bdDiag_multiply_hdf5

bdDiag_multiply_hdf5

HDF5_ALGEBRA

1 Description

Performs optimized diagonal multiplication between two datasets stored in HDF5 format. Automatically detects whether inputs are matrices (extracts diagonals) or vectors (direct operation) and uses the most efficient approach. This function performs element-wise multiplication and is ~50-250x faster than traditional matrix operations.

2 Usage

bdDiag_multiply_hdf5(filename, group, A, B, groupB = NULL, target = NULL, outgroup = NULL, outdataset = NULL, paral = NULL, threads = NULL, overwrite = NULL)

3 Arguments

Parameter Description
filename String. Path to the HDF5 file containing the datasets.
group String. Group path containing the first dataset (A).
A String. Name of the first dataset (matrix or vector).
B String. Name of the second dataset (matrix or vector).
groupB Optional string. Group path containing dataset B. If NULL, uses same group as A.
target Optional string. Where to write result: “A”, “B”, or “new” (default: “new”).
outgroup Optional string. Output group path. Default is “OUTPUT”.
outdataset Optional string. Output dataset name. Default is “A_*_B” with .diag suffix if appropriate.
paral Optional logical. Whether to use parallel processing. Default is FALSE.
threads Optional integer. Number of threads for parallel processing. If NULL, uses maximum available threads.
overwrite Optional logical. Whether to overwrite existing datasets. Default is FALSE.

4 Value

List with components:

  • fn: Character string with the HDF5 filename
  • ds: Character string with the full dataset path to the diagonal multiplication result (group/dataset)

5 Details

This function provides flexible diagonal multiplication with automatic optimization: - Operation modes: - Matrix * Matrix: Extract diagonals → vector multiplication → save as vector - Matrix * Vector: Extract diagonal → vector multiplication → save as vector
- Vector * Vector: Direct vector multiplication (most efficient) - Performance features: - Uses optimized vector operations for maximum efficiency - Automatic type detection and dimension validation - Memory-efficient processing for large datasets - Parallel processing support for improved performance - Mathematical properties: - Element-wise multiplication (not matrix multiplication) - Commutative operation: A * B = B * A - Handles overflow according to IEEE 754 standards - Preserves sign information correctly

6 Examples

Code
library(BigDataStatMeth)

# Create test matrices
N <- 1000
set.seed(123)
A <- matrix(rnorm(N*N), N, N)
B <- matrix(rnorm(N*N), N, N)

# Save to HDF5
bdCreate_hdf5_matrix("test.hdf5", A, "data", "matrixA",
                     overwriteFile = TRUE)
bdCreate_hdf5_matrix("test.hdf5", B, "data", "matrixB",
                     overwriteFile = FALSE)

# Multiply diagonals (element-wise)
result <- bdDiag_multiply_hdf5("test.hdf5", "data", "matrixA", "matrixB",
                              outgroup = "results",
                              outdataset = "diagonal_product",
                              paral = TRUE)