R/coclusterBinary.R

In blockcluster: Co-Clustering Package for Binary, Categorical, Contingency and Continuous Data-Sets

#' @include coclusterStrategy.R
#' @include optionclasses.R
#' 
NULL


#' Co-Clustering function for Binary data.
#' 
#' This function performs Co-Clustering (simultaneous clustering of rows and
#' columns ) for Binary data-sets using latent block models.
#' It can also be used to perform semi-supervised co-clustering.  
#' 
#' @param data Input data as matrix (or list containing data matrix)
#' @param semisupervised Boolean value specifying whether to perform
#' semi-supervised co-clustering or not. Make sure to provide row and/or
#' column labels if specified value is true. The default value is false.
#' @param rowlabels Integer Vector specifying the class of rows.
#' The class number starts from zero. Provide -1 for unknown row class. 
#' @param collabels Integer Vector specifying the class of columns.
#' The class number starts from zero. Provide -1 for unknown column class.
#' @param model This is the name of model. The following models exists for
#' Binary data:
#' \tabular{rlll}{
#'     Model  \tab Data-type \tab Proportions \tab Dispersion/Variance \cr
#'     pik_rhol_epsilonkl(Default) \tab binary \tab unequal \tab unequal \cr
#'     pik_rhol_epsilon \tab binary \tab unequal \tab equal \cr
#'     pi_rho_epsilonkl \tab binary \tab equal \tab unequal \cr
#'     pi_rho_epsilon \tab binary \tab equal \tab equal \cr
#' }
#' @param a First hyper-parameter in case of Bayesian settings. Default is 1 (no prior).
#' @param b Second hyper-parameter in case of Bayesian settings. Default is 1 (no prior).
#' 
#' @param nbcocluster Integer vector specifying the number of row and column clusters respectively.
#' @param strategy Object of class \code{\linkS4class{strategy}}.
#' @param nbCore number of thread to use (OpenMP must be available), 0 for all cores. Default value is 1.

#' @return Return an object of \code{\linkS4class{BinaryOptions}}.
#' 
# @export
#' 
# @exportPattern "^[[:alpha:]]+"
# @useDynLib RCocluster
#' 
#' @examples
#' 
#' ## Simple example with simulated binary data
#' ## load data
#' data(binarydata)
#' ## usage of coclusterBinary function in its most simplest form
#' out<-coclusterBinary(binarydata,nbcocluster=c(2,3))
#' ## Summarize the output results
#' summary(out)
#' ## Plot the original and Co-clustered data 
#' plot(out)
#' 
coclusterBinary<-function( data, semisupervised = FALSE
                         , rowlabels = integer(0), collabels = integer(0)
				         , model = NULL
				         , nbcocluster
				         , strategy = coclusterStrategy()
				         , a=1, b=1, nbCore = 1) 
{
	#Check for data and get dimensions
	if(missing(data)){ stop("Data is missing.")}
	if(!is.list(data))
  {
		if(!is.matrix(data)) { stop("Data should be matrix.")}
    dimData = dim(data)
	}
  else
	{
		if(!is.matrix(data[[1]])) { stop("Data should be matrix.")}
		if(!is.numeric(data[[2]])||!is.numeric(data[[3]]))
    { stop("Row/Column effects should be numeric vectors.")}
		if(length(data[[2]])!=dim(data[[1]])[1]||length(data[[3]])!=dim(data[[1]])[2])
    { stop("Dimension mismatch in Row/column effects  and Data.")}
    dimData = dim(data[[1]])
	}

  #check for row and column labels
  if (semisupervised)
  {
    if(missing(rowlabels)&&missing(collabels))
    { stop("Missing row and column labels. At-least one should be provided to perform semi-supervised Co-clustering.")}
    if(!missing(rowlabels)&&!is.numeric(rowlabels))
    { stop("Row labels should be a numeric vector.")}
    if(!missing(collabels)&&!is.numeric(collabels))
    { stop("Column labels should be a numeric vector.")}
    if(missing(rowlabels))      rowlabels = rep(-1,dimData[1])
    else if(missing(collabels)) collabels = rep(-1,dimData[2])
    # check    
    if(dimData[1]!=length(rowlabels))
    {  stop("rowlabels length does not match number of rows in data (also ensure to put -1 in unknown labels)")}
    if(dimData[2]!=length(collabels))
    { stop("collabels length does not match number of columns in data (also  ensure to put -1 in unknown labels)")}
  }
  #check for number of coclusters
	if(missing(nbcocluster))     { stop("Mention number of CoClusters.")}
	if(dimData[1]<nbcocluster[1]){ stop("Number of Row clusters exceeds numbers of rows.")}
	if(dimData[2]<nbcocluster[2]){ stop("Number of Column clusters exceeds numbers of columns.")}
  if(nbcocluster[1]<1 || nbcocluster[2]<1) { stop("Number of cluster must be at least 1.")}
  
	#check for Algorithm name (and make it compatible with version 1)
	if(strategy@algo=="XEMStrategy")
  {
    warning("The algorithm 'XEMStrategy' is renamed as BEM!")
    strategy@algo == "BEM"
  }
  else if(strategy@algo == "XCEMStrategy")
  {
    warning("The algorithm 'XCEMStrategy' is renamed as BCEM!")
    strategy@algo = "BCEM"
  }
  else if(strategy@algo!="BEM" && strategy@algo!="BCEM" && strategy@algo!="BSEM" && strategy@algo!="BGibbs")
  {
    stop("Incorrect Algorithm, Valide algorithms are: BEM, BCEM, BSEM, BGibbs") 
  }
	#check for stopping criteria
	if(strategy@stopcriteria!="Parameter" && strategy@stopcriteria!="Likelihood")
		stop("Incorrect stopping criteria, Valid stopping criterians are: Parameter, Likelihood")
  # check for model
	if(is.null(model)){ model = "pik_rhol_epsilonkl"}
	else 
	{
		if(  model!="pik_rhol_epsilonkl"
			&& model!="pik_rhol_epsilon"
			&& model!="pi_rho_epsilonkl"
			&& model!="pi_rho_epsilon")
		{
			stop("Incorrect Model, Valid Binary models are:pik_rhol_epsilonkl, pik_rhol_epsilon
								pi_rho_epsilonkl, pi_rho_epsilon")
		}
	}
  # checking for strategy
	if(length(strategy@initmethod)==0){ strategy@initmethod = "emInitStep"}
  ## else
  ## {
  ##   if(strategy@initmethod!="cemInitStep" && strategy@initmethod!="emInitStep")
  ##     stop("In coclusterBinary. Incorrect initialization method, valid method(s) are: cemInitStep, emInitStep")
  ## }
	# checking for hyperparameters
  if ((a <=0)||(b<=0)) { stop("hyper-parameters must be positive")}
  #strategy@hyperparam = c(a,b);
	#  check nbCore
	if(!is.numeric(nbCore) && length(nbCore) != 1) stop("nbCore must be an integer")
  # create object
  inpobj<-new( "BinaryOptions",data = data
	           , rowlabels = rowlabels, collabels = collabels
	           , semisupervised = semisupervised,
		           datatype = "binary", model = model, nbcocluster = nbcocluster
					   , strategy = strategy
				     , hyperparam = c(a,b))

  .Call("CoClustmain",inpobj, nbCore,PACKAGE = "blockcluster")
  cat(inpobj@message,"\n")
  return(inpobj)
}