R/Dist2All.R

In BIDistances: Bioinformatic Distances

# [dist2All, KNN] <- Dist2All(X, Data, SelectFeatures, method = "euclidean", p=2, knn=1, GPU=FALSE)
#
# Description:
# Returns the distances between a vector X and all rows of matrix Data.
#
# INPUT
# X		            Vector [1:n]; numeric;
# Data		        Matrix [1:n, 1:d]; numeric;
# SelectFeatures  [1:d] element equals TRUE if feature should be considered, otherwise zero
#
# OPTIONAL
# method          String for method, passed on to parallelDist::parallelDist(). Default is "euclidean".
# p               scalar, The ppth root of the sum of the ppth powers of the differences of the components.
# knn             scalar, which are the k nearest neighbors in Data
#
# OUTPUT
#List V with
# dist2All        [1:n]; numeric;  Distance of X to every row of Data
# KNN             [1:knn]; knn indices of the k nearest neighbors in Data
#
#ToDo: benchmark gegen mlpack::knn
#
#Author MCT, 2023

Dist2All = function(X, Data, SelectFeatures, method = "euclidean",p=2,knn=1) {

  if(!is.matrix(Data)){
    Data=as.matrix(Data)
    warning('dist2All: Data has to be matrix. applying as.matrix')
  }
  if(missing(SelectFeatures)) {
    SelectFeatures = rep(TRUE, ncol(Data))
  }
  if(mode(Data)!="numeric"){
    mode(Data)=="numeric"
    warning('dist2All: Data has to be numeric matrix. applying numerical mode.')
    }
  if(length(X)!=ncol(Data)){stop('dist2All: X has to be of same length as cols of A, i.e. number of colums == length X. Function stops!')}
  if(mode(X)!=mode(Data)){
    mode(X)=="numeric"
    warning('dist2All: X and Data have to be numeric vector and matrix respectively. Function stops!')

    }
  xData <- rbind(X, Data)
  xData <- xData[, SelectFeatures]
  if(sum(is.finite(xData[1,]))!=length(xData[1,])){
    stop('dist2All: X has values not are not finite. Function stops!')
  }
  if(!is.matrix(xData)) {
    xData = as.matrix(xData)
  }
  if(method=="euclidean"){
      distToAll <- as.vector(parallelDist::parallelDist(xData, method = method,p=p))[1:nrow(Data)]
  }else{
  if( isTRUE(requireNamespace('parallelDist'))){
    distToAll <- as.vector(parallelDist::parallelDist(xData, method = method,p=p))[1:nrow(Data)]

  }else{
    warning("dist2All: parallelDist not installed, faling back to minkowski distances.")
    dist2x    = apply(Data[, which( SelectFeatures == 1)], 1, function(row,X,p){sum((row-X)^p)}, X, p) # Summe der quadrierte Differenzen  zu  X
    distToAll = (dist2x)^(1/p)         # p=2, Euclid-sche Distanzen der Daten    zu X
  }
  }
  return(list("distToAll" = distToAll,
              "KNN"       = order(distToAll,decreasing = F)[1:knn]))
}