add_recursive_ml_forecast_model: Add randomForest forecast model

In ing-bank/tsforecast: Time Series Forecasting Pipeline

Description

add_recursive_ml_forecast_model is a function to add a recursive machine learning model from the caret package. It tunes the parameters with default settings in a cross validated way.

Usage

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

Arguments

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.
caret_model_tag	A character string specifying the tag of the model to be used from the caret meta package.
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.
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.
keep_fc_model_objects	Boolean, which is set to TRUE in order to keep original fc_model objects in the main_forecasting_table after running the forecast. This is needed for scenario analysis in multivariate forecasting. However, it may lead to memory issues, as the main_forecasting_table increases in size.
verbose	Boolean, indicating whether or not the function should print messages while running.

Value

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

Examples

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") %>%
   dplyr::slice(1:189) %>%
   tstools::transform_data_to_ts_object()
ts_object_valid <- 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") %>%
   dplyr::slice(190:191) %>%
   tstools::transform_data_to_ts_object()
add_recursive_ml_forecast_model(
   fc_models = list(),
   caret_model_tag = "svmRadialSigma",
   ts_object_train = ts_object_train,
   ts_object_valid = ts_object_valid,
   fc_name = "Example_svm",
   model_type = "multivariate",
   periods_ahead = 2,
   verbose = TRUE
 )

ing-bank/tsforecast documentation built on Sept. 18, 2020, 9:40 a.m.