rGNN: Sampling from a Generative Neural Network
In gnn: Generative Neural Networks

Description Usage Arguments Value Author(s) Examples

Sampling method for objects of S3 class "gnn_GNN".

1 2	## S3 method for class 'gnn_GNN' rGNN(x, size, prior = NULL, pobs = FALSE, ...)

`x`	object of `S3` class `"gnn_GNN"`.
`size`	sample size, a positive `integer`. Ignored if `prior` is a `matrix`.
`prior`	one of `NULL` the default, generates independent N(0,1) realizations as prior sample. `matrix` passes the given sample through the GNN `x`. Such a matrix is internally (if `prior = NULL`) and typically obtained via `rPrior()`.
`pobs`	`logical` indicating whether the pseudo-observations of the generated samples should be returned.
`...`	additional arguments passed to the underlying `rPrior()` if `prior = NULL`.

(size, dim(x)[1])-matrix of samples.

Marius Hofert

if(TensorFlow_available()) { # rather restrictive (due to R-Forge, winbuilder)
library(gnn) # for being standalone

## Define dummy model
d <- 2 # bivariate case
GMMN <- FNN(c(d, 300, d)) # Feedforward NN with MMD loss (a GMMN; random weights)

## Sampling
n <- 3
(X1 <- rGNN(GMMN, size = n)) # default (independent N(0,1) samples as prior)
(X2 <- rGNN(GMMN, size = n, # passing additional arguments to rPrior()
            qmargins = qexp, method = "sobol", seed = 271))
(X3 <- rGNN(GMMN, prior = matrix(rexp(n * d), ncol = d))) # providing 'prior'
stopifnot(dim(X1) == c(n, d), dim(X2) == c(n, d), dim(X3) == c(n, d))

}