Cross-Validation Schemes

These functions represent different cross-validation schemes that can be used with origami. They should be used as options for the fold_fun argument to make_folds, which will call the requested function specify n, based on its arguments, and pass any remaining arguments (e.g. V or pvalidation) on.

folds_vfold(n, V = 10L)

folds_resubstitution(n)

folds_loo(n)

folds_montecarlo(n, V = 1000L, pvalidation = 0.2)

folds_bootstrap(n, V = 1000L)

folds_rolling_origin(n, first_window, validation_size, gap = 0L, batch = 1L)

folds_rolling_window(n, window_size, validation_size, gap = 0L, batch = 1L)

folds_rolling_origin_pooled(
  n,
  t,
  id = NULL,
  time = NULL,
  first_window,
  validation_size,
  gap = 0L,
  batch = 1L
)

folds_rolling_window_pooled(
  n,
  t,
  id = NULL,
  time = NULL,
  window_size,
  validation_size,
  gap = 0L,
  batch = 1L
)

folds_vfold_rolling_origin_pooled(
  n,
  t,
  id = NULL,
  time = NULL,
  V = 10L,
  first_window,
  validation_size,
  gap = 0L,
  batch = 1L
)

folds_vfold_rolling_window_pooled(
  n,
  t,
  id = NULL,
  time = NULL,
  V = 10L,
  window_size,
  validation_size,
  gap = 0L,
  batch = 1L
)

Arguments

n: An integer indicating the number of observations.
V: An integer indicating the number of folds.
pvalidation: A numeric indicating the proportion of observation to be placed in the validation fold.
first_window: An integer indicating the number of observations in the first training sample.
validation_size: An integer indicating the number of points in the validation samples; should be equal to the largest forecast horizon.
gap: An integer indicating the number of points not included in the training or validation samples. The default is zero.
batch: An integer indicating increases in the number of time points added to the training set in each iteration of cross-validation. Applicable for larger time-series. The default is one.
window_size: An integer indicating the number of observations in each training sample.
t: An integer indicating the total amount of time to consider per time-series sample.
id: An optional vector of unique identifiers corresponding to the time vector. These can be used to subset the time vector.
time: An optional vector of integers of time points observed for each subject in the sample.

Value

A list of Folds.

Arguments

Value

See also