R/helpers.R

In davharris/mistnet2: Neural Networks with Latent Random Variables

# Functions used internally that don't need to be made visible
# to the user


# Add each element of `vector` to the corresponding column, instead
# of adding the whole vector to each column as the recycling rule would do
`%plus%` = function(matrix, vector){
  rcpp_add_biases(matrix, vector)
}

# several times faster than R's built-in `ifelse` function; useful for
# activation functions that are called many times
ifelse_matrix = function(test, yes, no){

  # Without this assertion, any nonzero entries would be coerced to TRUE
  if (!is.logical(test)) {
    stop("test is not logical")
  }

  ifelse_matrix_cpp(test, yes, no)
}

build_par_list = function(par, par_list){
  if (is.numeric(par)) {
    # Relist the parameters, using par_list
    parameters = relist(flesh = par, skeleton = purrr::compact(par_list))
  } else {
    if (is.list(par)) {
      # Just use the list
      parameters = par
    } else {
      stop("par is not a list or numeric vector")
    }
  }

  parameters
}