R/cec.R

In CEC: Cross-Entropy Clustering

#' @title Cross-Entropy Clustering
#' 
#' @aliases cec-class
#'
#' @description \code{cec} performs Cross-Entropy Clustering on a data matrix. 
#'  See \code{Details} for an explanation of Cross-Entropy Clustering.
#'
#' @param x A numeric matrix of data. Each row corresponds to a distinct
#'  observation; each column corresponds to a distinct variable/dimension. It
#'  must not contain \code{NA} values.
#'
#' @param centers Either a matrix of initial centers or the number of initial
#'  centers (\code{k}, single number \code{cec(data, 4, ...)}) or a vector for
#'  variable number of centers (\code{cec(data, 3:10, ...)}). It must not
#'  contain \code{NA} values.
#'
#'  If \code{centers} is a vector, \code{length(centers)} clusterings will be
#'  performed for each start (\code{nstart} argument) and the total number of
#'  clusterings will be \code{length(centers) * nstart}.
#'
#'  If \code{centers} is a number or a vector, initial centers will be generated
#'  using a method depending on the \code{centers.init} argument.
#'
#' @param type The type (or types) of clustering (density family). This can be
#'  either a single value or a vector of length equal to the number of centers.
#'  Possible values are: "covariance", "fixedr", "spherical", "diagonal",
#'  "eigenvalues", "all" (default).
#'
#'  Currently, if the \code{centers} argument is a vector, only a single type can
#'  be used.
#'
#' @param iter.max The maximum number of iterations of the clustering algorithm.
#'
#' @param nstart The number of clusterings to perform (with different initial
#'  centers). Only the best clustering (with the lowest cost) will be returned.
#'  A value grater than 1 is valid only if the \code{centers} argument is a
#'  number or a vector.
#'
#'  If the \code{centers} argument is a vector, \code{length(centers)}
#'  clusterings will be performed for each start and the total number of
#'  clusterings will be \code{length(centers) * nstart}.
#'
#'  If the split mode is on (\code{split = TRUE}), the whole procedure (initial
#'  clustering + split) will be performed \code{nstart} times, which may take
#'  some time.
#'
#' @param centers.init The method used to automatically initialize the centers.
#'  Possible values are: "kmeans++" (default) and "random".
#'
#' @param param The parameter (or parameters) specific to a particular type of
#'  clustering. Not all types of clustering require parameters. The types that
#'  require parameter are: "covariance" (matrix parameter), "fixedr" (numeric
#'  parameter), "eigenvalues" (vector parameter). This can be a vector or a list
#'  (when one of the parameters is a matrix or a vector).
#'
#' @param card.min The minimal cluster cardinality. If the number of
#'  observations in a cluster becomes lower than card.min, the cluster is
#'  removed. This argument can be either an integer number or a string ending
#'  with a percent sign (e.g. "5\%").
#'
#' @param keep.removed If this parameter is TRUE, the removed clusters will be
#'  visible in the results as NA in the "centers" matrix (as well as the
#'  corresponding values in the list of covariances).
#'
#' @param interactive If \code{TRUE}, the result of clustering will be plotted after
#'  every iteration.
#'
#' @param threads The number of threads to use or "auto" to use the default
#'  number of threads (usually the number of available processing units/cores)
#'  when performing multiple starts (\code{nstart} parameter).
#'
#'  The execution of a single start is always performed by a single thread, thus
#'  for \code{nstart = 1} only one thread will be used regardless of the value
#'  of this parameter.
#'
#' @param split If \code{TRUE}, the function will attempt to discover new
#'  clusters after the initial clustering, by trying to split single clusters
#'  into two and check whether it lowers the cost function.
#'
#'  For each start (\code{nstart}), the initial clustering will be performed and
#'  then splitting will be applied to the results. The number of starts in the
#'  initial clustering before splitting is driven by the
#'  \code{split.initial.starts} parameter.
#'
#' @param split.depth The cluster subdivision depth used in split mode. Usually,
#'  a value lower than 10 is sufficient (when after each splitting, new clusters
#'  have similar sizes). For some data, splitting may often produce clusters
#'  that will not be split further, in that case a higher value of
#'  \code{split.depth} is required.
#'
#' @param split.tries The number of attempts that are made when trying to split
#'  a cluster in split mode.
#'
#' @param split.limit The maximum number of centers to be discovered in split
#'  mode.
#'
#' @param split.initial.starts The number of 'standard' starts performed before
#'  starting the splitting process.
#'
#' @param readline Used only in the interactive mode. If \code{readline} is
#'  TRUE, at each iteration, before plotting it will wait for the user to press
#'  <Return> instead of the standard 'before plotting' waiting
#'  (\code{graphics::par(ask = TRUE)}).
#'
#' @details Cross-Entropy Clustering (CEC) aims to partition \emph{m} points
#'  into \emph{k} clusters so as to minimize the cost function (energy
#'  \emph{\strong{E}} of the clustering) by switching the points between
#'  clusters. The presented method is based on the Hartigan approach, where we
#'  remove clusters which cardinalities decreased below some small prefixed
#'  level.
#'
#'  The energy function \emph{\strong{E}} is given by:
#'
#'  \deqn{E(Y_1,\mathcal{F}_1;...;Y_k,\mathcal{F}_k) = \sum\limits_{i=1}^{k}
#'  p(Y_i) \cdot (-ln(p(Y_i)) + H^{\times}(Y_i\|\mathcal{F}_i))}{ E(Y1, F1; ...;
#'  Yk, Fk) = \sum(p(Yi) * (-ln(p(Yi)) + H(Yi | Fi)))}
#'
#'  where \emph{Yi} denotes the \emph{i}-th cluster, \emph{p(Yi)} is the ratio
#'  of the number of points in \emph{i}-th cluster to the total number points,
#'  \emph{\strong{H}(Yi|Fi)} is the value of cross-entropy, which represents the
#'  internal cluster energy function of data \emph{Yi} defined with respect to a
#'  certain Gaussian density family \emph{Fi}, which encodes the type of
#'  clustering we consider.
#'
#'  The value of the internal energy function \emph{\strong{H}} depends on the
#'  covariance matrix (computed using maximum-likelihood) and the mean (in case
#'  of the \emph{mean} model) of the points in the cluster. Seven
#'  implementations of \emph{\strong{H}} have been proposed (expressed as a type
#'  - model - of the clustering):
#'
#'  \itemize{
#'   \item{"all": }{All Gaussian densities. Data will form ellipsoids with
#'   arbitrary radiuses.}
#'   \item{"covariance": }{Gaussian densities with a fixed given covariance. The
#'   shapes of clusters depend on the given covariance matrix (additional
#'   parameter).}
#'   \item{"fixedr": }{Special case of 'covariance', where the covariance matrix
#'   equals \emph{rI} for the given \emph{r} (additional parameter). The
#'   clustering will have a tendency to divide data into balls with approximate
#'   radius proportional to the square root of \emph{r}.}
#'   \item{"spherical": }{Spherical (radial) Gaussian densities (covariance
#'   proportional to the identity). Clusters will have a tendency to form balls
#'   of arbitrary sizes.}
#'   \item{"diagonal": }{Gaussian densities with diagonal covariane. Data will
#'   form ellipsoids with radiuses parallel to the coordinate axes.}
#'   \item{"eigenvalues": }{Gaussian densities with covariance matrix having
#'   fixed eigenvalues (additional parameter). The clustering will try to divide
#'   the data into fixed-shaped ellipsoids rotated by an arbitrary angle.}
#'   \item{"mean": }{Gaussian densities with a fixed mean. Data will be covered
#'   with ellipsoids with fixed centers.}
#'  }
#'
#'  The implementation of \code{cec} function allows mixing of clustering types.
#'
#' @return An object of class \code{cec} with the following attributes:
#'  \code{data}, \code{cluster}, \code{probability}, \code{centers},
#'  \code{cost.function}, \code{nclusters}, \code{iterations}, \code{cost},
#'  \code{covariances}, \code{covariances.model}, \code{time}.
#'
#' @seealso \code{\link{CEC-package}}, \code{\link{plot.cec}}, 
#'  \code{\link{print.cec}}
#'
#' @references Spurek, P. and Tabor, J. (2014) Cross-Entropy Clustering
#'  \emph{Pattern Recognition} \bold{47, 9} 3046--3059
#'
#' @keywords cluster models multivariate package
#'
#' @examples
#' ## Example of clustering a random data set of 3 Gaussians, with 10 random
#' ## initial centers and a minimal cluster size of 7% of the total data set.
#'
#' m1 <- matrix(rnorm(2000, sd = 1), ncol = 2)
#' m2 <- matrix(rnorm(2000, mean = 3, sd = 1.5), ncol = 2)
#' m3 <- matrix(rnorm(2000, mean = 3, sd = 1), ncol = 2)
#' m3[,2] <- m3[, 2] - 5
#' m <- rbind(m1, m2, m3)
#'
#' plot(m, cex = 0.5, pch = 19)
#'
#' ## Clustering result:
#' Z <- cec(m, 10, iter.max = 100, card.min = "7%")
#' plot(Z)
#'
#' # Result:
#' Z
#'
#' ## Example of clustering mouse-like set using spherical Gaussian densities.
#' m <- mouseset(n = 7000, r.head = 2, r.left.ear = 1.1, r.right.ear = 1.1,
#' left.ear.dist = 2.5, right.ear.dist = 2.5, dim = 2)
#' plot(m, cex = 0.5, pch = 19)
#' ## Clustering result:
#' Z <- cec(m, 3, type = 'sp', iter.max = 100, nstart = 4, card.min = '5%')
#' plot(Z)
#' # Result:
#' Z
#'
#' ## Example of clustering data set 'Tset' using 'eigenvalues' clustering type.
#' data(Tset)
#' plot(Tset, cex = 0.5, pch = 19)
#' centers <- init.centers(Tset, 2)
#' ## Clustering result:
#' Z <- cec(Tset, 5, 'eigenvalues', param = c(0.02, 0.002), nstart = 4)
#' plot(Z)
#' # Result:
#' Z
#'
#' ## Example of using cec split method starting with a single cluster.
#' data(mixShapes)
#' plot(mixShapes, cex = 0.5, pch = 19)
#' ## Clustering result:
#' Z <- cec(mixShapes, 1, split = TRUE)
#' plot(Z)
#' # Result:
#' Z
#'
#' @export
cec <- function(x,
                centers,
                type = c("covariance", "fixedr", "spherical", "diagonal",
                         "eigenvalues", "mean", "all"),
                iter.max = 25,
                nstart = 1,
                param,
                centers.init = c("kmeans++", "random"),
                card.min = "5%",
                keep.removed = FALSE,
                interactive = FALSE,
                threads = 1,
                split = FALSE,
                split.depth = 8,
                split.tries = 5,
                split.limit = 100,
                split.initial.starts = 1,
                readline = TRUE) {

    ### CHECK ARGUMENTS
    if (!methods::hasArg(x))
        stop("Missing required argument: 'x'.")

    if (!methods::hasArg(centers)) {
        centers <- 1
        split <- TRUE
    }

    if (iter.max < 0)
        stop("Illegal argument: iter.max must be greater than 0.")

    if (!is.matrix(x))
        stop("Illegal argument: 'x' must be a matrix.")

    if (ncol(x) < 1)
        stop("Illegal argument: 'x' must have at least 1 column.")

    if (nrow(x) < 1)
        stop("Illegal argument: 'x' must have at least 1 row.")

    if (!all(stats::complete.cases(x)))
        stop("Illegal argument: 'x' should not contain NA values.")

    if (!all(stats::complete.cases(centers)))
        stop("Illegal argument: 'centers' should not contain NA values.")

    var.centers <- NULL
    centers.mat <- NULL

    if (!is.matrix(centers)) {
        if (length(centers) > 1) {
            var.centers <- centers
        } else {
            var.centers <- c(centers)
        }

        for (i in centers) {
            if (i < 1) {
                stop("Illegal argument: 'centers' must only contain integers greater
             than 0")
            }
        }

        centers.initialized <- FALSE
    } else {
        if (ncol(x) != ncol(centers)) {
            stop("Illegal argument: 'x' and 'centers' must have the same number of columns.")
        }

        if (nrow(centers) < 1) {
            stop("Illegal argument: 'centers' must have at least 1 row.")
        }

        var.centers <- c(nrow(centers))
        centers.mat <- centers
        centers.initialized <- TRUE
    }

    if (!(attr(regexpr("[\\.0-9]+%{0,1}", perl = TRUE, text = card.min), "match.length") == nchar(card.min))) {
        stop("Illegal argument: 'card.min' in wrong format.")
    }

    if (centers.initialized) {
        init.method.name <- "none"
    } else if (methods::hasArg(centers.init)) {
        init.method.name <- switch(match.arg(centers.init), `kmeans++` = "kmeanspp", random = "random")
    } else {
        init.method.name <- "kmeanspp"
    }

    if (!methods::hasArg(type)) {
        type <- "all"
    }

    if (length(type) > 1 && length(var.centers) > length(type)) {
        stop("Illegal argument: 'type' with length > 1 should be equal or greater
         than the length of the vector of variable number of centers ('centers'
         as a vector).")
    }

    ### INTERACTIVE MODE
    if (interactive) {
        if (split == TRUE) {
            stop("The interactive mode is not available in split mode.")
        }

        if (length(var.centers) > 1) {
            stop("The interactive mode is not available for variable centers.")
        }

        if (nstart > 1) {
            stop("The interactive mode is not available for multiple starts.")
        }

        return(cec.interactive(x, centers, type, iter.max, 1, param, centers.init,
                               card.min, keep.removed, readline))
    }

    ### NON-INTERACTIVE MODE
    n <- ncol(x)
    m <- nrow(x)

    if (substr(card.min, nchar(card.min), nchar(card.min)) == "%") {
        card.min <- as.integer(as.double(substr(card.min, 1, nchar(card.min) - 1)) * m/100)
    } else {
        card.min <- as.integer(card.min)
    }

    card.min <- max(card.min, n + 1)
    k <- max(var.centers)
    startTime <- proc.time()

    centers.r <- list(init.method = init.method.name,
                      var.centers = as.integer(var.centers),
                      mat = centers.mat)

    if (threads == "auto") {
        threads <- 0
    }

    control.r <- list(min.card = as.integer(card.min),
                      max.iters = as.integer(iter.max),
                      starts = as.integer(nstart),
                      threads = as.integer(threads))

    models.r <- create.cec.params.for.models(k, n, type, param)

    if (split) {
        for (i in 1:k) {
            if (models.r[[i]]$type != models.r[[1]]$type) {
                stop("Mixing model types is currently not supported in split mode")
            }
        }

        split.r <- list(depth = as.integer(split.depth),
                        limit = as.integer(split.limit),
                        tries = as.integer(split.tries),
                        initial.starts = as.integer(split.initial.starts))
        Z <- .Call(cec_split_r, x, centers.r, control.r, models.r, split.r)
    } else {
        Z <- .Call(cec_r, x, centers.r, control.r, models.r)
    }

    k.final <- nrow(Z$centers)
    execution.time <- as.vector((proc.time() - startTime))[3]
    Z$centers[is.nan(Z$centers)] <- NA
    tab <- tabulate(Z$cluster)
    probability <- vapply(tab, function(c.card) {
        c.card/m
    }, 0)

    # TODO: change this temporary hack
    model.one <- models.r[[1]]

    if (!keep.removed) {
        cluster.map <- 1:k.final
        na.rows <- which(is.na(Z$centers[, 1]))

        if (length(na.rows) > 0) {
            for (i in 1:length(na.rows)) {
                for (j in na.rows[i]:k.final) {
                    cluster.map[j] <- cluster.map[j] - 1
                }
            }

            Z$cluster <- as.integer(vapply(Z$cluster, function(asgn) {
                as.integer(cluster.map[asgn])
            }, 0))

            Z$centers <- matrix(Z$centers[-na.rows, ], , n)
            Z$covariances <- Z$covariances[-na.rows]
            probability <- probability[-na.rows]
            models.r <- models.r[-na.rows]
        }
    }

    covs <- length(Z$covariances)
    covariances.model <- rep(list(NA), covs)
    means.model <- Z$centers

    # TODO: change this temporary hack
    if (split) {
        models.r <- rep(list(model.one), covs)
    }

    for (i in 1:covs) {
        covariances.model[[i]] <- model.covariance(models.r[[i]]$type, Z$covariances[[i]],
                                                   Z$centers[i, ], models.r[[i]]$params)
        means.model[i, ] <- model.mean(models.r[[i]]$type, Z$centers[i, ], models.r[[i]]$params)
    }

    structure(
        list(data = x, cluster = Z$cluster, centers = Z$centers, probability = probability, cost.function = Z$energy,
             nclusters = Z$nclusters, iterations = Z$iterations, time = execution.time, covariances = Z$covariances,
             covariances.model = covariances.model, means.model = means.model),
        class = "cec")
}


#' @title Interactive Cross-Entropy Clustering
#'
#' @description Internal function to run \code{\link{cec}} interactively.
#'  
#' @noRd
cec.interactive <- function(x, 
                            centers, 
                            type = c("covariance", "fixedr", "spherical",
                                     "diagonal", "eigenvalues", "all"),
                            iter.max = 40, 
                            nstart = 1, 
                            param, 
                            centers.init = c("kmeans++", "random"),
                            card.min = "5%", 
                            keep.removed = FALSE, 
                            readline = TRUE) {
    old.ask <- graphics::par()["ask"]
    n <- ncol(x)
    
    if (n != 2) {
        stop("interactive mode available only for 2-dimensional data")
    }
    
    i <- 0
    
    if (!is.matrix(centers)) {
        centers <- init.centers(x, centers, centers.init)
    }
    
    if (readline) {
        ignore <- readline(prompt = "After each iteration you may:\n - press <Enter> for next iteration \n - write number <n> (may be negative one) and press <Enter> for next <n> iterations \n - write 'q' and abort execution.\n Press <Return>.\n")
        graphics::par(ask = FALSE)
    } else {
        graphics::par(ask = TRUE)
    }
    
    while (TRUE) {
        Z <- cec(x, centers, type, i, 1, param, centers.init, card.min, keep.removed, FALSE)
        
        if (i > Z$iterations | i >= iter.max) {
            break
        }
        
        desc <- ""
        
        if (i == 0) { 
            desc <- "(position of center means before first iteration)"
        }
        
        cat("Iterations:", Z$iterations, desc, "cost function:", Z$cost, " \n ")
        
        plot(Z, ellipses = TRUE)
        
        if (readline) {
            line <- readline(prompt = "Press <Enter> OR write number OR write 'q':")
            lineint <- suppressWarnings(as.integer(line))
            
            if (!is.na(lineint)) {
                i <- i + lineint - 1
                
                if (i < 0) {
                    i = -1
                }
            } else if (line == "q" | line == "quit") {
                break
            }
        }
        
        i <- i + 1
    }
    
    plot(Z, ellipses = "TRUE")
    
    if (readline) {
        ignore <- readline(prompt = "Press <Enter>:")
    }
    
    graphics::par(old.ask)
    Z
}