Creating sparse matrix from a list of sparse vectors

问题

I have a list of sparse vectors (in R). I need to convert this list to a sparse matrix. Doing it via a for-loop takes a long time.

sm<-spMatrix(length(tc2),n.col)
for(i in 1:length(tc2)){
    sm[i,]<-(tc2[i])[[1]];  
}

Is there a better way?

回答1:

Here is a two step solution:

Use lapply() and as(..., "sparseMatrix") to convert the list of sparseVectors to a list of one column sparseMatrices.
Use do.call() and cBind() to combine the sparseMatrices in a single sparseMatrix.

require(Matrix)

# Create a list of sparseVectors
ss <- as(c(0,0,3, 3.2, 0,0,0,-3), "sparseVector")
l <- replicate(3, ss)

# Combine the sparseVectors into a single sparseMatrix
l <- lapply(l, as, "sparseMatrix")
do.call(cBind, l)

# 8 x 3 sparse Matrix of class "dgCMatrix"
#                    
# [1,]  .    .    .  
# [2,]  .    .    .  
# [3,]  3.0  3.0  3.0
# [4,]  3.2  3.2  3.2
# [5,]  .    .    .  
# [6,]  .    .    .  
# [7,]  .    .    .  
# [8,] -3.0 -3.0 -3.0

回答2:

This scenario, cbinding a bunch of vectors, is set up perfectly for dumping the information right into a sparse, column-oriented matrix (dgCMatrix class).

Here's a function that will do it:

sv.cbind <- function (...) {
    input <- lapply( list(...), as, "dsparseVector" )
    thelength <- unique(sapply(input,length))
    stopifnot( length(thelength)==1 )
    return( sparseMatrix( 
            x=unlist(lapply(input,slot,"x")), 
            i=unlist(lapply(input,slot,"i")), 
            p=c(0,cumsum(sapply(input,function(x){length(x@x)}))),
            dims=c(thelength,length(input))
        ) )
}

From a quick test, this looks to be about 10 times faster than coercion + cBind:

require(microbenchmark)
xx <- lapply( 1:10, function (k) {
            sparseVector( x=rep(1,100), i=sample.int(1e4,100), length=1e4 )
        } )
microbenchmark( do.call( sv.cbind, xx ), do.call( cBind, lapply(xx,as,"sparseMatrix") ) )
# Unit: milliseconds
#                                            expr       min        lq      mean   median       uq       max neval cld
#                           do.call(sv.cbind, xx)  1.398565  1.464517  1.540172  1.49487  1.55911  3.455421   100  a 
#  do.call(cBind, lapply(xx, as, "sparseMatrix")) 16.037890 16.356268 16.956326 16.59854 17.49956 20.256253   100   b

回答3:

Thanks to Josh O'Brien for suggesting a solution: create 3 lists, then create sparseMatrix. I include the code for this here:

vectorList2Matrix<-function(vectorList){
 nzCount<-lapply(vectorList, function(x) length(x@j));
 nz<-sum(do.call(rbind,nzCount));
 r<-vector(mode="integer",length=nz);
 c<-vector(mode="integer",length=nz);
 v<-vector(mode="integer",length=nz);
 ind<-1;
 for(i in 1:length(vectorList)){
    ln<-length(vectorList[[i]]@i);
    if(ln>0){
     r[ind:(ind+ln-1)]<-i;
     c[ind:(ind+ln-1)]<-vectorList[[i]]@j+1
     v[ind:(ind+ln-1)]<-vectorList[[i]]@x
     ind<-ind+ln;
    }
 }
 return (sparseMatrix(i=r,j=c,x=v));
}

来源：https://stackoverflow.com/questions/8843700/creating-sparse-matrix-from-a-list-of-sparse-vectors

标签

sparse-matrix