mult_mat_ref: Multiply the rows or columns of a _matrix_ times a _vector_,...

In algoquant/HighFreq: High Frequency Time Series Management

Multiply the rows or columns of a matrix times a vector, element-wise and in place, without copying the data in memory.

Description

Multiply the rows or columns of a matrix times a vector, element-wise and in place, without copying the data in memory.

Usage

mult_mat_ref(vectorv, matrixv, byrow = TRUE)

Arguments

vectorv	A numeric vector.
matrixv	A numeric matrix.
byrow	A Boolean argument: if TRUE then multiply the rows of matrixv by vectorv, otherwise multiply the columns (the default is byrow = TRUE.)

Details

The function mult_mat_ref() multiplies the rows or columns of a matrix times a vector, element-wise and in place, without copying the data in memory.

It accepts a pointer to the argument matrixv, and it overwrites the old matrix values with the new values. It performs the calculation in place, without copying the matrix in memory, which can significantly increase the computation speed for large matrices.

If byrow = TRUE (the default), then function mult_mat_ref() multiplies the rows of the argument matrixv times the argument vectorv. Otherwise it multiplies the columns of matrixv.

In R, matrix multiplication is performed by columns. Performing multiplication by rows is often required, for example when multiplying asset returns by portfolio weights. But performing multiplication by rows requires explicit loops in R, or it requires matrix transpose. And both are slow.

The function mult_mat_ref() uses RcppArmadillo C++ code, so when multiplying large matrix columns it's several times faster than vectorized R code, and it's even much faster compared to R when multiplying the matrix rows.

The function mult_mat_ref() performs loops over the matrix rows and columns using the Armadillo operators each_row() and each_col(), instead of performing explicit for() loops (both methods are equally fast).

Value

Void (no return value - modifies the data in place).

Examples

## Not run: 
# Create vector and matrix data
matrixv <- matrix(round(runif(25e4), 2), nc=5e2)
vectorv <- round(runif(5e2), 2)

# Multiply the matrix rows using R
matrixr <- t(vectorv*t(matrixv))
# Multiply the matrix rows using C++
HighFreq::mult_mat_ref(vectorv, matrixv, byrow=TRUE)
all.equal(matrixr, matrixv)
# Compare the speed of Rcpp with R code
library(microbenchmark)
summary(microbenchmark(
    Rcpp=HighFreq::mult_mat_ref(vectorv, matrixv, byrow=TRUE),
    Rcode=t(vectorv*t(matrixv)),
    times=10))[, c(1, 4, 5)]  # end microbenchmark summary
    
# Multiply the matrix columns using R
matrixr <- vectorv*matrixv
# Multiply the matrix columns using C++
HighFreq::mult_mat_ref(vectorv, matrixv, byrow=FALSE)
all.equal(matrixr, matrixv)
# Compare the speed of Rcpp with R code
library(microbenchmark)
summary(microbenchmark(
    Rcpp=HighFreq::mult_mat_ref(vectorv, matrixv, byrow=FALSE),
    Rcode=vectorv*matrixv,
    times=10))[, c(1, 4, 5)]  # end microbenchmark summary

## End(Not run)

algoquant/HighFreq documentation built on Feb. 9, 2024, 8:15 p.m.