How to scale a NumericMatrix in-place with Rcpp?

问题

This is what I'm doing now

library(Rcpp)

A <- diag(c(1.0, 2.0, 3.0))
rownames(A) <- c('X', 'Y', 'Z')
colnames(A) <- c('A', 'B', 'C')

cppFunction('
void scaleMatrix(NumericMatrix& A, double x) {
    A = A * x;
}')

Unfortunately It doesn't work :(

> A
  A B C
X 1 0 0
Y 0 2 0
Z 0 0 3
> scaleMatrix(A, 2)
> A
  A B C
X 1 0 0
Y 0 2 0
Z 0 0 3

I learned from Rcpp FAQ, Question 5.1 that Rcpp should be able to change the object I passed by value. Stealing an example from Dirk's answer to my previous question:

> library(Rcpp)
> cppFunction("void inplaceMod(NumericVector x) { x = x * 2; }")
> x <- as.numeric(1:5)
> inplaceMod(x)
> x
[1]  2  4  6  8 10

I'm confused: it is possible to modify a NumericVector in-place, but not a NumericMatrix?

回答1:

You can preserve the row and column names by using NumericVector instead of NumericMatrix, keeping in mind that a matrix in R is just a vector with attached dimensions. You can do this switch either when going from R to C++ (scaleVector below) or within C++ (scaleMatrix below taken from a now deleted answer by @Roland):

library(Rcpp)
cppFunction('
NumericVector scaleVector(NumericVector& A, double x) {
    A = A * x;
    return A;
}')

cppFunction('
NumericMatrix scaleMatrix(NumericMatrix& A, double x) {
    NumericVector B = A;
    B = B * x;
    return A;
}')

If one applies these two function to your matrix, the row and column names are preserved. However, the matrix is not changed in place:

A <- diag(1:3)
rownames(A) <- c('X', 'Y', 'Z')
colnames(A) <- c('A', 'B', 'C')

scaleMatrix(A, 2)
#>   A B C
#> X 2 0 0
#> Y 0 4 0
#> Z 0 0 6
scaleVector(A, 2)
#>   A B C
#> X 2 0 0
#> Y 0 4 0
#> Z 0 0 6
A
#>   A B C
#> X 1 0 0
#> Y 0 2 0
#> Z 0 0 3

The reason for that is that diag(1:3) is actually an integer matrix, so a copy is made when you transfer it to a numeric matrix (or vector):

is.integer(A)
#> [1] TRUE

If one uses a numeric matrix to begin with, modification is done in place:

A <- diag(c(1.0, 2.0, 3.0))
rownames(A) <- c('X', 'Y', 'Z')
colnames(A) <- c('A', 'B', 'C')

scaleMatrix(A, 2)
#>   A B C
#> X 2 0 0
#> Y 0 4 0
#> Z 0 0 6
scaleVector(A, 2)
#>   A B  C
#> X 4 0  0
#> Y 0 8  0
#> Z 0 0 12
A
#>   A B  C
#> X 4 0  0
#> Y 0 8  0
#> Z 0 0 12

来源：https://stackoverflow.com/questions/57818049/how-to-scale-a-numericmatrix-in-place-with-rcpp