[R] General indexing in multidimensional arrays

[Jannis]

Thanks, Gene, for your hint! I indeed did not check any possible 
situation and my function was not returning what I intened it to return. 
This updated version, however, should. I am sure there are much easier 
ways (or ready made functions) to do the same.



ind.datacube = function(
##title<< create logical index matrices for multidimensional datacubes
   datacube       ##<< array: datacube from which to extract the subparts
   , logical.ind  ##<< logical array: TRUE/FALSE index matrix for a 
subset of the dimensions
                  ##   of datacube. The size of logical.ind`s dimesnions 
has to match the
                  ##   sizes of the corresponding dimensions in datacube.
   , dims='auto'  ##<< integer vector or 'auto' : indices of the 
dimensions in datacube corresponding
                  ##   to the dimensions of logical.ind. If set to 
'auto' this matching is tried to
                  ##   be accomplished by comparing the sizes of the 
dimensions of the two objects.
)
{
     if (sum(logical.ind) == 0) {
         stop('No TRUE value in index matrix!')
     } else {
         if (dims[1] == 'auto')
         {
             if (is.null(dim(logical.ind)[1])) {
                 size.ind     = length(logical.ind)
                 logical.ind  = matrix(logical.ind,ncol=1)
             } else {
                 size.ind = dim(logical.ind)
             }
             dims = match(size.ind, dim(datacube))
             if (sum(duplicated(size.ind)) > 0 || sum(duplicated(dims)) 
 > 0 )
                 stop('dimensions do not match unambigously. Supply dims 
manually!')
         }
         dims.all <- setdiff(1:length(dim(datacube)),dims)
         ind.matrix.choice <- which(logical.ind, arr.ind = TRUE)

         dims.all.expand <- list()
         for (i in 1:length(dims.all))
            dims.all.expand[[i]] <- 1:dim(datacube)[dims.all[i]]
         dims.all.grid <-  as.matrix(do.call(expand.grid,   
dims.all.expand))

         expgrid.dims.all <- as.matrix(do.call(expand.grid, 
dims.all.expand))
         dims.all.mat <-  
matrix(rep(dims.all.grid,times=2),ncol=length(dims.all))
         ind.matrix.all <- 
cbind(ind.matrix.choice[rep(1:dim(ind.matrix.choice)[1],each=dim(dims.all.grid)[1]),] 
,
                                 dims.all.mat)
         ind.matrix.ord <- ind.matrix.all[,order(c(dims,dims.all))]
     }
     colnames(ind.matrix.ord) <- paste('dim',1:length(dim(datacube)),sep='')
     ##value<< integer index matrix which can be used to index datacube
     ind.matrix.ord
}


data <- array(rnorm(64),dim=c(4,4,4))
indices <- matrix(FALSE,ncol=4,nrow=4)
indices[1,3] <- TRUE
indices[4,1] <- TRUE
#result <- data[indices,]
ind.datacube(data, indices, dims=c(1,2))




On 08/04/2011 09:20 PM, Gene Leynes wrote:
> data<- array(rnorm(64),dim=c(4,4,4))