I am converting a list of numbers to a matrix. The list of numbers is encoded text. Each word has a number associated with it, like 'the': 1, 'it': 2, etc. I want to get a matrix of values where the presence of an encoded word is represented by a '1'. So if one of our encoded texts looked like:
c(1, 4, 2)
Then the corresponding matrix (with a max word index of 10) would look like:
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
[1,] 1 1 0 1 0 0 0 0 0 0
So here is how I have it currently:
encoded.text <- list(c(1, 3, 2), c(1, 7, 8))
result <- matrix(0, nrow = length(encoded.text), ncol = 10)
for (i in 1:length(encoded.text)) {
result[i, encoded.text[[i]]] <- 1
}
I'm wondering, is there a better/more efficient way than the for loop to do this?
Here is one option with row/column indexing. We unlist the 'encoded.text' for the column index, while replicate the sequence of the list with the lengths of the list as row index. cbind it to make a row/column index matrix, extract the values of 'result' based on the index and assign it to 1
m1 <- cbind(rep(seq_along(encoded.text), lengths(encoded.text)),
unlist(encoded.text))
result[m1] <- 1
result
# [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
#[1,] 1 1 1 0 0 0 0 0 0 0
#[2,] 1 0 0 0 0 0 1 1 0 0
NOTE: The apply/Map etc are just loops as for loop. It wouldn't give any peformance increment except just to add as an answer
n <- 1e6
test <- rep(encoded.text, n)
testresult <- matrix(0, nrow = length(test), ncol = 10)
testresult2 <- copy(testresult)
testresult3 <- copy(testresult)
system.time({
m2 <- cbind(rep(seq_along(test), lengths(test)),
unlist(test))
testresult[m2] <- 1
})
# user system elapsed
# 0.290 0.098 0.388
system.time({
testresult2[do.call(rbind, Map(cbind, seq_len(length(test)), test))] <- 1
})
# user system elapsed
# 8.383 0.462 8.787
system.time({
for (i in 1:length(test)) {
testresult3[i, test[[i]]] <- 1
}
})
# user system elapsed
# 0.648 0.131 0.778
If we increase 'n' and rerun again (after constructing the data)
n <- 1e7
system.time({
m2 <- cbind(rep(seq_along(test), lengths(test)),
unlist(test))
testresult[m2] <- 1
})
# user system elapsed
# 2.699 1.225 3.990 # almost 2 times efficient now
system.time({
testresult2[do.call(rbind, Map(cbind, seq_len(length(test)), test))] <- 1
})
# user system elapsed
# 88.584 5.047 94.384
system.time({
for (i in 1:length(test)) {
testresult3[i, test[[i]]] <- 1
}
})
# user system elapsed
# 5.734 0.742 6.461
-microbenchmark on n <- 1e7 constructed data
ak <- function() {
m2 <- cbind(rep(seq_along(test), lengths(test)),
unlist(test))
testresult[m2] <- 1
}
wfw <- function() {
for (i in 1:length(test)) {
testresult3[i, test[[i]]] <- 1
}
}
library(microbemchmark)
microbenchmark(ak(), wfw(), unit = 'relative', times = 20L)
#Unit: relative
# expr min lq mean median uq max neval cld
# ak() 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 20 a
# wfw() 1.946415 1.945528 1.927263 1.926645 1.910907 1.940207 20 b