R/make_split.R
make_split.Rd
The function creates the split indices for train and test samples
(i.e. partitioning into time slices) for time series cross-validation. The
user can choose between stretch
and slide
. The first is an
expanding window approach, while the latter is a fixed window approach.
The user can define the window sizes for training and testing via
n_init
and n_ahead
, as well as the step size for increments
via n_step
.
make_split(
main_frame,
context,
type,
value,
n_ahead,
n_skip = 0,
n_lag = 0,
mode = "slide",
exceed = TRUE
)
A tibble
containing the time series data.
A named list
with the identifiers for seried_id
, value_id
and index_id
.
The type for the initial split. Possible values are "first"
, "last"
, "prob"
.
Numeric value specifying the split.
The forecast horizon (n-steps-ahead, must be positive).
The number of periods to skip between windows (must be zero or positive integer).
A value to include a lag between the training and testing set. This is useful if lagged predictors will be used during training and testing.
Character value. Define the setup of the training window for time series cross validation. stretch
is equivalent to an expanding window approach and slide
is a fixed window approach.
Logical value. If TRUE
, out-of-sample splits exceeding the sample size are created.
A list
containing the indices for train and test as integer vectors.