R/impute.nas.by.surrounding.R

In SIM: Integrated Analysis on two human genomic datasets

impute.nas.by.surrounding <- function(dataset, window.size=5) {
    # Create a matrix to hold the results.
    #retval=as.data.frame( matrix(0, nrow(dataset), ncol(dataset)) )
    #rownames(retval)=rownames(dataset)
    #colnames(retval)=colnames(dataset)
    
    
    # The following schematic shows the meaning of the variables used in the
    # rest of the function. A (part of a) vector is depicted. "start" indicates
    # the start of the window under scrutiny. "end" indicates the "end" of the
    # window. "offset" indicates the center.
    #
    #
    # |  X  |  X  |  X  |  X  |  X  |  NA  |  X  |  X  |  X  |  X  |
    # |     |     |     |start|     |offset|     | end |     |     |
    #
    # window.radius=(window.size - 1) / 2
    # start =offset - window.radius
    # end   =offset + window.radius
    #
    method <-  median
    window.radius <- (window.size - 1) / 2
    my.seq <- seq(1,nrow(dataset),1)
    
    # Iterate over the columns (samples) in the dataset.
    for (col.idx in 1:ncol(dataset)) {
        
        # If the current column contains NAs, process it ...
        NAs=my.seq[is.na(dataset[,col.idx])]
        
        if ( length(NAs) > 0 ) {
            cat("Processing column: ", col.idx, sep="", "\n")
            
            for(i in 1:length(NAs)){
                offset=NAs[i]
                # We'll only have to impute things if the entry at "offset" happens to
                # be an NA.
                start=offset - window.radius
                end  =offset + window.radius
                
                # If we're at the beginning of the stream, add a couple of items from
                # the other end until we're at "window.size".
                if (start < 0) {
                    end  =end + abs(start) + 1
                    start=0
                }
                
                # If we're at the end of the stream, add a couple of items from the
                # other end until we're at "window.size".
                if (end > nrow(dataset)) {
                    start=start - (end - nrow(dataset)) - 1
                    end  =nrow(dataset)
                }
                
                window=dataset[start:end, col.idx]
                NAs.in.window=is.na(window)
                
                # If there's only a single NA in the window, we can impute it.
                # If not, we'll let it be and remove the row later on (just before
                # we return the dataset).
                if ( sum(NAs.in.window) == 1 ) {
                    # Before computing anything, we need to remove the NA from the
                    # window. Otherwise, it'll screw up computations.
                    indices=!is.na(window)
                    valid.window=window[indices]
                    
                    imputed.value=method(valid.window)
                    
                    dataset[offset, col.idx]=imputed.value
                }
            }
        }
    }
    return( na.omit(dataset) )
}


#impute.nas.by.surrounding <- function(x, window.size=5)
#{
#		
#}
#