gp_fit: gp_fit fits a Gaussian process to the observed data

In cboettig/nonparametric-bayes: Nonparametric Bayes inference for ecological models

Description

API not yet stable! Arguments to this function will probably be updated to be much more generic

Usage

gp_fit(obs, X, pars = c(sigma_n = 1, tau = 1, l = 1), method = c("direct",
  "sequential", "kernlab", "cholesky", "conditional"), fit = TRUE,
  out_var = NULL, ...)

Arguments

obs	a data frame of observations with columns obs$x and obs$y
X	the desired points over which to predict
pars	a named numeric specifying "sigma_n" for the (additive) noise and "l" for the covariance length scale
method	select the method to use
out_var	optionally force kernlab method to generate Cf using a different variance than sigma_n^2 assumed in the fit
fit	logical, argument for the kernlab::gausspr method only
...	potential additional arguments, unimplemented

Details

so far treates the prior as mean 0 and covariance given by cov(X)

Value

a list with items "mu", the expected Y values at X (mean of the posterior Gaussian process), Sigma, the covariance matrix for the posterior Gaussian process, and "loglik", the log likelihood of observering the given data under the process, marginalized over the prior

cboettig/nonparametric-bayes documentation built on May 13, 2019, 2:09 p.m.