post_sample: Generates joint posterior samples from a trained GP/DGP

In deepgp: Bayesian Deep Gaussian Processes using MCMC

Generates joint posterior samples from a trained GP/DGP

Description

Acts on a gp, gpvec, dgp2, dgp2vec, dgp3, or dgp3vec object. Generates joint samples from the posterior distribution at the provided locations.

Usage

post_sample(object, x_new, nper = 1, ...)

## S3 method for class 'gp'
post_sample(object, x_new, nper = 1, grad = FALSE, cores = 1, ...)

## S3 method for class 'gpvec'
post_sample(
  object,
  x_new,
  nper = 1,
  m = NULL,
  ord_new = NULL,
  grad = FALSE,
  cores = 1,
  ...
)

## S3 method for class 'dgp2'
post_sample(
  object,
  x_new,
  nper = 1,
  grad = FALSE,
  mean_map = TRUE,
  cores = 1,
  ...
)

## S3 method for class 'dgp2vec'
post_sample(
  object,
  x_new,
  nper = 1,
  m = NULL,
  ord_new = NULL,
  grad = FALSE,
  mean_map = TRUE,
  cores = 1,
  ...
)

## S3 method for class 'dgp3'
post_sample(object, x_new, nper = 1, mean_map = TRUE, cores = 1, ...)

## S3 method for class 'dgp3vec'
post_sample(
  object,
  x_new,
  nper = 1,
  m = NULL,
  ord_new = NULL,
  mean_map = TRUE,
  cores = 1,
  ...
)

Arguments

object	object from fit_one_layer, fit_two_layer, or fit_three_layer with burn-in already removed
x_new	vector or matrix of predictive input locations
nper	the number of samples to generate from each MCMC iteration. The total number of samples will equal nper*object$nmcmc
...	N/A
grad	logical indicating whether to additionally calculate/return samples of the gradient (one and two layer models only)
cores	number of cores to use for SNOW parallelization
m	size of Vecchia conditioning sets (only for fits with vecchia = TRUE), defaults to the twice the m used for MCMC
ord_new	optional ordering for Vecchia approximation with lite = FALSE, must correspond to rows of x_new, defaults to random, is applied to all layers in deeper models
mean_map	logical indicating whether to map hidden layers using conditional mean (mean_map = TRUE) or using a random sample from the full MVN distribution (two or three layer models only)

Details

By default, one sample is generated per each MCMC iteration. This may be increased with the nper argument.

SNOW parallelization reduces computation time but requires more memory storage.

Value

If grad = FALSE, returns matrix of samples. Rows correspond to x_new locations. If grad = TRUE, returns a list with y and dydx containing the respective samples.

References

Sauer, A. (2023). Deep Gaussian process surrogates for computer experiments. *Ph.D. Dissertation, Department of Statistics, Virginia Polytechnic Institute and State University.*

Examples

# Simple step function
f <- function(x) {
  return(pnorm((x - 0.5) / 0.065))
}

# Training data
x <- seq(0, 1, length = 5)
y <- f(x)

# Testing data
xx <- seq(0, 1, length = 100)
yy <- f(xx)

plot(xx, yy, type = "l")
points(x, y, col = 2)

# Conduct MCMC
fit <- fit_two_layer(x, y, nmcmc = 2000, true_g = 1e-6, cov = "exp2")
plot(fit, hidden = TRUE)
fit <- trim(fit, 1000, 2)

# Generate posterior samples, including gradients
samples <- post_sample(fit, xx, grad = TRUE, cores = 1)

# Plot samples
par(mfrow = c(1, 2))
matplot(xx, t(samples$y), type = "l")
points(x, y, pch = 20)
matplot(xx, t(samples$dy), type = "l")

deepgp documentation built on July 30, 2026, 5:12 p.m.