R/specificationMixture.R

In luisdamiano/BayesHMM: Full Bayesian Inference for Hidden Markov Models

#' Specify a mixture model.
#' @keywords internal
#' @inherit specify
#' @family models
mixture <- function(K, R, observation = NULL, initial = NULL,
                    transition = NULL, name = "") {
  x <- specify(K, R, observation, initial, transition, name)
  class(x) <- append(class(x), "MixtureSpecification", 0)
  x
}

# Undocumented internal methods -------------------------------------------

block_data.MixtureSpecification <- function(spec) {
  "
  int<lower = 1> K; // number of hidden states
  int<lower = 1> R; // dimension of the observation vector
  "
}

block_functions.MixtureSpecification <- function(spec) {
  ""
}

block_parameters.MixtureSpecification <- function(spec) {
  "
  simplex[K] pi;                   // mixing proportion
  "
}

block_tparameters.MixtureSpecification <- function(spec) {
  "
  vector[K] logpi;
  logpi = log(pi);
  "
}

block_generated.MixtureSpecification <- function(spec) {
  ""
}

chunk_calculate_target.MixtureSpecification <- function(spec) {
  ""
}

chunk_increase_target.MixtureSpecification <- function(spec) {
  "
  for (t in 1:T) {
    vector[K] accumulator = logpi;
    for (k in 1:K) {
      accumulator[k] += loglike[k, t];
    }
    target += log_sum_exp(accumulator);
  }
  "
}

chunk_zpredictive.MixtureSpecification <- function(spec) {
  "
  for(t in 1:T) {
    zpred[t] = categorical_rng(pi);
  }
  "
}