Create a DPMirt model specification

Constructs a complete specification for a Bayesian IRT model, including NIMBLE model code, constants, data, initial values, and monitor configuration. This is the first step in the step-by-step workflow.

Usage

dpmirt_spec(
  data,
  model = c("rasch", "2pl", "3pl"),
  prior = c("normal", "dpm"),
  parameterization = c("irt", "si"),
  identification = NULL,
  alpha_prior = NULL,
  base_measure = list(s2_mu = 2, nu1 = 2.01, nu2 = 1.01),
  item_priors = list(),
  M = 50L,
  data_format = c("auto", "matrix", "long"),
  ...
)

# S3 method for class 'dpmirt_spec'
print(x, ...)

Arguments

data

A matrix or data.frame of binary (0/1) responses with persons in rows and items in columns. Can also be a long-format data.frame with columns for person, item, and response.

model

Character. IRT model type: "rasch", "2pl", or "3pl".

prior

Character. Latent trait prior: "normal" (parametric) or "dpm" (Dirichlet Process Mixture).

parameterization

Character. "irt" for standard IRT or "si" for slope-intercept. Only relevant for 2PL/3PL.

identification

Character or NULL. Identification strategy: "constrained_item", "constrained_ability", or "unconstrained". If NULL, uses model-specific default (constrained_item for Rasch, unconstrained for 2PL/3PL). Some combinations are intentionally unavailable: DPM models reject "constrained_ability", and 3PL models currently reject "constrained_item".

alpha_prior

Alpha hyperprior specification. NULL for default Gamma(1,3), a numeric vector c(a, b) for Gamma(a, b), or a DPprior_fit object. Only used when prior = "dpm". Automatic alpha prior elicitation via mu_K is handled by dpmirt.

base_measure

List with DPM base measure hyperparameters: s2_mu, nu1, nu2. Defaults from Paganin et al. (2023).

item_priors

Reserved. Custom item-prior schemas are not currently implemented. Use NULL or list() to use DPMirt's fixed item priors.

M

Integer. Maximum number of clusters for CRP truncation. Only used when prior = "dpm".

data_format

Character. "auto", "matrix", or "long".

...

Additional arguments (currently unused).

x

A dpmirt_spec object.

Value

A dpmirt_spec S3 object containing:

code

A nimbleCode object.

constants

List of constants (N, I, M, etc.).

data

List with the response data.

inits

List of initial values.

monitors

Character vector of parameters to track.

monitors2

Character vector of parameters for thinned monitoring.

config

List of all model configuration options.

See also

dpmirt, dpmirt_compile

Other model fitting: dpmirt(), dpmirt_compile(), dpmirt_fit-methods, dpmirt_resume(), dpmirt_sample()

Examples

if (FALSE) { # \dontrun{
sim <- dpmirt_simulate(200, 20, model = "rasch", seed = 42)

# Rasch-Normal specification
spec <- dpmirt_spec(sim$response, model = "rasch", prior = "normal")
print(spec)

# Rasch-DPM with custom alpha prior
spec_dpm <- dpmirt_spec(sim$response, model = "rasch", prior = "dpm",
                        alpha_prior = c(1, 3))

# 2PL specification
sim2 <- dpmirt_simulate(300, 25, model = "2pl", seed = 42)
spec_2pl <- dpmirt_spec(sim2$response, model = "2pl", prior = "normal",
                        parameterization = "irt")
} # }