jags.model {rjags}

Create a JAGS model object

Description

jags.model is used to create an object representing a Bayesian graphical model, specified with a BUGS-language description of the prior distribution, and a set of data.

Usage

 jags.model(file, data, inits,
            n.chains = 1, n.adapt=1000, quiet=FALSE)

Arguments

file

the name of the file containing a description of the model in the JAGS dialect of the BUGS language.

Alternatively, file can be a readable text-mode connection, or a complete URL.

data

a list or environment containing the data. Any numeric objects in data corresponding to node arrays used in file are taken to represent the values of observed nodes in the model.

inits

optional specification of initial values in the form of a list or a function (see Initialization below). If omitted, initial values will be generated automatically. It is an error to supply an initial value for an observed node.

n.chains

the number of parallel chains for the model

n.adapt

the number of iterations for adaptation. See adapt for details. If n.adapt = 0 then no adaptation takes place.

quiet

if TRUE then messages generated during compilation will be suppressed, as well as the progress bar during adaptation.

Value

jags.model returns an object inheriting from class jags which can be used to generate dependent samples from the posterior distribution of the parameters

An object of class jags is a list of functions that share a common environment. This environment encapsulates the state of the model, and the functions can be used to query or modify the model state.

ptr()

Returns an external pointer to an object created by the JAGS library

data()

Returns a list containing the data that define the observed nodes in the model

model()

Returns a character vector containing the BUGS-language representation of the model

state(internal=FALSE)

Returns a list of length equal to the number of parallel chains in the model. Each element of the list is itself a list containing the current parameter values in that chain. if internal=TRUE then the returned lists also include the RNG names (.RNG.name) and states (.RNG.state). This is not the user-level interface: use the coef.jags method instead.

update(n.iter)

Updates the model by n.iter iterations. This is not the user-level interface: use the update.jags method instead.

Initialization

There are various ways to specify initial values for a JAGS model. If no initial values are supplied, then they will be generated automatically by JAGS. See the JAGS User Manual for details. Otherwise, the options are as follows:

1. A list of numeric values. Initial values for a single chain may supplied as a named list of numeric values. If there are multiple parallel chains then the same list is re-used for each chain.

2. A list of lists. Distinct initial values for each chain may be given as a list of lists. In this case, the list should have the same length as the number of chains in the model.

3. A function. A function may be supplied that returns a list of initial values. The function is called repeatedly to generate initial values for each chain. Normally this function should call some random number generating functions so that it returns different values every time it is called. The function should either have no arguments, or have a single argument named chain. In the latter case, the supplied function is called with the chain number as argument. In this way, initial values may be generated that depend systematically on the chain number.

Random number generators

Each chain in a model has its own random number generator (RNG). RNGs and their initial seed values are assigned automatically when the model is created. The automatic seeds are calculated from the current time.

If you wish to make the output from the model reproducible, you may specify the RNGs to be used for each chain, and their starting seeds as part of the inits argument (see Initialization above). This is done by supplementing the list of initial parameter values for a given chain with two additional elements named “.RNG.name”, and “.RNG.seed”:

.RNG.name

a character vector of length 1. The names of the RNGs supplied in the base module are:

• “base::Wichmann-Hill”

• “base::Marsaglia-Multicarry”

• “base::Super-Duper”

• “base::Mersenne-Twister”

If the lecuyer module is loaded, it provides “lecuyer::RngStream”

.RNG.seed

a numeric vector of length 1 containing an integer value.

Note that it is also possible to specify “.RNG.state” rather than “.RNG.seed” - see for example the output of parallel.seeds

Author(s)

Martyn Plummer