sd_Bayes: Create Stan file for Bayesian inference

In readsdr: Translate Models from System Dynamics Software into 'R'

Create Stan file for Bayesian inference

Description

Create Stan file for Bayesian inference

Usage

sd_Bayes(
  filepath,
  meas_mdl,
  estimated_params,
  data_params = NULL,
  data_inits = NULL,
  const_list = NULL,
  forecast = FALSE
)

Arguments

filepath	A string that indicates a path to a file with extension .stmx or .xmile. Vensim files (.mdl) are not xmile files. They must be exported from Vensim with extension .xmile
meas_mdl	A list of strings. Each string corresponds to a sampling statement written in Stan language.
estimated_params	A list of lists. Each sublist describes each parameter that will be estimated in the inference stage. To construct this description, the user can avail of the function 'sd_prior'.
data_params	An optional string vector defining which model parameters will be configured through the Stan data block. That is, the user will provide fixed values for such parameters at every Stan run.
data_inits	An optional string vector defining which model parameters that only affect initial values (of stocks) will be configured through the Stan data block. That is, the user will provide fixed values for such parameters at every Stan run.
const_list	A list in which each element's name is the name of the constant to override and the element's value correspond to the new value.
forecast	An optional boolean that indicates whether the Stan file supports a forecast. If TRUE, the data block requires the user to supply an integer value for n_fcst. This variable corresponds to the number of periods that will be predicted.

Value

A string

Negative binomial measurement component

While this package aims to avoid making decisions for users whenever possible, I have taken the liberty to automate the transformation of phi (the concentration parameter) when using the Negative Binomial distribution (alternative parameterisation) as a measurement component. sd_Bayes() automatically creates an inverse phi parameter for computational efficiency, which will be subject to inference (instead of phi). Additionally, I have provided a default prior for this inv_phi but users can override it as needed.

Time

Simulation of the ordinary differential equation (ODE) model starts at time 0.

Examples

  filepath         <- system.file("models/", "SEIR.stmx", package = "readsdr")
  mm1              <- "y ~ neg_binomial_2(net_flow(C), phi)"
  meas_mdl         <- list(mm1)
  estimated_params <- list(
    sd_prior("par_beta", "lognormal", c(0, 1)),
    sd_prior("par_rho", "beta", c(2, 2)),
    sd_prior("I0", "lognormal", c(0, 1), "init"))
  sd_Bayes(filepath, meas_mdl, estimated_params)

readsdr documentation built on May 29, 2024, 2:45 a.m.