Subspace Identification

SubspaceIdentification is intended to detect changes in an estimated dynamical-system subspace rather than directly comparing signal shapes.

Currently unavailable

The constructor raises NotImplementedError; the offline implementation is unfinished.

Parameters

window_length : int

Subsequence length in samples.

n_windows : int, optional

Number of subsequences. The intended default is window_length.

lag : int, optional

Separation between compared regions.

rank : int, default 5

Estimated system-subspace rank.

scale : bool, default True

Intended min-max scaling before scoring.

method : {"offline"}, default "offline"

Intended implementation variant.

random_rank : int, optional

Randomized approximation dimension.

Minimal Example

from changepoynt.algorithms.si import SubspaceIdentification

try:
    detector = SubspaceIdentification(window_length=40)
except NotImplementedError:
    detector = None  # Subspace Identification is not yet available

Reference: Kawahara, Yairi, and Machida (2007).