R/mlbench-class.R

In mlbench: Machine Learning Benchmark Problems

#
#  Copyright (C) 1997-2010 Friedrich Leisch
#  $Id: mlbench-class.R 4612 2010-10-08 09:51:20Z leisch $
#

mlbench.xor <- function(n, d=2){

  x <- matrix(runif(n*d,-1,1),ncol=d,nrow=n)
  if((d != as.integer(d)) || (d<2))
    stop("d must be an integer >=2")

  z <- rep(0, length=n)
  for(k in 1:n){
    if(x[k,1]>=0){
      tmp <- (x[k,2:d] >=0)
      z[k] <- 1+sum(tmp*2^(0:(d-2)))
    }
    else {
      tmp <- !(x[k,2:d] >=0)
      z[k] <- 1 + sum(tmp*2^(0:(d-2)))
    }
  }

  retval <- list(x=x, classes=factor(z))
  class(retval) <- c("mlbench.xor", "mlbench")
  retval
}

mlbench.circle <- function(n, d=2){

  x <- matrix(runif(n*d,-1,1),ncol=d,nrow=n)
  if((d != as.integer(d)) || (d<2))
    stop("d must be an integer >=2")

  z <- rep(1, length=n)

  r <- (2^(d-1) * gamma(1+d/2) / (pi^(d/2)))^(1/d)
  z[apply(x, 1, function(x) sum(x^2)) > r^2] <- 2

  retval <- list(x=x, classes=factor(z))
  class(retval) <- c("mlbench.circle", "mlbench")
  retval
}

mlbench.2dnormals <- function(n, cl=2, r=sqrt(cl), sd=1){
  
  e <- sample(0:(cl-1), size=n, replace=TRUE)
  m <- r * cbind(cos(pi/4 + e*2*pi/cl), sin(pi/4 + e*2*pi/cl))
  x <- matrix(rnorm(2*n, sd=sd), ncol=2) + m

  retval <- list(x=x, classes=factor(e+1))
  class(retval) <- c("mlbench.2dnormals", "mlbench")
  retval
}


mlbench.1spiral <- function(n, cycles=1, sd=0)
{
    w <- seq(0, by=cycles/n, length=n)
    x <- matrix(0, nrow=n, ncol=2)

    x[,1] <- (2*w+1)*cos(2*pi*w)/3;
    x[,2] <- (2*w+1)*sin(2*pi*w)/3;

    if(sd>0){
        e <- rnorm(n, sd=sd)

        xs <- cos(2*pi*w)-pi*(2*w+1)*sin(2*pi*w)
        ys <- sin(2*pi*w)+pi*(2*w+1)*cos(2*pi*w)
  
        nrm <- sqrt(xs^2+ys^2)
        x[,1] <- x[,1] + e*ys/nrm
        x[,2] <- x[,2] - e*xs/nrm
    }
    x
}

mlbench.spirals <- function(n, cycles=1, sd=0)
{
    x <-  matrix(0, nrow=n, ncol=2)
    c2 <- sample(1:n, size=n/2, replace=FALSE)
    cl <- factor(rep(1, length=n), levels=as.character(1:2))
    cl[c2] <- 2

    x[-c2,] <- mlbench.1spiral(n=n-length(c2), cycles=cycles, sd=sd)
    x[c2,] <- - mlbench.1spiral(n=length(c2), cycles=cycles, sd=sd)
        
    retval <- list(x=x, classes=cl)
    class(retval) <- c("mlbench.spirals", "mlbench")
    retval
}

mlbench.ringnorm <- function(n, d=20)
{
    x <-  matrix(0, nrow=n, ncol=d)
    c2 <- sample(1:n, size=n/2, replace=FALSE)
    cl <- factor(rep(1, length=n), levels=as.character(1:2))
    cl[c2] <- 2

    a <- 1/sqrt(d)
    x[-c2,] <- matrix(rnorm(n=d*(n-length(c2)), sd=2), ncol=d)
    x[c2,]  <- matrix(rnorm(n=d*length(c2), mean=a), ncol=d)
    
    retval <- list(x=x, classes=cl)
    class(retval) <- c("mlbench.ringnorm", "mlbench")
    retval
}

mlbench.twonorm <- function (n, d = 20) 
{
    x <- matrix(0, nrow = n, ncol = d)
    c2 <- sample(1:n, size = n/2, replace = FALSE)
    cl <- factor(rep(1, length = n), levels = as.character(1:2))
    cl[c2] <- 2
    a <- 2/sqrt(d)
    x[-c2, ] <- matrix(rnorm(n = d * (n - length(c2)), mean = a, sd = 1), 
        ncol = d)
    x[c2, ] <- matrix(rnorm(n = d * length(c2), mean = -a), ncol = d)
    retval <- list(x = x, classes = cl)
    class(retval) <- c("mlbench.twonorm", "mlbench")
    retval
}

mlbench.threenorm <- function (n, d = 20) 
{
    x <- matrix(0, nrow = n, ncol = d)
    c2 <- sample(1:n, size = n/2, replace = FALSE)
    cl <- factor(rep(1, length = n), levels = as.character(1:2))
    cl[c2] <- 2
    c1 <- (1:n)[-c2]
    a <- 2/sqrt(d)
    for (i in c1)
      {
        distr <- as.logical(round(runif(1,0,1)))
        if ( distr )
          x[i, ] <- rnorm(n = d, mean = a) 
        else
          x[i, ] <- rnorm(n = d, mean = -a) 
      }
    m <- rep(c(a, -a), d/2)
    if ((d %% 2)==1)
      m <- c(m, a)
    x[c2, ] <- matrix(rnorm(n = d * length(c2), mean = m),
                      ncol = d, byrow=TRUE)
    retval <- list(x = x, classes = cl)
    class(retval) <- c("mlbench.threenorm", "mlbench")
    retval
}

mlbench.waveform <- function (n)
{
    Rnuminstances <- n
    retval <- .C("waveform",
                 Rnuminstances = as.integer(Rnuminstances),
                 x = double(21*n),
                 type = integer(n),
                 PACKAGE = "mlbench")
    x <- matrix (retval$x, ncol=21, byrow = TRUE)
    retval <- list (x=x, classes=as.factor(retval$type+1))
    class(retval) <- c("mlbench.waveform","mlbench")
    return(retval)
}

mlbench.cassini <- function(n,relsize=c(2,2,1))
{
    cassinib <- function(x, a, c)
    {
        y <- numeric(2)
        y[1] <- -sqrt(-c^2  - x^2  + sqrt(a^4  + 4*c^2*x^2))
        y[2] <- sqrt(-c^2  - x^2  + sqrt(a^4  + 4*c^2*x^2))
        y
    }
    
    circle <- function(x, r)
        sqrt(r^2-x^2)
    
    
    big1<-relsize[1]
    big2<-relsize[2]
    small<-relsize[3]
    parts<-big1+big2+small
    npiece<-n/parts
    n1<-round(big1*npiece)
    n2<-round(big2*npiece)
    n3<-round(small*npiece)
    if ((n1+n2+n3)!=n) n3<-n3+1
    a<-1
    C<-0.97
    Cell<-sqrt((1+C^2)/3)
    aell <- Cell*sqrt(2)
    transl <- 1.1
    r <- 0.6
    tmima1<-matrix(0,ncol=2,nrow=n1)
    tmima2<-matrix(0,ncol=2,nrow=n2)
    tmima3<-matrix(0,ncol=2,nrow=n3)
    n1found <- 0
    while(n1found < n1)
    {
        x1 <- runif(1,min=-sqrt(a^2+C^2),max=sqrt(a^2+C^2))
        y1 <- runif(1,min=-transl-1,max=-transl+0.6)
        if ((y1 < cassinib(x1,a,C)[2]-transl) &&
            (y1 > cassinib(x1,aell,Cell)[1]-transl))
        {
            n1found <- n1found +1
            tmima1[n1found,]<-c(x1,y1)
        }
    }
    
    n2found <- 0
    while(n2found < n2)
    {
        x2 <- runif(1,min=-sqrt(a^2+C^2),max=sqrt(a^2+C^2))
        y2 <- runif(1,max= transl+1,min=transl-0.6)
        if ((y2 > cassinib(x2,a,C)[1]+transl) &&
            (y2 < cassinib(x2,aell,Cell)[2]+transl))
        {
            n2found <- n2found +1
            tmima2[n2found,]<-c(x2,y2)
        }
    }
    
    n3found <- 0
    while(n3found < n3)
    {
        x3<-runif(1,min=-r,max=r)
        y3<-runif(1,min=-r,max=r)
        if ((y3 > -circle(x3,r)) &&
            (y3 < circle(x3,r)))
        {
            n3found <- n3found +1
            tmima3[n3found,]<-c(x3,y3) 
        }
    }
    teliko <- rbind(tmima1,tmima2,tmima3)
    cl <- factor(c(rep(1,n1),rep(2,n2),rep(3,n3)))
    retval<-list(x=teliko,classes=cl)
    class(retval) <- c("mlbench.cassini","mlbench")
    retval
}

mlbench.cuboids <- function (n, relsize=c(2,2,2,1))
{
    big1 <- relsize[1]
    big2 <- relsize[2]
    big3 <- relsize[3]
    small <- relsize[4]
    parts<-big1+big2++big3+small
    npiece<-n/parts
    n1<-round(big1*npiece)
    n2<-round(big2*npiece)
    n3<-round(big3*npiece)
    n4<-round(small*npiece)
    if ((n1+n2+n3+n4)!=n) n4<-n4+1
    
    x1 <- cbind(runif(n1,min=0,max=1),runif(n1,min=0.75,max=1.0),runif(n1,min=0.75,max=1))
    
    x2 <- cbind(runif(n2,min=0.75,max=1.0),runif(n2,min=0,max=0.25),runif(n2,min=0,max=1))
    
    x3 <- cbind(runif(n3,min=0.0,max=0.25),runif(n3,min=0.0,max=1),runif(n3,min=0,max=0.25))
    
    x4 <- cbind(runif(n4,min=0.4,max=0.6),runif(n4,min=0.4,max=0.6),runif(n4,min=0.4,max=0.6))
    
    x<-rbind(x1,x2,x3,x4)
    retval <-list(x=x,classes=factor(c(rep(1,n1),rep(2,n2),
                      rep(3,n3),rep(4,n4))))
    class(retval) <- c("mlbench.cuboids","mlbench")
    return(retval)
}


mlbench.smiley <- function(n=500, sd1=.1, sd2=.05)
{
    n1 <- round(n/6)
    n2 <- round(n/4)
    n3 <- n - 2 * n1 - n2

    x1 <- cbind(rnorm(n1, -.8, sd1), rnorm(n1, 1, sd1))
    x2 <- cbind(rnorm(n1,  .8, sd1), rnorm(n1, 1, sd1))

    x3 <- cbind(runif(n2, -.2, .2), runif(n2, 0, .75))
    x3[,1] <- x3[,1]*(1-x3[,2])
    
    x4 <- runif(n3, -1, 1)
    x4 <- cbind(x4, x4^2 - 1 + rnorm(n3, 0, sd2))

    x <- 
    retval <- list(x = rbind(x1, x2, x3, x4),
                   classes=factor(c(rep(1,n1),rep(2,n1),rep(3,n2),rep(4,n3))))
    class(retval) <- c("mlbench.smiley", "mlbench")
    retval
}
    
                           
mlbench.shapes <- function(n=500)
{
    n1 <- round(n/4)
    n2 <- n-3*n1
    
    x1 <- cbind(rnorm(n1, -1, .2), rnorm(n1, 1.5, .2))
    x2 <- cbind(runif(n1, -1.5, -0.5), runif(n1, -2, 0))

    x3 <- cbind(runif(n1, -1, 1), runif(n1, 1, 2))
    x3[,1] <- x3[,1]*(2-x3[,2])+1
    
    x4 <- runif(n2, 0.5, 2)
    x4 <- cbind(x4, cos(4*x4)-x4+runif(n2,-.2,.2))

    retval <- list(x = rbind(x1, x2, x3, x4),
                   classes=factor(c(rep(1,n1),rep(2,n1),rep(3,n1),rep(4,n2))))
    class(retval) <- c("mlbench.shapes", "mlbench")
    retval
}

###**********************************************************

## Original ist bincombinations in e1071
hypercube <- function(d) {

  retval <- matrix(0, nrow=2^d, ncol=d)
  
  for(n in 1:d){
    retval[,n] <- rep(c(rep(0, (2^d/2^n)), rep(1, (2^d/2^n))),
                      length=2^d)
  }
  retval
}



mlbench.hypercube <- function(n=800, d=3, sides=rep(1,d), sd=0.1)
{
    m <- hypercube(d)
    n1 <- round(n/2^d)
    
    sides <- rep(sides, length=d)
    z <- NULL
    
    for(k in 1:nrow(m))
    {
        m[k,] <- m[k,]*sides
        z1 <- matrix(rnorm(d*n1, sd=sd), ncol=d)
        z1 <- sweep(z1, 2, m[k,], "+")
        z <- rbind(z, z1)
    }
    retval <- list(x=z,
                   classes=factor(rep(1:nrow(m), rep(n1, nrow(m)))))
    class(retval) <- c("mlbench.hypercube", "mlbench")
    retval
}

## for backwards compatibility
mlbench.corners <- function(...) mlbench.hypercube(...)

###**********************************************************

simplex <- function(d, sides, center = TRUE)
{
    m <- matrix(0, d+1, d)
    cent <- rep(0,d)

    m[2,1] <- sides
    cent[1] <- sides/2
    b <- sides/2

    if(d>=2)
    {
        for(i in 2:d)
        {
            m[i+1,] <- cent
            m[i+1,i] <- sqrt(sides^2-b^2)
            cent[i] <- 1/(i+1)* m[i+1,i]
            b <- (1- 1/(i+1)) * m[i+1,i]
        }
    }
    if(center)
      m <- t(t(m) - cent)
    m
}

mlbench.simplex <- function (n = 800, d = 3, sides = 1, sd = 0.1, center=TRUE)
{
    m <- simplex(d=d , sides=sides, center=center)
    n1 <- round(n/2^d)
    z <- NULL
    for (k in 1:nrow(m)) {
        z1 <- matrix(rnorm(d * n1, sd = sd), ncol = d)
        z1 <- sweep(z1, 2, m[k, ], "+")
        z <- rbind(z, z1)
    }
    retval <- list(x = z, classes = factor(rep(1:nrow(m), rep(n1,
        nrow(m)))))
    class(retval) <- c("mlbench.simplex", "mlbench")
    retval
}

                           
###**********************************************************
                           

bayesclass <- function(z) UseMethod("bayesclass")

bayesclass.noerr <- function(z) z$classes

bayesclass.mlbench.xor <- bayesclass.noerr
bayesclass.mlbench.circle <- bayesclass.noerr
bayesclass.mlbench.cassini <- bayesclass.noerr
bayesclass.mlbench.cuboids <- bayesclass.noerr

    
bayesclass.mlbench.2dnormals <- function(z){

    ncl <- length(levels(z$classes))
    z <- z$x
    for(k in 1:nrow(z)){
        z[k,] <- z[k,] / sqrt(sum(z[k,]^2))
    }        
    winkel <- acos(z[,1] * sign(z[,2])) + pi * (z[,2]<0)
    winkel <- winkel - pi/ncl - pi/4
    winkel[winkel < 0] <- winkel[winkel<0] + 2*pi
    retval <- (winkel)%/%(2 * pi/ncl)
    factor((retval+1)%%ncl+1)
}

bayesclass.mlbench.ringnorm <- function (z)
  {
    z <- z$x
    ndata <- dim(z)[1]
    ndim <- dim(z)[2]
    a <- 1/sqrt(ndim)
    center1 <- rep(0,ndim)
    center2 <- rep(a,ndim)
    m1 <- mahalanobis(z, center1, (4*diag(ndim)), inverted=FALSE) +
        ndim*log(4)
    m2 <- mahalanobis(z, center2, diag(ndim), inverted=FALSE)
    as.factor ((m1 > m2) +1)
  }

bayesclass.mlbench.twonorm <- function (z)
  {
    z <- z$x
    ndata <- dim(z)[1]
    bayesclass <- integer(ndata)
    ndim <- dim(z)[2]
    a <- 2/sqrt(ndim)
    center1 <- rep(a,ndim)
    center2 <- rep(-a,ndim)
    for (i in 1:ndata)
      {
        dist1 <- sum((z[i, ] - center1) ^2)
        dist2 <- sum((z[i, ] - center2) ^2)
        bayesclass[i] <- (dist1 > dist2) +1
      }
    as.factor(bayesclass)
  }

## Code by Julia Schiffner
bayesclass.mlbench.threenorm <- function(z)
{
    z <- z$x
    ndim <- dim(z)[2]
    a <- 2/sqrt(ndim)

    center1a <- rep(a, ndim)
    center1b <- rep(-a, ndim)
    center2 <- rep(c(a, -a), ndim/2)
    
    if ((ndim%%2) == 1)
        center2 <- c(center2, a)

    m1 <- 0.5 * exp(-0.5 * mahalanobis(z, center1a, diag(ndim),
                                       inverted = FALSE)) +
          0.5 * exp(-0.5 * mahalanobis(z, center1b,
                                       diag(ndim), inverted = FALSE))

    m2 <- exp(-0.5 * mahalanobis(z, center2, diag(ndim), inverted = FALSE))
    
    as.factor((m1 < m2) + 1)
}

###**********************************************************

as.data.frame.mlbench <- function(x, row.names=NULL, optional=FALSE, ...)
{
    data.frame(x=x$x, classes=x$classes)
}


plot.mlbench <- function(x, xlab="", ylab="", ...)
{
    if(ncol(x$x)>2){
        pairs(x$x, col=as.integer(x$classes), ...)
    }
    else{
        plot(x$x, col=as.integer(x$classes), xlab=xlab, ylab=ylab, ...)
    }
}