R: Extract Fitted Values of a Dynamite Model

fitted.dynamitefit {dynamite}

R Documentation

Extract Fitted Values of a Dynamite Model

Description

Fitted values for a dynamitefit object, i.e., E(y_t | newdata, \theta) where \theta contains all the model parameters. See also predict.dynamitefit() for multi-step predictions.

Usage

## S3 method for class 'dynamitefit'
fitted(
  object,
  newdata = NULL,
  n_draws = NULL,
  thin = 1,
  expand = TRUE,
  df = TRUE,
  ...
)

Arguments

`object`	[`dynamitefit`] The model fit object.
`newdata`	[`data.frame`] Data used in predictions. If `NULL` (default), the data used in model estimation is used for predictions as well. There should be no new time points that were not present in the data that were used to fit the model, and no new group levels can be included.
`n_draws`	[`integer(1)`] Number of posterior samples to use, default is `NULL` which uses all samples without permuting (with chains concatenated). If `n_draws`is smaller than `ndraws(object)`, a random subset of `n_draws` posterior samples are used.
`thin`	[`integer(1)`] Use only every `thin` posterior sample. This can be beneficial with when the model object contains large number of samples. Default is `1` meaning that all samples are used.
`expand`	[`logical(1)`] If `TRUE` (the default), the output is a single `data.frame` containing the original `newdata` and the predicted values. Otherwise, a `list` is returned with two components, `simulated` and `observed`, where the first contains only the predicted values, and the second contains the original `newdata`. Setting `expand` to `FALSE` can help conserve memory because `newdata` is not replicated `n_draws` times in the output. This argument is ignored if `funs` are provided.
`df`	[`logical(1)`] If `TRUE` (default) the output consists of `data.frame` objects, and `data.table` objects otherwise.
`...`	Ignored.

Value

A data.frame containing the fitted values.

Examples

data.table::setDTthreads(1) # For CRAN
fitted(gaussian_example_fit, n_draws = 2L)

set.seed(1)
# Please update your rstan and StanHeaders installation before running
# on Windows
if (!identical(.Platform$OS.type, "windows")) {
  fit <- dynamite(
    dformula = obs(LakeHuron ~ 1, "gaussian") + lags(),
    data = data.frame(LakeHuron, time = seq_len(length(LakeHuron)), id = 1),
    time = "time",
    group = "id",
    chains = 1,
    refresh = 0
  )

  if (requireNamespace("dplyr") &&
    requireNamespace("tidyr") &&
    base::getRversion() >= "4.1.0") {

    # One-step ahead samples (fitted values) from the posterior
    # (first time point is fixed due to lag in the model):
    fitted(fit) |>
      dplyr::filter(time > 2) |>
      ggplot2::ggplot(ggplot2::aes(time, LakeHuron_fitted, group = .draw)) +
      ggplot2::geom_line(alpha = 0.5) +
      # observed values
      ggplot2::geom_line(ggplot2::aes(y = LakeHuron), colour = "tomato") +
      ggplot2::theme_bw()

    # Posterior predictive distribution given the first time point:
    predict(fit, type = "mean") |>
      dplyr::filter(time > 2) |>
      ggplot2::ggplot(ggplot2::aes(time, LakeHuron_mean, group = .draw)) +
      ggplot2::geom_line(alpha = 0.5) +
      # observed values
      ggplot2::geom_line(ggplot2::aes(y = LakeHuron), colour = "tomato") +
      ggplot2::theme_bw()
  }
 }