CODEWITHSUNDEEP
rmatrix-multiplication
MNU
MNU

Reputation: 764

How to multiple each row of one matrix with another matrix in R?

For the following two matrix, 

yy=matrix(c(1:40), nrow = 10, ncol = 8)
tt=diag(1:4)

I would like to create a new matrix yy_new=matrix(NA, nrow = 10, ncol=ncol(tt)) by multiplying each row and first 4 column of yy by tt . Example, for the first row yy_new=yy[1,1:4]%*%tt, second row is yy_new=yy[2,1:4]%*%tt. Finally I want the mean of yy_new at each column as yy_new=apply(yy_new,2,mean). The folowing loop is working well, but for large data set it is time comsuming.

 yy_new=matrix(NA, nrow = 10, ncol=ncol(tt))         
 for ( it in 1:10){
      for ( tim in 1:4){
        yy_new[it, tim]=yy[it,tim]*tt[tim,tim]
      }
    }
yy_new=apply(yy_new,2,mean)

similarly, I want another matrix yy_new1 by considering the last four column of yy

 yy_new1=matrix(NA, nrow = 10, ncol=ncol(tt))

How can I do it efficiently using any built-in function or customized function? Any help is appreciated.

Upvotes: 1

Views: 658

Answers (1)

Maurits Evers
Maurits Evers

Reputation: 50668

Here is a shorter (and faster) version for yy_new

yy_new <- rowMeans(apply(yy[, 1:4], 1, function(row) row %*% tt))

Similarly for the last 4 columns of yy

yy_new1 <- rowMeans(apply(yy[, (ncol(yy)-3):ncol(yy)], 1, function(row) row %*% tt))

Note that rowMeans and colMeans are generally faster than apply(..., 1, mean) and apply(..., 2, mean).

Here are results from a microbenchmark comparison

library(microbenchmark)
res <- microbenchmark(
    rowMeans_apply = {
        yy_new = rowMeans(apply(yy[, 1:4], 1, function(row) row %*% tt))
    },
    for_loop = {
        yy_new=matrix(NA, nrow = 10, ncol=ncol(tt))
         for ( it in 1:10){
              for ( tim in 1:4){
                yy_new[it, tim]=yy[it,tim]*tt[tim,tim]
              }
            }
    }
)
res
#Unit: microseconds
#           expr      min       lq      mean   median       uq       max neval
# rowMeans_apply   73.148   82.097  116.8959  101.329  123.863  1348.141   100
#       for_loop 3985.521 4141.633 5017.9808 4421.285 5020.425 18574.364   100

Update

In response to your comment, you could do something like this:

f <- function(x) rowMeans(apply(x, 1, function(row) row %*% tt))
sapply(split.default(as.data.frame(yy), rep(1:2, each = 4)), f)
#         1     2
#[1,]   5.5   5.5
#[2,]  31.0  31.0
#[3,]  76.5  76.5
#[4,] 142.0 142.0

Explanation: split.default here splits the data.frame into the first 4 and last 4 columns and stores them as two data.frames in a list; then we use sapply to loop through the list elements and calculate the required quantity as requested. The resulting output object is a matrix.

Upvotes: 1

Related Questions