compute_window_size

dtaianomaly.windowing.compute_window_size(X: ndarray, window_size: int | str, lower_bound: int = 10, relative_lower_bound: float = 0.0, upper_bound: int = 1000, relative_upper_bound: float = 1.0, threshold: float = 0.89, default_window_size: int = None) → int[source]

Compute the window size of the given time series [11].

Given a time series, automatically compute the window size. This can be done either automatically, using a number of procedures for estimating the window size, or based on a user-defined value.

Parameters:

Xarray-like of shape (n_samples, n_attributes)

Input time series.

window_sizeint or str

The method by which a window size should be computed. Valid options are:

int: Simply return the given window size.
'fft': Compute the window size by selecting the dominant Fourier frequency.
'acf': Compute the window size as the leg with the highest autocorrelation.
'mwf': Computes the window size using the Multi-Window-Finder method [18].
'suss': Computes the window size using the Summary Statistics Subsequence method [10].

lower_boundint, default=10

The lower bound on the automatically computed window size. Only used if window_size equals 'fft', 'acf', 'mwf' or 'suss'.

relative_lower_boundfloat, default=0.0

The lower bound on the automatically computed window size, relative to the length of the given time series. Only used if window_size equals 'fft', 'acf', 'mwf' or 'suss'.

upper_boundint, default=1000

The lower bound on the automatically computed window size. Only used if window_size equals 'fft', 'acf', or 'mwf'.

relative_upper_boundfloat, default=1.0

The upper bound on the automatically computed window size, relative to the length of the given time series. Only used if window_size equals 'fft', 'acf', or 'mwf'.

thresholdfloat, default=0.89

The threshold for selecting the optimal window size using 'suss'.

default_window_sizeint, default=None

The default window size, in case an invalid automatic window size was computed. By default, the value is set to None, which means that an error is thrown.

Returns:

int: The computed window size.

Warning

Automatically computing the windwow size only works for univariate time series!

See also

dominant_fourier_frequency: Compute the window size based on the fft.
highest_autocorrelation: Compute the window size based on the acf.
multi_window_finder: Compute the window size with mwf.
summary_statistics_subsequences: Compute the window size with suss.

Examples

>>> from dtaianomaly.data import demonstration_time_series
>>> from dtaianomaly.windowing import compute_window_size
>>> X, _ = demonstration_time_series()
>>> compute_window_size(X, 64)
64
>>> compute_window_size(X, 'fft')
107
>>> compute_window_size(X, 'acf')
112
>>> compute_window_size(X, 'mwf')
112
>>> compute_window_size(X, 'suss')
62