R/ilr.R

In OxWearables/epicoda: Supports epidemiological analyses using compositional exposure variables

#' ilr_trans: Performs ilr transformations using pivot coordinates.
#'
#' Takes compositional columns and returns them after ilr transformation using pivot coordinates.
#'
#' This doesn't use compositions::ilr to insulate it from changes in the choice of
#' basis used. As pivot coordinates are currently standard in physical behaviour epidemiology this
#' is specified.
#'
#' @param data Compositional columns to transform.
#' @return ilr-transformed columns.
ilr_trans <- function(data){
  dClr <- clr_trans(data)
  mClr <- as.matrix(dClr)
  transformation_matrix <- create_transformation_matrix(ncol(data))
  dTransformed <- data.frame(mClr %*% transformation_matrix)
  colnames(dTransformed) <- paste0("ilr_", 1:(ncol(data)-1))
  return(dTransformed)
}


#' ilr_trans_inv: Inverts ilr transformation (pivot coordinates)
#'
#' Takes ilr-transformed columns and inverts the ilr transformation.
#'
#' @param data Dataset to have the transformation reversed.
#' @return Compositional columns
ilr_trans_inv <- function(data){
  transformation_matrix <- create_transformation_matrix(ncol(data) + 1)
  transformation_matrix_t <- t(transformation_matrix)

  matrixA <- rbind(transformation_matrix_t, 1)
  aug_data <- cbind(data, "zero_col" = 0)

  Ainv <- solve(matrixA)

  data_ready <- t(as.matrix(aug_data))
  dUntransformed <- data.frame(t(Ainv %*% data_ready))
  colnames(dUntransformed) <- paste0("clr_comp_", 1:(ncol(dUntransformed)-1))
  dReversed <- clr_trans_inv(dUntransformed)

  return(dReversed)
}