predict.dstm: Predict Method for DSTM Fits

In ideq: Bayesian Dynamic Spatio-Temporal Models, Including the Integrodifference Equation Model

Description

Generates samples from the posterior predictive distribution at future time points for (1) the observation vector and (2) the state vector.

Usage

## S3 method for class 'dstm'
predict(
  object,
  K = 1,
  only_K = FALSE,
  return_ys = TRUE,
  return_thetas = FALSE,
  burnin = NULL,
  ...
)

Arguments

object	A 'dstm' object
K	(integer scalar) The number of future time periods for which to generate predictions
only_K	(logical scalar) Whether to return predictions for time period T+K only (as opposed to T+1, T+2, ..., T+K)
return_ys	(logical scalar) Whether to return samples from the posterior predictive distribution of the observation vector (ys)
return_thetas	(logical scalar) Whether to return samples from the posterior predictive distribution of the state vector (thetas)
burnin	(integer scalar) The number of samples to discard as burn-in. If object$burnin exists, this argument will override it.
...	Arguments passed to other methods (necessary for S3 generic compatibility)

Details

The posterior predictive samples are returned in a matrix or 3-D array, depending on whether samples from multiple time points are requested. The dimensions are always in the following order:

1. The index of the value within the state or observation vector.

2. The time period

3. The sample number

Examples

data("ide_standard", "ide_locations")

# IDE example
mod_ide <- dstm_ide(ide_standard, ide_locations)
predict(mod_ide)
predict(mod_ide, K=4, return_thetas=TRUE)

# EOF example
mod_eof <- dstm_eof(ide_standard, n_samples=2)
predict(mod_eof, K=2, only_K=TRUE, burnin=1)

ideq documentation built on Dec. 20, 2019, 1:07 a.m.