[R] sparse matrix from vector outer product

[Martin Maechler]

>>>>> Philipp A <flying-sheep at web.de>
>>>>>     on Wed, 14 Jan 2015 14:02:40 +0000 writes:

    > Hi,
    > creating a matrix from two vectors a, b by multiplying each combination can
    > be done e.g. via

    > a %*% t(b)

    > or via

    > outer(a, b)  # default for third argument is '*'

really the best (most efficient) way would be

   tcrossprod(a, b)

    > But this yields a normal matrix.
of course.

Please always use small self-contained example code,
here, e.g.,

a <- numeric(17); a[3*(1:5)] <- 10*(5:1)
b <- numeric(12); b[c(2,3,7,11)] <- 1:3


    > Is there an efficient way to create sparse matrices (from the Matrix
    > package) like that?

    > Right now i’m doing

    > a.sparse = as(a, 'sparseVector')
    > b.sparse = as(t(b), 'sparseMatrix')
    > a.sparse %*% b.sparse

    > but this strikes me as wasteful.

not really wasteful I think. But there is a nicer and more efficient way :

require(Matrix)
tcrossprod(as(a, "sparseVector"),
           as(b, "sparseVector"))

now also gives

 17 x 12 sparse Matrix of class "dgCMatrix"

  [1,] .  .   . . . .   . . . .  . .
  [2,] .  .   . . . .   . . . .  . .
  [3,] . 50 100 . . . 150 . . . 50 .
  [4,] .  .   . . . .   . . . .  . .
  [5,] .  .   . . . .   . . . .  . .
  [6,] . 40  80 . . . 120 . . . 40 .
  [7,] .  .   . . . .   . . . .  . .
  [8,] .  .   . . . .   . . . .  . .
  [9,] . 30  60 . . .  90 . . . 30 .
 [10,] .  .   . . . .   . . . .  . .
 [11,] .  .   . . . .   . . . .  . .
 [12,] . 20  40 . . .  60 . . . 20 .
 [13,] .  .   . . . .   . . . .  . .
 [14,] .  .   . . . .   . . . .  . .
 [15,] . 10  20 . . .  30 . . . 10 .
 [16,] .  .   . . . .   . . . .  . .
 [17,] .  .   . . . .   . . . .  . .
>