RBaM (version 1.1.1)

prediction: prediction object constructor.

Description

Creates a new instance of a 'prediction' object

Usage

prediction(
  X,
  spagFiles,
  data.dir = getwd(),
  data.fnames = paste0("X", 1:length(X), ".pred"),
  fname = paste0("Config_Pred_", paste0(sample(c(letters, LETTERS, 0:9), 6), collapse =
    ""), ".txt"),
  doParametric = FALSE,
  doStructural = rep(FALSE, length(spagFiles)),
  transposeSpag = TRUE,
  priorNsim = NULL,
  envFiles = paste0(tools::file_path_sans_ext(spagFiles), ".env"),
  consoleProgress = TRUE,
  spagFiles_state = NULL,
  transposeSpag_state = TRUE,
  envFiles_state = switch(is.null(spagFiles_state) + 1,
    paste0(tools::file_path_sans_ext(spagFiles_state), ".env"), NULL),
  parSamples = NULL
)

Value

An object of class 'prediction'.

Arguments

X

data frame or list of dataframes / matrices, representing the values taken by the input variables.

• If X is a dataframe, then each column is interpreted as one input variable, and consequently inputs are not replicated (=> no input uncertainty).

• If X is a list, then each element of the list is a matrix associated with one input variable, and the columns of this matrix are replications (=> input uncertainty is propagated). All matrices in the list should have the same number of rows, columns are recycled if needed.

spagFiles

Character vector (size nY, the number of output variables). Name of the files containing the spaghettis for each output variable. NOTE: provide file names only, not full paths. Using a '.spag' extension is a good practice.

data.dir

Character, directory where a copies of the dataset X will be written if required (1 file per variable in X). Default is the current working directory, but you may prefer to use the BaM workspace.

data.fnames

Character, data file names.

fname

Character, configuration file name.

doParametric

Logical, propagate parametric uncertainty? If FALSE, maxpost parameters are used.

doStructural

Logical, propagate structural uncertainty for each output variable? (size nY)

transposeSpag

Logical. If FALSE, spaghettis are written horizontally (row-wise), otherwise they will be transposed so that each spaghetti is a column.

priorNsim

Integer, number of samples from the prior distribution for 'prior prediction' experiments. If negative or NULL (default), posterior samples are used.

envFiles

Character vector (size nY, the number of output variables). Name of the files containing the envelops (e.g. prediction intervals) computed from the spaghettis for each output variable. By default, same name as spaghetti files but with a '.env' extension. If NULL, envelops are not computed.

consoleProgress

Logical, print progress in BaM.exe console?

spagFiles_state

Character vector (size nState, the number of state variables), same as spagFiles but for states rather than outputs. If NULL, states are not predicted. Note that only parametric uncertainty is propagated for state variables since they are not observed. Consequently, structural uncertainty = 0 and total uncertainty = parametric uncertainty.

transposeSpag_state

Logical. Same as transposeSpag, but for states rather than outputs.

envFiles_state

Character vector (size nState, the number of state variables), same as envFiles, but for states rather than outputs.

parSamples

data frame, parameter samples that will replace the MCMC-generated one for this prediction.

Examples

#--------------
# Example using the twoPopulations dataset, containing 101 values for
# 3 input variables (time t, temperature at site 1 T1, temperature at site 2 T2)
# and 2 output variables (population at site 1 P1, population at site 2 P2).
pred=prediction(X=twoPopulations[,1:3],spagFiles=c('P1.spag','P2.spag'))
#--------------
# Alternative example showing how to propagate uncertainty in some of
# the input variables (here, temperatures T1 and T2)
# Create 100 noisy replicates for T1, representing uncertainty
T1rep=matrix(rnorm(101*100,mean=twoPopulations$T1,sd=0.1),nrow=101,ncol=100)
# Same for T2
T2rep=matrix(rnorm(101*100,mean=twoPopulations$T2,sd=0.1),nrow=101,ncol=100)
# Create prediction object
pred=prediction(X=list(twoPopulations$t,T1rep,T2rep),spagFiles=c('P1.spag','P2.spag'))