version 1.0.1

DESCRIPTION

@@ -0,0 +1,19 @@
+Package: evclass
+Type: Package
+Title: Evidential Distance-Based Classification
+Version: 1.0.1
+Date: 2016-06-21
+Author: Thierry Denoeux
+Maintainer: Thierry Denoeux <tdenoeux@utc.fr>
+Description: Different evidential distance-based classifiers, which provide
+    outputs in the form of Dempster-Shafer mass functions. The methods are: the
+    evidential K-nearest neighbor rule and the evidential neural network.
+License: GPL-3
+Depends: R (>= 3.1.0)
+Imports: FNN
+LazyData: TRUE
+RoxygenNote: 5.0.1
+NeedsCompilation: no
+Packaged: 2016-06-21 09:49:58 UTC; Thierry
+Repository: CRAN
+Date/Publication: 2016-06-21 17:42:52

MD5

@@ -0,0 +1,32 @@
+796e0a2c90a6bd52ddb6912593f0598f *DESCRIPTION
+6dfd314ed4c8aaa318cb72b37c1bdd9d *NAMESPACE
+bcd052b70a02a0e61c6457b927c334c8 *R/EkNNfit.R
+a0a4669c272eff4f63a0bdcacafe84e7 *R/EkNNinit.R
+cc918301931ef0dfa197a2f3d791b3d0 *R/EkNNval.R
+52a431a233d5075e603d3b52f9c9782c *R/classds.R
+115d774e5bce42f17e4c44ba20b1c5e2 *R/evclass.R
+6f31c165a0d7a2a3e86c41b4e697ed38 *R/foncgradr2n.R
+aeaad4cb6952aa9e16d979f5b01f0950 *R/glass-data.R
+2980f6f9293ffb4e0dc4d828494c1c99 *R/gradientds.R
+20a8bb25740f8f2445eda7ea0460ec20 *R/harris.R
+7c919b97eb312a99911bba45aa026664 *R/ionosphere-data.R
+848b2e2c11ce455a332dd8b0e7306f75 *R/optimds.R
+7535b780d2fd7f9fc526016f40975819 *R/proDSfit.R
+4523d06f8b0b94024d8d57b9819c9aac *R/proDSinit.R
+cf689fe5918edf79efcfb88395fcbd20 *R/proDSval.R
+a988fc277f3bfdb7f16cd8807d7bec4e *R/vehicles-data.R
+988e27d0f83a7ff327e4f38f089aeefd *data/glass.RData
+b7fa49999b8679c81898cc5262ee0c81 *data/ionosphere.RData
+1665c19c5134e390d23a6f938557c3b5 *data/vehicles.RData
+692b0234dd3ae533733ef4382a3109b6 *man/EkNNfit.Rd
+df9e6f51bbe7a511a71c62669b0369e9 *man/EkNNinit.Rd
+5c6c7e59a33b0c199ea37c15e93b6f1a *man/EkNNval.Rd
+cf56f366ae4cec7a5916b3be9c7313c5 *man/evclass.Rd
+ecf423e10f75e6dab5a7ced8a253575b *man/glass.Rd
+144a8069c347be152abc7543780db1c2 *man/ionosphere.Rd
+4f20dd9c17a3c66b6eda7e1c24012814 *man/proDSfit.Rd
+19dc1316032dc586a3b83fd4d9b31945 *man/proDSinit.Rd
+68c05de616fa459bba567ce123daebed *man/proDSval.Rd
+6d00db39a18df54421981daa1b761cdc *man/vehicles.Rd
+15c64d602e7264b11a8ca115a92a62f8 *vignettes/Introduction.Rmd
+7a59c365f0124c990e37d0f2e8c9768c *vignettes/tdenoeux.bib

NAMESPACE

@@ -0,0 +1,12 @@
+# Generated by roxygen2: do not edit by hand
+
+export(EkNNfit)
+export(EkNNinit)
+export(EkNNval)
+export(proDSfit)
+export(proDSinit)
+export(proDSval)
+import(FNN)
+importFrom(stats,dist)
+importFrom(stats,kmeans)
+importFrom(stats,runif)

R/EkNNfit.R

@@ -0,0 +1,66 @@
+#' Training of the EkNN classifier
+#'
+#'\code{EkNNfit} optimizes the parameters of the EkNN classifier.
+#'
+#'If the argument \code{param} is not supplied, the function \code{\link{EkNNinit}} is called.
+#'
+#' @param x Input matrix of size n x d, where n is the number of objects and d the number of
+#' attributes.
+#' @param y Vector of class labels (of length n). May be a factor, or a vector of
+#' integers.
+#' @param K Number of neighbors.
+#' @param param Initial parameters (default: NULL).
+#' @param alpha Parameter \eqn{\alpha} (default: 0.95)
+#' @param lambda Parameter of the cost function. If \code{lambda=1}, the
+#' cost function measures the error between the plausibilities and the 0-1 target values.
+#' If \code{lambda=1/M}, where M is the number of classes (default), the piginistic probabilities
+#' are considered in the cost function. If \code{lambda=0}, the beliefs are used.
+#' @param optimize Boolean. If TRUE (default), the parameters are optimized.
+#' @param options A list of parameters for the optimization algorithm: maxiter
+#' (maximum number of iterations), eta (initial step of gradient variation),
+#' gain_min (minimum gain in the optimisation loop), disp (Boolean; if TRUE, intermediate
+#' results are displayed during the optimization).
+#'
+#' @return A list with five elements:
+#'   \describe{
+#'   \item{param}{The optimized parameters.}
+#'   \item{cost}{Final value of the cost function.}
+#'   \item{err}{Leave-one-out error rate.}
+#'   \item{ypred}{Leave-one-out predicted class labels.}
+#'   \item{m}{Leave-one-out predicted mass functions. The first M columns correspond
+#'   to the mass assigned to each class. The last column corresponds to the mass
+#'   assigned to the whole set of classes.}
+#'  }
+#'
+#'@references T. Denoeux. A k-nearest neighbor classification rule based on Dempster-Shafer
+#'theory. IEEE Transactions on Systems, Man and Cybernetics, 25(05):804--813, 1995.
+#'
+#' L. M. Zouhal and T. Denoeux. An evidence-theoretic k-NN rule with parameter
+#' optimization. IEEE Transactions on Systems, Man and Cybernetics Part C,
+#' 28(2):263--271,1998.
+#'
+#'Available from \url{https://www.hds.utc.fr/~tdenoeux}.
+#'
+#'@author Thierry Denoeux.
+#'
+#' @export
+#' @import FNN
+#'
+#' @seealso \code{\link{EkNNinit}}, \code{\link{EkNNval}}
+#'
+#' @examples ## Iris dataset
+#' data(iris)
+#' x<-iris[,1:4]
+#' y<-iris[,5]
+#' fit<-EkNNfit(x,y,K=5)
+EkNNfit<-function(x,y,K,param=NULL,alpha=0.95,lambda=1/max(as.numeric(y)),optimize=TRUE,
+                  options=list(maxiter=300,eta=0.1,gain_min=1e-6,disp=TRUE)){
+  y<-as.integer(y)
+  x<-as.matrix(x)
+  if(is.null(param)) param<-EkNNinit(x,y,alpha)
+  knn<-get.knn(x,k=K)
+  knn$nn.dist<-knn$nn.dist^2
+  if(optimize) opt<-optimds(x,y,param,knn,K,lambda,options)
+  class <- classds(opt$param,knn,y,K)
+  return(list(param=opt$param,cost=opt$cost,err=class$err,ypred=class$ypred,m=class$m))
+}

R/EkNNinit.R

@@ -0,0 +1,54 @@
+#' Initialization of parameters for the EkNN classifier
+#'
+#'\code{EkNNinit} returns initial parameter values for the EkNN classifier.
+#'
+#'Each parameter \eqn{\gamma_k} is set ot the inverse of the square root of the mean
+#'Euclidean distances wihin class k. Note that \eqn{\gamma_k} here is the square root
+#'of the \eqn{\gamma_k} as defined in (Zouhal and Denoeux, 1998). By default, parameter alpha is set
+#'to 0.95. This value normally does not have to be changed.
+#'
+#'
+#' @param x Input matrix of size n x d, where n is the number of objects and d the number of
+#' attributes.
+#' @param y Vector of class lables (of length n). May be a factor, or a vector of
+#' integers.
+#' @param alpha Parameter \eqn{\alpha}.
+#'
+#' @return A list with two elements:
+#'   \describe{
+#'   \item{gamma}{Vector of parameters \eqn{\gamma_k}, of length c, the number of classes.}
+#'   \item{alpha}{Parameter \eqn{\alpha}, set to 0.95.}
+#'  }
+#'
+#'@references T. Denoeux. A k-nearest neighbor classification rule based on Dempster-Shafer
+#'theory. IEEE Transactions on Systems, Man and Cybernetics, 25(05):804--813, 1995.
+#'
+#' L. M. Zouhal and T. Denoeux. An evidence-theoretic k-NN rule with parameter
+#' optimization. IEEE Transactions on Systems, Man and Cybernetics Part C,
+#' 28(2):263--271,1998.
+#'
+#'Available from \url{https://www.hds.utc.fr/~tdenoeux}.
+#'
+#'@author Thierry Denoeux.
+#'
+#' @export
+#' @importFrom stats dist
+#'
+#' @seealso \code{\link{EkNNfit}}, \code{\link{EkNNval}}
+#'
+#' @examples ## Iris dataset
+#' data(iris)
+#' x<-iris[,1:4]
+#' y<-iris[,5]
+#' param<-EkNNinit(x,y)
+#' param
+EkNNinit<-function(x,y,alpha=0.95){
+  y<-as.numeric(y)
+  M<-max(y)
+  gamm<-rep(0,M)
+  for(k in 1:M){
+    D<-dist(x[y==k,])
+    gamm[k]<-1/sqrt(mean(D))
+  }
+  return(list(gamma=gamm,alpha=alpha))
+}

R/EkNNval.R

@@ -0,0 +1,94 @@
+#' Classification of a test set by the EkNN classifier
+#'
+#'\code{EkNNval} classifies instances in a test set using the EkNN classifier.
+#'
+#' If class labels for the test set are provided, the test error rate is also returned.
+#' If parameters are not supplied, they are given default values by \code{\link{EkNNinit}}.
+#'
+#'
+#' @param xtrain Matrix of size ntrain x d, containing the values of the d attributes for the
+#' training data.
+#' @param ytrain Vector of class labels for the training data (of length ntrain). May
+#' be a factor, or a vector of integers.
+#' @param xtst Matrix of size ntst x d, containing the values of the d attributes for the
+#' test data.
+#' @param K Number of neighbors.
+#' @param ytst Vector of class labels for the test data (optional). May
+#' be a factor, or a vector of integers.
+#' @param param Parameters, as returned by \code{\link{EkNNfit}}.
+#'
+#' @return A list with three elements:
+#'   \describe{
+#'   \item{m}{Predicted mass functions for the test data. The first M columns correspond
+#'   to the mass assigned to each class. The last column corresponds to the mass
+#'   assigned to the whole set of classes.}
+#'   \item{ypred}{Predicted class labels for the test data.}
+#'   \item{err}{Test error rate.}
+#'  }
+#'
+#'
+#'@references T. Denoeux. A k-nearest neighbor classification rule based on Dempster-Shafer
+#'theory. IEEE Transactions on Systems, Man and Cybernetics, 25(05):804--813, 1995.
+#'
+#' L. M. Zouhal and T. Denoeux. An evidence-theoretic k-NN rule with parameter
+#' optimization. IEEE Transactions on Systems, Man and Cybernetics Part C,
+#' 28(2):263--271,1998.
+#'
+#'Available from \url{https://www.hds.utc.fr/~tdenoeux}.
+#'
+#'@author Thierry Denoeux.
+#'
+#' @export
+#' @import FNN
+#'
+#' @seealso \code{\link{EkNNinit}}, \code{\link{EkNNfit}}
+#'
+#' @examples ## Iris dataset
+#' data(iris)
+#' train<-sample(150,100)
+#' xtrain<-iris[train,1:4]
+#' ytrain<-iris[train,5]
+#' xtst<-iris[-train,1:4]
+#' ytst<-iris[-train,5]
+#' K<-5
+#' fit<-EkNNfit(xtrain,ytrain,K)
+#' test<-EkNNval(xtrain,ytrain,xtst,K,ytst,fit$param)
+EkNNval <- function(xtrain,ytrain,xtst,K,ytst=NULL,param=NULL){
+
+  ytrain<-as.numeric(ytrain)
+  if(!is.null(ytst)) ytst<-as.numeric(ytst)
+
+  if(is.null(param)) param<-EkNNinit(xtrain,ytrain)
+
+  Napp<-nrow(xtrain)
+  M<-max(ytrain)
+  N<-nrow(xtst)
+
+  knn<-get.knnx(xtrain, xtst, k=K)
+  knn$nn.dist<-knn$nn.dist^2
+  is<-t(knn$nn.index)
+  ds<-t(knn$nn.dist)
+
+  m = rbind(matrix(0,M,N),rep(1,N))
+
+
+  for(i in 1:N){
+    for(j in 1:K){
+      m1 <- rep(0,M+1)
+      m1[ytrain[is[j,i]]] <- param$alpha*exp(-param$gamma[ytrain[is[j,i]]]^2*ds[j,i])
+      m1[M+1] <- 1 - m1[ytrain[is[j,i]]]
+      m[1:M,i] <- m1[1:M]*m[1:M,i] + m1[1:M]*m[M+1,i] + m[1:M,i]*m1[M+1]
+      m[M+1,i] <- m1[M+1] * m[M+1,i]
+      m<-m/matrix(colSums(m),M+1,N,byrow=TRUE)
+    }
+  }
+  m<-t(m)
+  ypred<-max.col(m[,1:M])
+  if(!is.null(ytst)) err<-length(which(ypred != ytst))/N else err<-NULL
+
+  return(list(m=m,ypred=ypred,err=err))
+
+}
+
+
+

R/classds.R

@@ -0,0 +1,35 @@
+classds<-function(param,knn,y,K){
+  N<-length(y)
+  M=max(y)
+  L<-rep(0,N)
+  mk <- rbind(matrix(0,M,N),rep(1,N))
+  is<-t(knn$nn.index)
+  ds<-t(knn$nn.dist)
+
+  for(k in 1:K){
+    Is <- is[k,]
+    Is = y[Is]
+    Tk <- matrix(0,M,N)
+    for(j in 1:M){
+      pos <- which(Is==j)
+      if(length(pos) != 0) Tk[j,pos] <- rep(1,length(pos))
+    }
+    G <- matrix(param$gamm^2,M,N) * Tk
+    gam <- apply(G,2,max)
+    s <- param$alpha*exp(-gam *ds[k,])
+    m <- rbind(Tk*matrix(s,M,N,byrow=TRUE),1-s)
+    mk <- rbind( mk[1:M,]*(m[1:M,]+matrix(m[M+1,],M,N,byrow=TRUE))+
+                   m[1:M,]*matrix(mk[M+1,],M,N,byrow=TRUE),mk[M+1,]*m[M+1,])
+    Kn <- colSums(mk)
+    mk <- mk/ matrix(Kn,M+1,N,byrow=TRUE)
+  }
+  mk<-t(mk)
+  L<-max.col(mk[,1:M])
+  err=length(which(L != y))/N
+  return(list(m=mk,ypred=L,err=err))
+}
+
+
+
+
+

R/evclass.R

@@ -0,0 +1,32 @@
+#' evclass: A package for evidential classification
+#'
+#' The evclass package currently contains functions for two evidential classifiers: the evidential
+#' K-nearest neighbor (EK-NN) rule (Denoeux, 1995; Zouhal and Denoeux, 1998) and the evidential
+#' neural network (Denoeux, 2000). In contrast with classical statistical classifiers, evidential
+#' classifier quantify the uncertainty of the classification using Dempster-Shafer mass functions.
+#'
+#' The main functions are: \code{\link{EkNNinit}}, \code{\link{EkNNfit}} and \code{\link{EkNNval}}
+#' for the initialization, training and evaluation of the EK-NN classifier, and
+#' \code{\link{proDSinit}}, \code{\link{proDSfit}} and \code{\link{proDSval}} for the
+#' evidential neural network classifier.
+#'
+#' @docType package
+#' @name evclass
+#'
+#' @seealso \code{\link{EkNNinit}}, \code{\link{EkNNfit}},
+#'\code{\link{EkNNval}}, \code{\link{proDSinit}}, \code{\link{proDSfit}}, \code{\link{proDSval}}.
+#'
+#' @references
+#'T. Denoeux. A k-nearest neighbor classification rule based on Dempster-Shafer
+#'theory. IEEE Transactions on Systems, Man and Cybernetics, 25(05):804--813, 1995.
+#'
+#'T. Denoeux. A neural network classifier based on Dempster-Shafer theory.
+#'IEEE Trans. on Systems, Man and Cybernetics A, 30(2):131--150, 2000.
+#'
+#' L. M. Zouhal and T. Denoeux. An evidence-theoretic k-NN rule with parameter
+#' optimization. IEEE Transactions on Systems, Man and Cybernetics Part C,
+#' 28(2):263--271,1998.
+#'
+#'Available from \url{https://www.hds.utc.fr/~tdenoeux}.
+#'
+NULL