m_KnearestNeighbors.R

   1 #' @include b_Algorithm.R
   2
   3 #' @title K Nearest Neighbors Algorithm
   4 #'
   5 #' @description K Nearest Neighbors Algorithm.
   6 #' Inherits \code{\link{Algorithm}}
   7 #'
   8 #' @field k Number of neighbors to consider. Default: \code{n^(2/3)}
   9 #'
  10 KnearestNeighbors = setRefClass(
  11     Class = "KnearestNeighbors",
  12
  13     fields = c(
  14         k = "numeric"
  15     ),
  16
  17     contains = "Algorithm",
  18
  19     methods = list(
  20         predictOne = function(X, Y, x)
  21         {
  22             "Find the neighbors of one row, and solve a constrained linear system to obtain weights"
  23
  24             distances = sqrt(apply(X, 1, function(z)(return (sum((z-x)^2)))))
  25             rankedHistory = sort(distances, index.return=TRUE)
  26             n = length(Y)
  27             k_ = ifelse(length(k) == 0 || k <= 0. || k > n, getKnn(n), as.integer(k))
  28             weight = ridgeSolve(matricize(X[rankedHistory$ix[1:k_],]), Y[rankedHistory$ix[1:k_]], LAMBDA)
  29
  30             return (sum(x * weight))
  31         },
  32         predict_noNA = function(XY, x)
  33         {
  34             X = XY[,names(XY) != "Measure"]
  35             K = ncol(XY) - 1
  36             if (K == 1)
  37                 X = as.matrix(X)
  38             else if (length(XY[["Measure"]]) == 1)
  39                 X = t(as.matrix(X))
  40             Y = XY[,"Measure"]
  41             x = matricize(x)
  42             res = c()
  43             for (i in 1:nrow(x))
  44                 res = c(res, predictOne(X, Y, x[i,]))
  45             return (res)
  46         }
  47     )
  48 )