pysad.transform.preprocessing.SeasonalTrendDecomposer

class pysad.transform.preprocessing.SeasonalTrendDecomposer(season_length, trend_window=None)[source]

Streaming seasonal and trend decomposition preprocessor.

The transformer subtracts a rolling mean trend estimate and a running average seasonal estimate for each position in the season. This is an online approximation intended for preprocessing before univariate anomaly scoring; it is not an STL implementation and does not reproduce the modified STL residual used by Seasonal ESD in [BHVK17].

Parameters:

season_length (int) – Number of observations in one seasonal period.
trend_window (int) – Number of recent observations used to estimate the trend. If None, defaults to season_length.

Methods

`__init__`(season_length[, trend_window])
`fit`(X)	Shortcut method that iteratively applies fit_partial to all instances in order.
`fit_partial`(X)	Fits the next timestep's values to train the decomposer.
`fit_transform`(X)	Shortcut method that iteratively applies fit_transform_partial to all instances in order.
`fit_transform_partial`(X)	Fits and transforms the next timestep.
`transform`(X)	Shortcut method that iteratively applies transform_partial to all instances in order.
`transform_partial`(X)	Transforms the next timestep by removing trend and seasonality.

fit(X)

Shortcut method that iteratively applies fit_partial to all instances in order.

Parameters:: X (np.float64 array of shape (num_instances, num_features)) – Input feature vectors.
Returns:: The fitted transformer
Return type:: object

fit_partial(X)[source]

Fits the next timestep’s values to train the decomposer.

Parameters:: X (np.float64 array of shape (num_features,)) – Input feature vector.
Returns:: self.
Return type:: object

fit_transform(X)

Shortcut method that iteratively applies fit_transform_partial to all instances in order.

Parameters:: X (np.float64 array of shape (num_instances, num_components)) – Input feature vectors.
Returns:: Projected feature vectors.
Return type:: np.float64 array of shape (num_instances, num_components)

fit_transform_partial(X)[source]

Fits and transforms the next timestep.

Parameters:: X (np.float64 array of shape (num_features,)) – Input feature vector.
Returns:: Residual feature vector.
Return type:: residual_X (np.float64 array of shape (num_features,))

transform(X)

Shortcut method that iteratively applies transform_partial to all instances in order.

Parameters:: X (np.float64 array of shape (num_instances, num_features)) – Input feature vectors.
Returns:: Projected feature vectors.
Return type:: np.float64 array of shape (num_instances, num_components)

transform_partial(X)[source]

Transforms the next timestep by removing trend and seasonality.

Parameters:: X (np.float64 array of shape (num_features,)) – Input feature vector.
Returns:: Residual feature vector.
Return type:: residual_X (np.float64 array of shape (num_features,))