add_prophet_forecast_model: Add prophet forecast model
In ing-bank/tsforecast: Time Series Forecasting Pipeline

Description Usage Arguments Value Examples

View source: R/INTRA_FORECAST_add_prophet_forecast_model.R

add_prophet_forecast_model is a function to add a single prophet forecast model to a (named) list of forecast models. The forecast model is created based on a parameter value to determine the flexibility of automatic changepoint selection, which is then used to forecast a specific number of periods ahead.

add_prophet_forecast_model(
  fc_models,
  ts_object_train,
  ts_object_valid = NULL,
  fc_name,
  model_type = c("univariate", "multivariate"),
  changepoint_prior_scale,
  periods_ahead = 1,
  periods_history = Inf,
  keep_fc_model_objects = FALSE,
  verbose = FALSE,
  log_message = ""
)

`fc_models`	A named list of forecast models, with for each forecast model a list with the model itself and a table with forecast values.
`ts_object_train`	A time series object, which contains only the training data.
`ts_object_valid`	A time series object, which contains the validation data. This is used for multivariate frameworks, thus it should have the forecasted/actual values of the external regressors as well.
`fc_name`	A character string specifying the name to be used for the new model that is added to the list of existing forecast models.
`model_type`	A character string indicating whether a univariate model (without external regressors) or a multivariate model (with external regressors) should be estimated.
`changepoint_prior_scale`	A positive numeric value modulating the flexibility of the automatic changepoint selection, where large values will allow many changepoints, while small values will allow few changepoints.
`periods_ahead`	A positive integer value indicating the number of periods to forecast ahead.
`periods_history`	A positive integer value indicating the number of historic datapoints to use for training, which is only relevant for specific forecast methods such as drift and mean.
`verbose`	Boolean, which is set to TRUE if status updates are valued, or set to FALSE if they are not.

A named list of forecast models, with for each forecast model a list with the model itself and a table with forecast values.

ts_object_train <- tstools::initialize_ts_forecast_data(
      data = dummy_gasprice,
      date_col = "year_month",
      col_of_interest = "gasprice",
      group_cols = c("state", "oil_company"),
      xreg_cols = c("spotprice", "gemprice")
   ) %>%
   dplyr::filter(grouping == "state = New York   &   oil_company = CompanyA") %>%
   tstools::transform_data_to_ts_object()
add_prophet_forecast_model(
   fc_models = list(),
   ts_object_train = ts_object_train,
   fc_name = "fc_prophet_050cps",
   changepoint_prior_scale = 0.050,
   periods_ahead = 12,
   verbose = T
)