R/reduce_to_sparse_matrix.R

In namgillee/BTTSoftImpute: R Software for Block Tensor Train Decomposition for Missing Data Estimation

#' reduce_to_sparse_matrix
#'
#' Convert a sparse tensor X into the sparse matrix format.
#'
#' @param X X has attributes $i, $j, $v, $nrow, $ncol, $dimnames
#' @return Output X has the same attributes as the input X, But $i and $j are forced to be in vector instead of matrix, and $nrow and $ncol are forced to be scalars instead of vectors.
#' @export
reduce_to_sparse_matrix <- function(X)
{
  M = length(X$nrow)
  N = length(X$ncol)

  if (M>1) {
    idata = X$i  #Consider that idata is a matrix
    cum_prod_rowsizes = cumprod(c(1,X$nrow[-M]))  #(1, J{1}, J{1}J{2}, ..., J{1}***J{N-1})
    idata = colSums( t(idata-1) *  cum_prod_rowsizes ) + 1 # SUM{ (j{n}-1)*J{1}J{2}*...*J{n-1} } + 1
    X$i = idata     #matrix(idata, nrow = length(idata))

    X$nrow = prod(X$nrow)
  } else if (is.array(X$i)) {
    X$i = as.vector(X$i)
  }

  if (N>1) {
    jdata = X$j  #Consider that jdata is a matrix
    cum_prod_colsizes = cumprod(c(1,X$ncol[-N]))  #(1, J{1}, J{1}J{2}, ..., J{1}***J{N-1})
    jdata = colSums( t(jdata-1) *  cum_prod_colsizes ) + 1 # SUM{ (j{n}-1)*J{1}J{2}*...*J{n-1} } + 1
    X$j = matrix(jdata, nrow = length(jdata))

    X$ncol = prod(X$ncol)
  } else if (is.array(X$j)) {
    X$j = as.vector(X$j)
  }

  #
  # X$dimnames  :  how to use it is not specified yet.
  #

  return(X)

}