transform

class SlidingWindowAggregator(metric_funcs, window_size, window_slices=None)

Bases: BaseEstimator, TransformerMixin

Sliding window aggregator for time series data.

Parameters:

  • metric_funcs (list[ArrayToArray]) – List of functions to apply to each window.

  • window_size (int | Iterable[int]) – Size of the sliding window.

  • window_slices (list[slice] | None, default: None) – List of slices to use for each window.

See also

get_window_slices() to obtain window size(s) and corresponding slices.

fit(X, y=None)

This method is required by the sklearn API and does not perform any actual fitting.

Parameters:
Return type:

Self

get_feature_names_out(input_features=None)

Get output feature names for transformation.

Parameters:

input_features (Iterable[str] | None, default: None) – Input feature names.

Return type:

ndarray

Returns:

Output feature names as a 1D array.

transform(X)

Transform the input data by applying the metric functions to sliding windows. The transformed data is returned as a 2D array, flattened along all axes except the first.

Parameters:

X (ndarray | DataFrame) – Input data to transform.

Return type:

ndarray

Returns:

Transformed data as a 2D array.

get_window_slices(num_windows_per_scale, *, time_scales=None, durations=None, time_scale_quantiles=None)

Find consecutive window slices for time scales, either explicitly specified or derived from durations and quantiles.

Parameters:

  • num_windows_per_scale (int) – Number of windows per time scale.

  • time_scales (Iterable[int] | None, default: None) – Explicit time scales.

  • durations (ndarray | None, default: None) – Durations to calculate time scales from.

  • time_scale_quantiles (Iterable[float] | None, default: None) – Quantiles of the durations to derive time scales.

Return type:

tuple[list[int], list[slice]]

Returns:

A tuple containing the (adjusted) time scales and the corresponding window slices.

Raises:

  • ValueError – If neither time_scales nor durations and time_scale_quantiles are specified.

  • ValueError – If time_scale_quantiles are specified but durations are not.