run_model {EcoDiet}

Run the EcoDiet model

Description

This function runs the EcoDiet model using a Markov chain Monte Carlo approximation through the 'jagsUI' package to provide an approximated distribution for the variables of interest.

Depending on the nb_iter entered, this function may take hours, or even days to run. We advise you to first test whether your model is compiling properly with the by-default parameters, as this should take 1-2 min to run depending on your data size.

To save time, this function can solicit several cores (if available) to parallelize chains. Note that progress bars won't be displayed if chains are parallelized.

A warning message is printed if the model has not converged in the end (if the Gelman-Rubin diagnostic of at least one variable is > 1.1). For each run, the default 'jagsUI' package messages summarize the '.txt' file used for the definition of the BUGS model, the configuration of the model (iteration, adaptation, burnin, thin rate), the time required to run the model, and main statistics for the variables.

You need to have run the preprocess_data and the write_model functions before using this function, as their outputs are used as the inputs for run_model.

Usage

run_model(
  model_file,
  data,
  inits = NULL,
  run_param = "test",
  variables_to_save = c("eta", "PI"),
  parallelize = FALSE,
  DIC.out = TRUE
)

Arguments

model_file

The file containing the BUGS definition of the EcoDiet model output by the write_model function

data

The preprocessed data list output by the preprocess_data() function

inits

A list containing the initial values of the variables. By default the initialisation values are NULL, which means that the chain initial values are drawn from the prior distributions.

run_param

A object that can be a list of the parameters to configure the JAGS model or a string acting as a shortcut characterizing the overall length of the run requested (e.g. "short" or "long"). If run_param is provided as a list, the user should provide at least nb_iter, i.e. the number of iterations to run (the more iterations, the better are the chances that the model will converge; very small by default to test if the model compiles properly), and nb_burnin, i.e. the number of burn-in steps to run (so that the variable approximations are not too influenced by the first initial random values). nb_thin, the thinning rate, is by default defined by the function. The number of adaptation steps nb_adapt can be specified but is not required (see jasgUI documentation for more details). If set manually, it should be at least set at 1000.

variables_to_save

A vector of variable names defining the variables to output. The number has a big number of variables but by default we only save the variables of interest that are the trophic link probabilities eta and the diet proportions PI. Only these saved variables are used to compute the Gelman-Rubin statistics that indicate whether the model has converged or not.

parallelize

Indicates whether chains should be parallelized using several cores. Recommended in case of complex models.

DIC.out

Indicates whether the DIC (Deviance Information Criterion) should be reported.

Value

A MCMC output formatted as a jagsUI object.

See Also

preprocess_data to preprocess the data, and write_model to define the model.

Examples

realistic_biotracer_data <- read.csv(system.file("extdata", "realistic_biotracer_data.csv",
                                               package = "EcoDiet"))
realistic_stomach_data <- read.csv(system.file("extdata", "realistic_stomach_data.csv",
                                             package = "EcoDiet"))

data <- preprocess_data(biotracer_data = realistic_biotracer_data,
                        trophic_discrimination_factor = c(0.8, 3.4),
                        literature_configuration = FALSE,
                        stomach_data = realistic_stomach_data)
                        
write_model(literature_configuration = FALSE)

mcmc_output <- run_model("EcoDiet_model.txt", data, run_param="test")