locppm {spatstat.local} | R Documentation |
Locally Fitted Poisson or Gibbs Point Process Model
Description
Fits Poisson or Gibbs point process model using local likelihood or pseudolikelihood.
Usage
locppm(..., sigma = NULL, f = 1/4,
vcalc = c("none", "t", "hessian", "hom", "lik", "full"),
locations=c("split", "fine", "coarse"),
ngrid = NULL, grideps = NULL, verbose = TRUE,
use.fft=FALSE, fft.algorithm="closepairs")
Arguments
... |
Arguments passed to |
sigma |
Standard deviation of Gaussian kernel for local likelihood. |
f |
Argument passed to |
vcalc |
Type of variance calculation to be performed. See Details. |
locations |
Spatial locations for local calculations. See Details. |
ngrid |
Dimensions of coarse grid, if used. See Details.
Incompatible with |
grideps |
Grid spacing of coarse grid, if used. See Details.
Incompatible with |
verbose |
Logical. If |
use.fft |
Logical value indicating whether to perform
computations using the Fast Fourier Transform.
With |
fft.algorithm |
Developer use only. |
Details
This function fits a Poisson or Gibbs point process model to point pattern data by local likelihood or local pseudolikelihood respectively.
This command should be used in the same way as
ppm
.
The point pattern data and the specification of the model
are given in the leading arguments ...
which are passed
directly to ppm
.
In all cases, the local estimates of the coefficients are
computed. However, because the variance calculations are
time-consuming, the default is not to perform them.
This is controlled by the argument vcalc
.
vcalc = "none"
:-
no variance calculations are performed.
vcalc = "t"
:-
the
t
statistic for each parameter is computed for the local model. vcalc = "hessian"
:-
the local Hessian matrix is computed, and its negative inverse is used as a surrogate for the local variance.
vcalc = "hom"
:-
No local fitting is performed. Calculations are performed only for the homogeneous (template) model. The variance of the local parameter estimates under the homogeneous model is computed.
vcalc = "lik"
:-
In addition to the calculations for
vcalc="hom"
described above, ifuse.fft=FALSE
the algorithm also computes the local composite likelihood ratio test statistic for the test of homogeneity. Ifuse.fft=TRUE
thenvcalc="lik"
is equivalent tovcalc="hom"
. vcalc = "full"
:-
all variance calculations are performed for the local model.
The spatial locations, where the model fits and variance calculations
are performed, are determined by the argument locations
.
locations = "fine"
:-
The calculations are performed at every quadrature point of the model. This can take a very long time.
locations = "coarse"
:-
The calculations are performed at the points of a coarse grid with dimensions specified by
ngrid
orgrideps
. locations = "split"
:-
The fitted coefficients are computed at every quadrature point of the model, but the variance calculations (if any) are performed at a coarse grid of locations, specified by
ngrid
orgrideps
. If neitherngrid
norgrideps
is specified, the default isngrid=10
.
If use.fft=FALSE
(the default), all desired quantities are
computed exactly, by an iterative algorithm that
fits a separate model at each spatial location. This can be quite
slow.
If use.fft=TRUE
, we only compute quantities that can be
obtained using the Fast Fourier Transform, resulting in much faster
calculations (sometimes 3 orders of magnitude faster) when
locations="fine"
.
Properties of the homogeneous model are
computed accurately. Properties of the locally-fitted model are
approximated by a first order Taylor expansion.
Value
An object of class "locppm"
representing the fitted model.
Author(s)
Adrian Baddeley Adrian.Baddeley@curtin.edu.au.
References
Baddeley, A. (2017) Local composite likelihood for spatial point patterns. Spatial Statistics 22, 261–295. DOI: 10.1016/j.spasta.2017.03.001
Baddeley, A., Rubak, E. and Turner, R. (2015) Spatial Point Patterns: Methodology and Applications with R. Chapman and Hall/CRC Press.
See Also
Examples
fit <- locppm(swedishpines, ~1, sigma=9, nd=20)
fit