README.md

In liujl93/spKrylov: Krylov Method for Large Spatial Datasets

spKrylov

Krylov Methods for Large Spatial Datasets

Installation

This package can be installed directly from R by running the following code:

Download the binary file

• Windows binaries: spKrylov_1.0.zip
• OS X binaries: spKrylov_1.0.tgz

## MacOS
install.packages("/path/to/spKrylov_1.0.tgz", repos = NULL, type = .Platform$pkgType)
## Windows
install.packages("/path/to/spKrylov_1.0.zip", repos = NULL, type = "win.binary")

Example

This package includes a data file sim_n_4900rep_1delta_6. It includes the following objects: - train: N by 3 matrix, first two columns stores the coordinates and the last column stores the observations - test: N_ho by 3 matrix, first two columns are coordinates and the last column are observations - delta: only distances smaller than delta are recorded, used in nearest.dist() - dist_mat: distance matrix of train data, only distances smaller than delta are recorded, stored in compressed form (class: spam)

Install Packages

rm(list=ls(all=TRUE))
library(spKrylov)
library(foreach)
library(spam)

Approximate Negative Log-likelihood

set.seed(2019)

n <- 70^2 # sample size
delta <- 6 #    only distances smaller than delta are recorded.
rep <- 1
data("sim_n_4900rep_1delta_6")
theta <- c(2, 0.2)

dist_mat <- nearest.dist(train[, 1:2],
  miles = FALSE,
  upper = NULL,
  delta = delta
)
#########################################################
### Approximate Negative Log-likelihood by Krylov Method
#########################################################
system.time(
  lik2 <- krylov_neg_loglik(
    theta = theta,
    y = train[, 3],
    dist_mat = dist_mat,
    cov.model = "exponential",
    cov.taper = "wend1",
    delta = delta,
    ctrl = list(
      cg.tol = 1e-6, cg.max_iter = 1000,
      cg.precond = "no_precond",
      logdet.m = 50, logdet.nr = 1
    ),
    gls = FALSE
  )
)

lik2

Find MLE

set.seed(2019)
n <- 70^2 # sample size
delta <- 6 #    only distances smaller than delta are recorded.
rep <- 1
data("sim_n_4900rep_1delta_6")
nr <- 1

dist_mat <- nearest.dist(train[, 1:2],
  miles = FALSE,
  upper = NULL,
  delta = delta
)

system.time(
  fit.mle <- krylov_mle(
    y = train[, 3],
    dist_mat = dist_mat,
    cov.model = "exponential",
    cov.taper = "wend1",
    theta.init = c(3, 0.15),
    theta.lower = c(0.0001, 0.0001),
    theta.upper = c(max(dist_mat), 1),
    delta = delta,
    ctrl = list(
      cg.tol = 1e-6, cg.max_iter = 1000,
      cg.precond = "no_precond",
      logdet.m = 30, logdet.nr = nr
    ),
    gls = FALSE
  )
)

fit.mle$result$theta

liujl93/spKrylov documentation built on Aug. 26, 2020, 8:46 p.m.