######################################################################
### k-NN
######################################################################

makeKNN <- function(Xtrain,Ttrain) {
  standardizeF <- makeStandardizeF(Xtrain)
  Xtrain <- standardizeF(Xtrain)
  list(Xtrain=Xtrain, Ttrain=Ttrain, classes=unique(Ttrain), standardizeF=standardizeF)
}

useKNN <- function(knn,Xtest, nNeighbors=1) {
  Xtest <- knn$standardizeF(Xtest)
  indices <- apply(Xtest, 1,
                   function(xtest) {
                     distancesSquared <-
                       rowSums((knn$Xtrain -
                                matrix(xtest,nrow(knn$Xtrain),ncol(knn$Xtrain),byrow=TRUE))^2)
                     order(distancesSquared)
                   })
  indices <- t(indices)
  predictions <- NULL
  for (k in nNeighbors)
    predictions <- cbind(predictions,
                         mostCommon(matrix(knn$Ttrain[c(indices[,1:k])],nrow(indices),k), knn$classes))
  predictions
}

mostCommon <- function(data, values=unique(c(data))) {
  if (missing(data)) {
    cat("Usage: mostCommon(data, <values>)  where most common is determined for each row\n")
  }
  if (is.null(dim(data))) {
    counts <- lapply(values, function(v) sum(v==data))
    results <- values[which.max(counts)]
  } else {
    ## multiple rows in data
    results <- NULL
    for (rowi in 1:nrow(data)) {
      datarow <- data[rowi,]
      counts <- lapply(values, function(v) sum(v==datarow))
      results <- c(results, values[which.max(counts)])
    }
  }
  matrix(results)
}