SAMCpack (version 0.1.1)

SAMCrsa: A Resampling-based Stochastic Approximation Method for Analysis of Large Geostatitical data

Description

Performs parameter estimation using a resampling-based Stochastic Approximation (RSA) method. It is a stochatic approximation method. At every iteration, only a subset of the data is drawn and used to update the estimation of the parameters. The data are assumed to have a powered exponential correlation structure.

Usage

SAMCrsa(coords, y, X = NULL, nsubset = max(ceiling(length(y)/5), 10),
  stepscale = 200, niter = 2500, warm = 100)

Arguments

coords

an \((n\times 2)\) matrix. 2D location coordinates.

y

a length-\(n\) vector of response value.

X

an \((n\times k)\) matrix of extra covariates.

nsubset

the size of the subset drawn from the data. It is recommended to be set to 300 or higher.

stepscale

gain factor control. It specifies the number of iterations when the gain factor begins to shrink. For example, one can be set it equal to 2 times the burn-in steps.

niter

the total number of iterations for stochastic approximation. In practice, it is recommended to be set to 2500 or higher.

warm

the number of burn-in iterations

Value

a named list containing

beta

the coefficient estimates of the mean effect. It is a vector of length equal to the number of coefficients plus 1.

phi

the shape estimate in the powered exponential correlation matrix.

sigmasq

the estimate of error variance.

tausq

the estimate of nugget variance.

References

SAMCrsaSAMCpack

Examples

# NOT RUN {
##### load example data pre-loaded
data(gdata)

##### run RSA
output = SAMCrsa(gdata$coords, gdata$y, gdata$X, nsubset=50, stepscale=40, niter=100, warm=20)

# }