crossval_ts: Generic cross-validation function for time series

In thierrymoudiki/crossval: Generic cross-validation functions

Description

Generic cross-validation for univariate time series

Usage

crossval_ts(
  y,
  x = NULL,
  fit_func = crossval::fit_lm,
  predict_func = crossval::predict_lm,
  fcast_func = NULL,
  fit_params = NULL,
  initial_window = 5,
  horizon = 3,
  fixed_window = TRUE,
  type_forecast = c("mean", "quantiles"),
  level = c(80, 95),
  seed = 123,
  eval_metric = NULL,
  cl = NULL,
  errorhandling = c("stop", "remove", "pass"),
  packages = c("stats", "Rcpp"),
  verbose = FALSE,
  show_progress = TRUE,
  ...
)

Arguments

y	response time series; a vector
x	input covariates' matrix (optional)
fit_func	a function for fitting the model
predict_func	a function for predicting values from the model
fcast_func	time series forecasting function
fit_params	a list; additional (model-specific) parameters to be passed to fit_func
initial_window	an integer; the initial number of consecutive values in each training set sample
horizon	an integer; the number of consecutive values in test set sample
fixed_window	a boolean; if FALSE, all training samples start at 1
type_forecast	a string; "mean" for mean forecast, "lower", "upper" for lower and upper bounds respectively
level	a numeric vector; confidence levels for prediction intervals.
seed	random seed for reproducibility of results
eval_metric	a function measuring the test errors; if not provided: RMSE for regression and accuracy for classification
cl	an integer; the number of clusters for parallel execution
errorhandling	specifies how a task evalution error should be handled. If value is "stop", then execution will be stopped if an error occurs. If value is "remove", the result for that task will not be returned. If value is "pass", then the error object generated by task evaluation will be included with the rest of the results. The default value is "stop".
packages	character vector of packages that the tasks depend on
verbose	logical flag enabling verbose messages. This can be very useful for troubleshooting.
show_progress	show evolution of the algorithm
...	additional parameters

Examples

require(forecast)
data("AirPassengers")

# Example 1 -----

res <- crossval_ts(y=AirPassengers, initial_window = 10, fcast_func = thetaf)
print(colMeans(res))


# Example 2 -----

fcast_func <- function (y, h, ...)
{
      forecast::forecast(forecast::auto.arima(y, ...),
      h=h, ...)
}

res <- crossval_ts(y=AirPassengers, initial_window = 10, fcast_func = fcast_func)
print(colMeans(res))


# Example 3 -----

fcast_func <- function (y, h, ...)
{
      forecast::forecast(forecast::ets(y, ...),
      h=h, ...)
}

res <- crossval_ts(y=AirPassengers, initial_window = 10, fcast_func = fcast_func)
print(colMeans(res))


# Example 4 -----

xreg <- cbind(1, 1:length(AirPassengers))
res <- crossval_ts(y=AirPassengers, x=xreg, fit_func = crossval::fit_lm,
predict_func = crossval::predict_lm,
initial_window = 10,
horizon = 3,
fixed_window = TRUE)
print(colMeans(res))


# Example 5 -----

res <- crossval_ts(y=AirPassengers, x=xreg, fcast_func = thetaf,
initial_window = 10,
horizon = 3,
fixed_window = TRUE, type_forecast="quantiles")
print(colMeans(res))


#' # Example 6 -----

xreg <- cbind(1, 1:length(AirPassengers))
res <- crossval_ts(y=AirPassengers, x=xreg, fit_func = crossval::fit_lm,
predict_func = crossval::predict_lm,
initial_window = 10,
horizon = 3,
fixed_window = TRUE, type_forecast="quantiles")
print(colMeans(res))

thierrymoudiki/crossval documentation built on Aug. 17, 2020, 5:51 a.m.