R/comparison.R

In hoenlab/SBIC: Structural Bayesian Information Criterion for Graphical Models

#' @name comparison 
#' @aliases comparison
#' @title Comparing the real and estimated adjacency matrix  
#' @description Comparing the two adjacency matrices for false discovery rate and positive selection rate. Used
#'   for model validation 
#' @author Jie Zhou
#' @param real The real matrix \code{p} by \code{p} adjacency matrix likely from simulated data 
#' @param estimate The estimated matrix \code{p} by \code{p} adjacency matrix likely estimated using the SBIC procedure
#' 
#' @return 
#'   A list of the following evaluation metrics 
#'   \item{PSR}{Positive Selection Rate}
#'   \item{FDR}{False Discovery rate}
#' @export 

comparison=function(real, estimate){
  ## M1 is the true net matrix
  ##Me is the estimated net matrix
  real=real+t(real)
  diag(real)=1
  estimate=estimate+t(estimate)
  diag(estimate)=1
  N1=ifelse(real==0,0,1)
  N2=ifelse(estimate==0,0,1)
   if (any(dim(N1)!=dim(N2))[1])
  stop("Two matrixes should have the same dimension")
  p=dim(real)[1]
  real=(sum(N1)-p)/2
  select=(sum(N2)-p)/2
  real_select=(sum(N2[N1==1])-p)/2
  fause_select=sum(N2[N1==0])/2
  if (real==0){return( list(PSR=1, FDR=0))}
  if (select==0) {return(list(PSR=0, FDR=0))}
  PSR=real_select/real
  FDR=fause_select/select
    return(list(PSR=PSR, FDR=FDR)) 
}