Preprocessing 0.2.0 splitters

Lodestar.Preprocessing 0.2.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

Home › Splitting

Splitters

Cross-validation and train/test splits over row indices.

public static class Splitters

Example — three stratified folds over twelve rows in three classes.

using Lodestar.Preprocessing;

// Six rows of class 0, three of class 1, three of class 2.
int[] labels = [0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2];

IReadOnlyList<FoldSplit> folds = Splitters.StratifiedKFold(labels, foldCount: 3);

string held = string.Join(",", folds[0].TestIndices);  // => 0,1,6,9
string fit = string.Join(",", folds[0].TrainIndices);  // => 2,3,4,5,7,8,10,11

Remarks — every fold above holds two rows of class 0 and one of each other class, which is what stratifying buys: Splitters.KFold on the same twelve rows would hand fold 0 four rows of class 0 and nothing else.

The unshuffled splitters are scikit-learn's, fold for fold and index for index, replayed from tests/oracles/preprocessing_splitters.json. The shuffled ones take the permutation as an argument rather than a seed — the splitting index has why, and which shuffled splits a permutation reproduces.

Applies to — net10.0, netstandard2.0.

See also — FoldSplit, TrainTestSplit, the splitting index, the Python equivalence table.

Members

Member	What it does
Splitters.KFold	Cuts the rows into contiguous folds.
Splitters.StratifiedKFold	Cuts folds that keep each class's share.
Splitters.TrainTest	Holds out the last rows as a test set.