R/NNLS_MF.R

In phytoclass: Estimate Chla Concentrations of Phytoplankton Groups

#' Performs the non-negative matrix factorisation for given phytoplankton 
#' pigments and pigment ratios, to attain an estimate of phytoplankton 
#' class abundances.
#'
#' @param Fn   Pigment to Chl a matrix
#' @param S   Sample data matrix – a matrix of pigment samples
#' @param cm  Weights for each column
#'
#' @return A list containing 
#' \enumerate{
#'  \item The F matrix (pigment: Chl *a*) ratios
#'  \item The root mean square error (RMSE)
#'  \item The C matrix (class abundances for each group)
#'  }
#' @export
#'
#' @examples
#' MC <- Matrix_checks(Sm,Fm)
#' Snew <- MC$Snew
#' Fnew <- MC$Fnew
#' cm <- Bounded_weights(Snew, weight.upper.bound = 30)
#' NNLS_MF(Fnew, Snew, cm)
#'
NNLS_MF <- function(Fn, S, cm=NULL){
    if (is.null(cm)) {
    cm <- as.vector(rep(1,ncol(S)))
  }


  b <- crossprod(t(Weight_error(Fn, cm)),t(Weight_error(S, cm)))
  C_new2 <-t(RcppML::nnls(crossprod(t(Weight_error(Fn, cm))),
                          b, 
                          cd_maxit = 1000,cd_tol =1e-8 ))
  C_new2 <- as.matrix(C_new2)
  Cn.s2 <- rowSums(C_new2)
  Cn2 <- C_new2/Cn.s2 #Row sums to one
  Cn2 <- as.matrix(Cn2)
  colnames(Cn2) <- rownames(Fn)
  error <- Metrics::rmse((S),C_new2%*%(Fn))
  return(list("F matrix " = Fn, "RMSE" = error,"C matrix" = Cn2))
}