from each smp column using the weight columns as the probabilities and add each column sample into a new data frame and a new column. I can do this using a for loop:

tempdf <- data.frame(matrix(0,ncol=0,nrow=1000))
for (k in 1:2){
  tempdf[,paste0("sim_",k)] <- sample(x[,paste0("smp_",k)],size=1000, replace=T, prob = x[,paste0("weight_",k)])
}

my question is how can I do this without a for loop in a more efficient way? I will be sampling 100k of multiple columns so I need something quite quick.

Upvotes: 2

Views: 57

Answers (3)

akrun
akrun

Reputation: 887118

Here is one option with tidyverse using map2, we subset the columns 'smp', 'weight', and use the correspoing 'weight' to sample the 'smp' columns

library(tidyverse)
map2_df(x %>% 
          dplyr::select(matches("^smp")), 
       x %>% 
          dplyr::select(matches("^weight")), ~ 
     sample(.x, size = 1000, replace = TRUE, prob = .y))

Upvotes: 0

Ronak Shah
Ronak Shah

Reputation: 388982

In base R, we can separate columns for "smp" and weigths and use mapply (which BTW internally is still a loop) to sample values. 

sample_col <- grep("^smp", names(x))
weigth_col <- grep("^weight", names(x))

mapply(function(p, q) sample(p, size = 1000, replace = TRUE, prob = q), 
                       x[,sample_col], x[,weigth_col])


#          smp_1     smp_2
# [1,] 62.499648 74.148250
# [2,] 88.216552 94.461613
# [3,] 55.232243 70.369581
# [4,] 28.035384 74.148250
# [5,] 39.848790 76.259859
# [6,] 39.848790 97.966850
# [7,] 88.216552 91.922002
# [8,] 20.461216 97.966850
# [9,] 66.902171 53.045304
#[10,] 54.655860 76.259859
#...

Upvotes: 0

Wimpel
Wimpel

Reputation: 27732

Here is a data.table approach.

In the answer ans, the variable-value (1 or 2) is your k.

library(data.table)
#melt to long format
DT <- melt( setDT(x) , 
            id.vars = NULL, 
            measure.vars = patterns( smp = "^smp", 
                                     weight = "^weight"))
#pull samples
ans <- DT[ , .( sim = sample( smp, 
                              size = 1000, 
                              replace = TRUE, 
                              prob = weight)), 
           by = .(variable) ]


#    variable      sim
# 1:        1 69.02905
# 2:        1 30.77661
# 3:        1 37.03205
# 4:        1 35.75249
# 5:        1 48.37707
# 6:        1 55.23224

Upvotes: 2

Related Questions