Preprocessing splitters timeseries

Development build. This page describes main, not a released package. The latest published Lodestar.Preprocessing is 0.2.0 — read its documentation.

Home › Preprocessing › Splitting

Splitters.TimeSeries

Cuts forward-chaining splits over time-ordered rows, as TimeSeriesSplit does.

public static IReadOnlyList<FoldSplit> TimeSeries(int sampleCount, int splitCount, int? testSize = null, int gap = 0, int? maxTrainSize = null)

Parameters — sampleCount is how many rows there are, oldest first. splitCount is scikit-learn's n_splits, at least two. testSize is the rows in each test block; null is sampleCount / (splitCount + 1), the reference's default. gap is how many rows to drop between a training block and its test block; a negative gap overlaps them, as the reference lets it. maxTrainSize keeps only the most recent rows of each training block; null or 0 keeps them all, since the reference reads max_train_size=0 as no cap.

Returns — one FoldSplit per split, oldest first, each index list ascending.

Exceptions — ArgumentOutOfRangeException when splitCount is below two, testSize below one, maxTrainSize negative, or splitCount + 1 above sampleCount. The negative cap is a refusal the reference does not make: it slices one into an empty training block without a word. ArgumentException when the test blocks and the gap leave the first split no training row, the reference's own refusal, or when the test blocks need more rows than there are — which only a negative gap lets past that check, and which the reference answers with an empty test fold.

Example — ten rows in three splits, then with a gap and a capped training window.

using Lodestar.Preprocessing;

IReadOnlyList<FoldSplit> splits = Splitters.TimeSeries(sampleCount: 10, splitCount: 3);
string firstTrain = string.Join(",", splits[0].TrainIndices);  // => 0,1,2,3
string firstTest = string.Join(",", splits[0].TestIndices);    // => 4,5
string lastTrain = string.Join(",", splits[2].TrainIndices);   // => 0,1,2,3,4,5,6,7

IReadOnlyList<FoldSplit> windowed = Splitters.TimeSeries(10, 3, testSize: 2, gap: 1, maxTrainSize: 3);
string window = string.Join(",", windowed[1].TrainIndices);    // => 2,3,4
string held = string.Join(",", windowed[1].TestIndices);       // => 6,7

Remarks — the test blocks are the last splitCount · testSize rows, cut in order, so any rows the division leaves over go to the first training block. Each split trains on everything before its block less gap rows, cut to the last maxTrainSize of them. With a gap of zero or more, no row a split trains on comes after a row it is tested on.

Applies to — net10.0, netstandard2.0.

See also — Splitters.KFold, FoldSplit.