Struct TimeSeriesSplit 

pub struct TimeSeriesSplit { /* private fields */ }

Time-series cross-validation (rolling-origin evaluation).

Successive splits grow their training window into the past while the test set is always the next chronological chunk. Order is never shuffled: for every split, every training index is strictly earlier than every test index (with an optional gap between them).

Input convention: samples are assumed to be in chronological row order. This splitter works purely on positions, not on explicit timestamps — sort your data by time before using it. (Positional order keeps the API simple and matches scikit-learn’s TimeSeriesSplit; a timestamp-aware variant is intentionally out of scope.)

Configuration mirrors scikit-learn:

• n_splits — number of train/test splits.
• max_train_size — cap the training window to a fixed size (a rolling window) instead of an ever-expanding one.
• gap — drop this many samples between the end of train and the start of test, modelling a real-world delay before an outcome/label is known.
• test_size — samples per test set (defaults to n_samples / (n_splits + 1)).

use model_selection_rs::splitters::{CvSplitter, TimeSeriesSplit};

let tss = TimeSeriesSplit::new(3).unwrap();
for (train, test) in tss.split(12).unwrap() {
    // train is always entirely before test
    assert!(train.iter().max() < test.iter().min());
}

Implementations

impl TimeSeriesSplit

pub fn new(n_splits: usize) -> Result<Self>

Create a TimeSeriesSplit with n_splits expanding-window splits.

Errors

Returns ModelSelectionError::InvalidSplitCount if n_splits < 1.

pub fn with_max_train_size(self, max_train_size: usize) -> Self

Cap the training window to max_train_size samples (rolling window).

pub fn with_gap(self, gap: usize) -> Self

Insert a gap of dropped samples between train and test.

pub fn with_test_size(self, test_size: usize) -> Self

Fix the number of samples in each test set.

Trait Implementations

impl Clone for TimeSeriesSplit

fn clone(&self) -> TimeSeriesSplit

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl CvSplitter for TimeSeriesSplit

fn split(&self, n_samples: usize) -> Result<Vec<(Vec<usize>, Vec<usize>)>>

Produce every (train_indices, test_indices) pair for n_samples rows.

fn n_splits(&self) -> usize

Number of splits this strategy yields.

impl Debug for TimeSeriesSplit

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.