Filling sparse matrices

[Jan-Ole Koslik]

Hi,

I want to initialize and fill a sparse matrix inside my likelihood function. 

I only need to fill value on the super- and subdiagonal.

I felt like the most elegant way would be

 A <- Matrix::Matrix(0, m, m) # m is integer
 A[cbind(1:(m-1), 2:m)] <- some advector of suitable length

but this gives

Error in .local(x, i, j, ..., value) :
  not-yet-implemented 'Matrix[<-' method

Alternatively I tried

  i_up   <- 1:(m - 1)
  i_down <- 2:m

  up_vals   <- ...
  down_vals <- ...

  i <- c(i_up, i_down)
  j <- c(i_up + 1, i_down - 1)
  x <- c(up_vals, down_vals) 
  Q <- Matrix::sparseMatrix(i = i, j = j, x = x, dims = c(m, m))

but this gives

Error in getClass(Class, where = topenv(parent.frame())) :
  “gTMatrix” is not a defined class

So my question is simply, at this point in time, is there a way to do this semi elegantly without a loop?

Cheers
  

[Jan-Ole Koslik]

And obviously, I could also construct A as a dense matrix and afterwards convert to sparse using something like A <- as(A, "sparseMatrix") but I don't know if that is wasteful.

[Kasper Kristensen]

I think this issue is explained under ?AD. Just wrap AD(.) around Matrix(0,m,m):

F <- MakeTape(function(x) {A <- AD(Matrix(0, 5, 5)); A[cbind(1:4, 2:5)] <- x; A}, 1:4)

[Jan-Ole Koslik]

Thanks!