r 中的稀疏矩阵大小与常规矩阵大小

Question

My regular matrix has object size 416 bytes, and when I use as(, "sparseMatrix") to turn it into sparse matrix, then the size for this sparse matrix goes up to 1720 bytes.

Is it normal? Shouldn't we expect a smaller storage size for the sparse matrix than the regular one?

Many thanks in advance!

Answer 1

matrix is one of the base data structures of R, and can be stored with very little metadata: it is a sequence of values with just a length for each dimension, and a data type.

A sparseMatrix object however contains more metadata, as you'll see with str() in the examples below. Most prominently, for each non-zero value an (x,y) position is stored in addition to the value itself. This alone will cause a threefold increase in memory use, if you're storing integers. This is only compensated when there are many zero values, as they are not stored at all.

Dense example

Compare for a matrix with no zero values:

> mat1 = matrix( sample(3*3), c(3, 3))
> smat1 <- as(mat1, "sparseMatrix")

> showMem(c('mat1', 'smat1'), bytes=T)
        size bytes
mat1   264 B   264
smat1 1.7 kB  1688

> mat1
     [,1] [,2] [,3]
[1,]    2    5    7
[2,]    8    6    1
[3,]    3    4    9

> str(mat1)
 int [1:3, 1:3] 2 8 3 5 6 4 7 1 9

> str(smat1)
Formal class 'dgCMatrix' [package "Matrix"] with 6 slots
  ..@ i       : int [1:9] 0 1 2 0 1 2 0 1 2
  ..@ p       : int [1:4] 0 3 6 9
  ..@ Dim     : int [1:2] 3 3
  ..@ Dimnames:List of 2
  .. ..$ : NULL
  .. ..$ : NULL
  ..@ x       : num [1:9] 2 8 3 5 6 4 7 1 9
  ..@ factors : list()

Or a larger version of such a matrix:

> mat2 = matrix( sample(1000*1000), c(1000, 1000))
> smat2 <- as(mat2, "sparseMatrix")

> showMem(c('mat2', 'smat2'), bytes=T)
       size    bytes
mat2   4 MB  4000216
smat2 12 MB 12005504

Sparse example

Here we create a more sparse matrix, with 6 zeroes and only 3 values. We can see that the sparseMatrix only stores the 3 values.

> mat3 = matrix( sample(3*3)%%3%%2, c(3, 3))
> smat3 <- as(mat3, "sparseMatrix")

> showMem(c('mat3', 'smat3'), bytes=T)
        size bytes
mat3   344 B   344
smat3 1.6 kB  1560

> mat3
     [,1] [,2] [,3]
[1,]    0    1    0
[2,]    0    0    0
[3,]    1    0    1

> str(mat3)
 num [1:3, 1:3] 0 0 1 1 0 0 0 0 1

> str(smat3)
Formal class 'dgCMatrix' [package "Matrix"] with 6 slots
  ..@ i       : int [1:3] 2 0 2
  ..@ p       : int [1:4] 0 1 2 3
  ..@ Dim     : int [1:2] 3 3
  ..@ Dimnames:List of 2
  .. ..$ : NULL
  .. ..$ : NULL
  ..@ x       : num [1:3] 1 1 1
  ..@ factors : list()

And finally a case where the sparseMatrix gives the expected memory savings:

> mat4 = matrix( sample(1000*1000)%%3%%2, c(1000, 1000))

> smat4 <- as(mat4, "sparseMatrix")

> table(mat4)
mat4
     0      1 
666666 333334 

> showMem(c('mat4', 'smat4'), bytes=T)
      size   bytes
mat4  8 MB 8000216
smat4 4 MB 4005512