Home Select various cells according to different combinations of columns and rows

Questions

Select various cells according to different combinations of columns and rows

January 13, 2024

I have vector of index, that I’d like to filter by. In each column a different row will be selected. The output will be one vector of filtered values with the length of the number of columns. Bud without using the slow cycle. For example

set.seed(123)
(M <- matrix(rnorm(25), 5))

            [,1]       [,2]       [,3]       [,4]       [,5]
[1,] -0.56047565  1.7150650  1.2240818  1.7869131 -1.0678237
[2,] -0.23017749  0.4609162  0.3598138  0.4978505 -0.2179749
[3,]  1.55870831 -1.2650612  0.4007715 -1.9666172 -1.0260044
[4,]  0.07050839 -0.6868529  0.1106827  0.7013559 -0.7288912
[5,]  0.12928774 -0.4456620 -0.5558411 -0.4727914 -0.6250393

indíces <- c(2, 3, 1, 4, 4)
vect <- c()

for(i in 1:5) {
  vect <- c(vect, M[indíces[i], i])
}
vect
[1] -0.2301775 -1.2650612  1.2240818  0.7013559 -0.7288912

I have larger data set and for cycle isn’t therefore ideal. But I couldn’t think of anything better, or find anything better.

>Solution :

We can use a vectorized operation using cbind: M[cbind(indices, 1:ncol(M))]

set.seed(123)
(M <- matrix(rnorm(25), 5))
indices <- c(2, 3, 1, 4, 4)

# vectorized
vect1 <- M[cbind(indices, 1:ncol(M))]

# loop
vect <- c()

for(i in 1:5) {
  vect <- c(vect, M[indíces[i], i])
}

# benchmark
library(microbenchmark)
mbm = microbenchmark(
vectorized =  M[cbind(indices, 1:ncol(M))],
loop = for(i in 1:5) {
  vect <- c(vect, M[indíces[i], i])
},
times=50
)
mbm

Unit: microseconds
       expr    min     lq     mean  median     uq    max neval cld
 vectorized    2.1    2.8    5.158    6.30    6.5   15.6    50  a 
       loop 2062.8 2089.2 2143.418 2114.55 2183.8 2391.0    50   b
autoplot(mbm)