R/compare_struct.R

In CorReg: Linear Regression Based on Linear Structure Between Variables

#' To compare sub-regression structures
#' 
#' @description Compares two sub-regression structures, considering one of them as the "true one".
#' @param trueZ first structure (binary adjacency matrix)
#' @param Zalgo second structure (binary adjacency matrix)
#' @param all (boolean) Also compute the ratio for each stat.
#' @param mode how to modify the structures before comparison. mode=c("NULL","hybrid","clique","sym")
#' It allows to compare groups instead of exact sub-regressions. Does nothing by default.
#' 
#' @return
#' \item{true1}{ Number of links that exist in both matrices }
#' \item{false1}{ Number of links that exist only in Zalgo }
#' \item{false0}{ Number of links that exist only in trueZ }
#' \item{deltadr}{Number of sub-regressions in trueZ -  Number of sub-regressions in Zalgo (i.e.: negative if too much sub-regressions in Zalgo)}
#' \item{true_left}{ Number of variables redundant in both matrices }
#' \item{false_left}{ Number of variables redundant in Zalgo but not in trueZ  }
#' \item{ratio_true1}{ ratio of links in trueZ that exist also in Zalgo  }
#' \item{ratio_true0}{ ratio of links not in trueZ that don't exist in  Zalgo.}
# ' \item{ratio_false1}{ ratio of links that exist only in Zalgo  }
# ' \item{ratio_false0}{ ratio of links that exist only in trueZ  }
#'
#'
#' @export
compare_struct <- function(trueZ = trueZ, Zalgo = Zalgo, all = TRUE, mode = "NULL") {
  if (mode == "hybrid") {
    trueZ = trueZ + t(trueZ) + t(trueZ) %*% trueZ # attention ? bien multiplier par la transpos?e ? gauche
    trueZ[trueZ > 1] = 1
    Zalgo = Zalgo + t(Zalgo) + t(Zalgo) %*% Zalgo # attention ? bien multiplier par la transpos?e ? gauche
    Zalgo[Zalgo > 1] = 1
  } else if (mode == "clique") { # on fait des cliques
    trueZ = cliquefaction(trueZ)
    Zalgo = cliquefaction(Zalgo)
  } else if (mode == "sym") { # on fait juste la d?sorientation
    trueZ = trueZ + t(trueZ)
    Zalgo = Zalgo + t(Zalgo)
  } else {
    # on garde les structures d'origine
  }
  trueZ = as.matrix(trueZ)
  Zalgo = as.matrix(Zalgo)
  res = as.matrix(trueZ - Zalgo)
  nbbon1 = sum(Zalgo * trueZ) # produit de hadamard
  nbtrop = -sum(res[res == -1]) # faux 1
  nbmank = sum(res[res == 1]) # faux 0
  nbbon0 = ncol(trueZ)^2 - nbbon1 - nbtrop - nbmank # vrai 0
  # dist=nbtrop+nbmank
  # on passe aux pourcentages
  taux_bon1 = nbbon1 / sum(trueZ) # taux de v?rit? d?couverte
  taux_bon0 = nbbon0 / (ncol(trueZ)^2 - sum(trueZ))
  taux_faux1 = nbtrop / sum(Zalgo) # taux d'ajout dans ce qui est dit doit tendre vers 0 (max si on ne dit que des aneries)
  taux_faux0 = 1 - taux_bon0 # taux d'oublis doit tendre vers 0 (max si on n a rien dit de vrai)
  vraissreg = which(colSums(trueZ) > 0)
  ssregalgo = which(colSums(Zalgo) > 0)
  deltap2 = length(vraissreg) - length(ssregalgo)
  bon_gauche = sum(duplicated(c(vraissreg, ssregalgo)))
  faux_gauche = length(ssregalgo) - bon_gauche
  if (all) {
    # return(list(ratio_true1=taux_bon1,ratio_true0=taux_bon0,ratio_false1=taux_faux1,ratio_false0=taux_faux0,true1=nbbon1,false1=nbtrop,false0=nbmank,deltadr=deltap2,true_left=bon_gauche,false_left=faux_gauche))
    return(list(ratio_true1 = taux_bon1, ratio_true0 = taux_bon0, true1 = nbbon1, false1 = nbtrop, false0 = nbmank, deltadr = deltap2, true_left = bon_gauche, false_left = faux_gauche))
  } else {
    return(list(ratio_true1 = taux_bon1, ratio_false1 = taux_faux1))
  }
}