Struct StratifiedGroupKFold 

pub struct StratifiedGroupKFold<L, G> { /* private fields */ }

K-fold that tries to satisfy two constraints at once: keep class proportions balanced across folds and never let a group straddle the train/test boundary.

Approximation

Perfectly satisfying both constraints simultaneously is not always possible — a group is indivisible, so its whole class makeup lands in one fold. This implementation therefore uses the same documented greedy heuristic as scikit-learn’s StratifiedGroupKFold, and does not claim exactness:

1. Count each class within each group.
2. Visit groups in order of decreasing spread (standard deviation) of their per-class counts, so the “lumpiest” groups are placed first while there is still freedom to balance around them.
3. Place each group in whichever fold keeps the per-fold class distribution closest to uniform (minimising the mean, over classes, of the standard deviation of each class’s per-fold share); ties break toward the smaller fold.

Group integrity is exact (groups are never split); only class balance is approximate. n_splits must not exceed the number of distinct groups.

use ndarray::array;
use model_selection_rs::splitters::{CvSplitter, StratifiedGroupKFold};

let y      = array![0, 0, 1, 1, 0, 1, 0, 1];
let groups = array![1, 1, 2, 2, 3, 3, 4, 4];
let sgkf = StratifiedGroupKFold::new(2, &y, &groups).unwrap();
let splits = sgkf.split(y.len()).unwrap();
assert_eq!(splits.len(), 2);

Implementations

impl<L: Eq + Hash + Clone, G: Eq + Hash + Clone> StratifiedGroupKFold<L, G>

pub fn new(n_splits: usize, y: &Array1<L>, groups: &Array1<G>) -> Result<Self>

Create a StratifiedGroupKFold over class labels y and groups.

Errors

Returns ModelSelectionError::InvalidSplitCount if n_splits < 2, or ModelSelectionError::ShapeMismatch if y and groups differ in length.

Trait Implementations

impl<L: Clone, G: Clone> Clone for StratifiedGroupKFold<L, G>

fn clone(&self) -> StratifiedGroupKFold<L, G>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<L: Eq + Hash + Clone, G: Eq + Hash + Clone> CvSplitter for StratifiedGroupKFold<L, G>

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<L: Debug, G: Debug> Debug for StratifiedGroupKFold<L, G>

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

Formats the value using the given formatter.