Observation/prediction matrices for mesh models.

Constructs observation/prediction weight matrices for models based on inla.mesh() and inla.mesh.1d() objects.

For a more modular approach, see fmesher::fm_basis(), fmesher::fm_row_kron(), fmesher::fm_block(), and the inlabru bru_mapper() system.

inla.spde.make.A(
  mesh = NULL,
  loc = NULL,
  index = NULL,
  group = NULL,
  repl = 1L,
  n.spde = NULL,
  n.group = NULL,
  n.repl = NULL,
  group.mesh = NULL,
  weights = NULL,
  A.loc = NULL,
  A.group = NULL,
  group.index = NULL,
  block = NULL,
  n.block = NULL,
  block.rescale = c("none", "count", "weights", "sum"),
  ...
)

Arguments

mesh

An inla.mesh() or inla.mesh.1d() object specifying a function basis on a mesh domain. Alternatively, an inla.spde object that includes a mesh (e.g. from inla.spde2.matern()).

loc

Observation/prediction coordinates. mesh and loc defines a matrix A.loc of mapping weights between basis function weights and field values. If loc is NULL, A.loc is defined as Diagonal(n.spde, 1).

index

For each observation/prediction value, an index into loc. Default is seq_len(nrow(A.loc)).

group

For each observation/prediction value, an index into the group model.

repl

For each observation/prediction value, the replicate index.

n.spde

The number of basis functions in the mesh model. (Note: may be different than the number of mesh vertices/nodes/knots.)

n.group

The size of the group model.

n.repl

The total number of replicates.

group.mesh

An optional inla.mesh.1d() object for the group model.

weights

Optional scaling weights to be applied row-wise to the resulting matrix.

A.loc

Optional precomputed observation/prediction matrix. A.loc can be specified instead of mesh+loc, optionally with index supplied.

A.group

Optional precomputed observation/prediction matrix for the group model. A.group can be specified instead of group and/or group.mesh, optionally with group.index supplied.

group.index

For each observation/prediction value, an index into the rows of A.group.

block

Optional indices specifying block groupings: Entries with the same block value are joined into a single row in the resulting matrix, and the block values are the row indices. This is intended for construction of approximate integration schemes for regional data problems. See inla.spde.make.block.A() for details.

n.block

The number of blocks.

block.rescale

Specifies what scaling method should be used when joining entries as grouped by a block specification. See inla.spde.make.block.A() for details.

...

Additional parameters. Currently unused.

Author

Finn Lindgren finn.lindgren@gmail.com

Examples


loc <- matrix(runif(10000 * 2) * 1000, 10000, 2)
mesh <- inla.mesh.2d(
    loc = loc,
    cutoff = 50,
    max.edge = c(50, 500)
)
A <- inla.spde.make.A(mesh, loc = loc)