Temporal K Fold Data Splitter

This module has two classes for splitting time series data temporally - where train/test or fold splits are created within each of the time series’ in the time series data. This splitting approach is for evaluating how well the algorithm performs on segments drawn from the same time series but excluded from the training set. The performance from this splitting approach should be similar to performance on the training data so long as the data in each series is relatively uniform.

Note that splitting along the temporal axis violates the assumption of independence between train and test samples, as successive samples in a sequence or series are not iid. However, temporal splitting is still useful in certain cases such as for the analysis of a single sequence / series.

class seglearn.split.TemporalKFold(n_splits=3)

K-fold iterator variant for temporal splitting of time series data

The time series’ are divided in time with no overlap, and are balanced.

By splitting the time series’, the number of samples in the data set is changed and so new arrays for the data and target are returned by the split function in addition to the iterator.

Parameters

n_splitsint > 1

number of folds

Examples

>>> from seglearn.split import TemporalKFold
>>> from seglearn.datasets import load_watch
>>> data = load_watch()
>>> splitter = TemporalKFold(n_splits=4)
>>> X, y, cv = splitter.split(data['X'], data['y'])

Methods

split(self, X, y)

Splits time series data and target arrays, and generates splitting indices

split(self, X, y)

Splits time series data and target arrays, and generates splitting indices

Parameters

Xarray-like, shape [n_series, …]

Time series data and (optionally) contextual data

yarray-like shape [n_series, ]

target vector

Returns

Xarray-like, shape [n_series * n_splits, ]

Split time series data and contextual data

yarray-like, shape [n_series * n_splits]

Split target data

cvlist, shape [2, n_splits]

Splitting indices

seglearn.split.temporal_split(X, y, test_size=0.25)

Split time series or sequence data along the time axis. Test data is drawn from the end of each series / sequence

Parameters

Xarray-like, shape [n_series, …]

Time series data and (optionally) contextual data

yarray-like shape [n_series, ]

target vector

test_sizefloat

between 0 and 1, amount to allocate to test

Returns

X_trainarray-like, shape [n_series, ]

X_testarray-like, shape [n_series, ]

y_trainarray-like, shape [n_series, ]

y_testarray-like, shape [n_series, ]