Prediction Window Structure

This module defines an abstract base class PredictionWindowStructure, which is used as a template for creating the prediction window mask used in Long-Range AutoRegression.

The following classes derive from this base class: FullWindowStructure, SubspaceWindowStructure, DistanceWindowStructure, and ExplicitWindowStructure. Each of these classes creates a prediction window mask based on some structure.

For more details on using these classes, see Advanced Features.

These classes are documented below:

Abstract Base Class

class aomodel.prediction_window_structure.PredictionWindowStructure

Bases: ABC

Abstract base class for all prediction window structures.

abstractmethod PredictionWindowStructure.get_mask()

Create the prediction window mask, a boolean numpy 2-D array of shape (vector_dimensionality, vector_dimensionality).

Parameters:

vector_dimensionality (int) – the dimensionality of the vector space to use.

Returns:

mask (ndarray) – the prediction window mask.

Inherited Classes

The inherited classes each have the get_mask method derived from PredictionWindowStructure. In each inherited class, the method takes in the same argument vector_dimensionality and returns the same array mask.

class aomodel.prediction_window_structure.FullWindowStructure(predicted_components=None)

Bases: PredictionWindowStructure

Include all possible prediction window indices for each predicted component.

Parameters:

predicted_components (ndarray, optional) – [Default=None] numpy 1-D integer array containing the indices of the data vector’s components for which the model will compute prediction weights.

• If set to None, all vector components are predicted.

class aomodel.prediction_window_structure.SubspaceWindowStructure(subspace_dimension)

Bases: PredictionWindowStructure

Restrict prediction to a sub-set of the first subspace_dimension components.

Parameters:

subspace_dimension (int) – the subspace dimension to restrict the prediction window to.

class aomodel.prediction_window_structure.DistanceWindowStructure(distance, predicted_components=None)

Bases: PredictionWindowStructure

For each predicted component, structure the prediction window to include only the “neighboring” vector components with a certain distance.

Parameters:

• distance (int) – the distance to restrict the prediction window to.

• predicted_components (ndarray, optional) – [Default=None] numpy 1-D integer array containing the indices of the data vector’s components for which the model will compute prediction weights.

  • If set to None, all vector components are predicted.

class aomodel.prediction_window_structure.ExplicitWindowStructure(mask)

Bases: PredictionWindowStructure

Creates the prediction window structure from an explicit prediction window mask.

Parameters:

mask (ndarray) – the prediction window mask.

---

Disclaimer: Approved for public release; distribution is unlimited. Public Affairs release approval # AFRL-2026-1309.