blockmult: Block matrix multiplication with Delayed Array Object

In BigDataStatMeth: Statistical Methods and Algorithms for Big Data

Block matrix multiplication with Delayed Array Object

Description

This function performs a block matrix-matrix multiplication with numeric matrix or Delayed Arrays

Usage

blockmult(
  a,
  b,
  block_size = NULL,
  paral = NULL,
  threads = NULL,
  bigmatrix = NULL,
  mixblock_size = NULL,
  outfile = NULL,
  onmemory = NULL
)

Arguments

a	a double matrix.
b	a double matrix.
block_size	(optional, defalut = 128) block size to make matrix multiplication, if 'block_size = 1' no block size is applied (size 1 = 1 element per block)
paral,	(optional, default = TRUE) if paral = TRUE performs parallel computation else performs seria computation
threads	(optional) only if bparal = true, number of concurrent threads in parallelization if threads is null then threads = maximum number of threads available
bigmatrix	(optiona, default = 5000) maximum number of rows or columns to consider as big matrix and work with hdf5 files, by default a matrix with more than 5000 rows or files is considered big matrix and computation is made in disk
mixblock_size	(optiona, default = 128), only if we are working with big matrix and parallel computation = true. Block size for mixed computation in big matrix parallel. Size of the block to be used to perform parallelized memory memory of the block read from the disk being processed.
outfile	(optional) file name to work with hdf5 if we are working with big matrix in disk.
onmemory	(optional) if onmemory = TRUE the multiplication is forced to execute in memory

Value

A List with :

• "matrix" Result matrix if execution has been performed in memory

• "filename" HDF5 filename if execution has been performed in disk, HDF5 file contains :

  • "INPUT"hdf5 group with input matrix A and B

  • "OUTPUT"hdf5 group with output matrix C

  with input and output matrix.

Examples

# with numeric matrix
m <- 500
k <- 1500
n <- 400
A <- matrix(rnorm(n*k), nrow=n, ncol=k)
B <- matrix(rnorm(n*k), nrow=k, ncol=n)

C <- blockmult(A,B,128, TRUE)

BigDataStatMeth documentation built on March 30, 2022, 1:07 a.m.