Split Data into Temporal Train and Test Sets

Performs a time-ordered (non-random) train/test split so that the training set always precedes the test set chronologically – the correct approach for time-series and any dataset where row order encodes time.

Usage

split_data(data, train_ratio = 0.8, verbose = TRUE)

Arguments

data

A data.frame (rows assumed ordered by time).

train_ratio

Numeric in (0, 1). Fraction of rows for training. Default 0.8 (80 / 20 split).

verbose

Logical (default TRUE).

Value

An invisible list with elements:

train

Training data.frame.

test

Test data.frame.

train_idx

Integer vector of training row indices.

test_idx

Integer vector of test row indices.

ratio

Named numeric c(train = ..., test = ...).

Examples

data(airquality)
clean <- airquality[complete.cases(airquality), ]
sp    <- split_data(clean, train_ratio = 0.7)
#> 
#> ============================================================
#>   TRAIN / TEST SPLIT  (temporal)
#> ============================================================
#> 
#>   Total rows  : 111
#>   Train ratio : 69%  (77 rows)
#>   Test  ratio : 31%  (34 rows)
#> 
nrow(sp$train); nrow(sp$test)
#> [1] 77
#> [1] 34