cv_arima: Cross validated prediction and evaluation with ARIMA...

In thfuchs/fcf: Time Series Forecasting using Recurrent Neural Networks by Keras

Description

Time-series cross-validation wrapper for auto.arima

Usage

cv_arima(
  data,
  cv_setting,
  col_id = NULL,
  col_date = "index",
  col_value = "value",
  transform = NULL,
  frequency = 4,
  h = NULL
)

Arguments

data	data.table object
cv_setting	cross validation settings. Named list requiring periods_train, periods_val periods_test and skip_span. See section "Cross validation settings" for details.
col_id	Optional ID column in data, default to "ticker"
col_date	Date column in data, default to "index"
col_value	Value column in data, default to "value"
transform	Transform data before estimation? One of NULL (default) and "normalize"
frequency	time series frequency, e.g. 4 for quarters and 12 for months
h	NULL if forecast horizon equals cv_setting$n_test, else named list of forecast horizons for accuracy measures

Value

list of type (model), h (forecast horizon, if specified), mape, smape, mase, smis and acd

Cross validation settings

Using rolling_origin to split the time series. Requiring:

• periods_train: Length of training set per split

• periods_val: Length of validation set per split

• periods_test: Length of test/hold-out set per split

• skip_span: Gaps between overlapping splits to reduce computational intensity and recundancy between data splits.
  
  

Note: periods_val only relevant for deep learning models. cv_baselines and cv_arima use sum of periods_train and periods_val for training and only periods_test as hold-out test set (no learning and feedback through validation by traditional statistical models)

Examples

## Not run: 
cv_setting <- list(
  periods_train = 90,
  periods_val = 10,
  periods_test = 10,
  skip_span = 5
)

fc_01 <- cv_arima(
  data = tsRNN::DT_apple,
  cv_setting = cv_setting
)
fc_01

# Multiple forecast horizons
fc_02 <- cv_arima(
  data = tsRNN::DT_apple,
  cv_setting = cv_setting,
  h = list(short = 1:2, long = 3:6)
)
fc_02

## End(Not run)

thfuchs/fcf documentation built on April 18, 2021, 1:43 p.m.