MSST

MSST extends SST to multivariate signals by forming a block-Hankel matrix from all channels. Use it when changes are expressed by joint temporal behavior rather than one channel alone.

Use the tuning guide for a parameter-by-parameter workflow and runtime advice.

Parameters

window_length : int

Subsequence length in samples.

n_windows : int, optional

Number of subsequences in each block-Hankel matrix. Defaults to window_length.

lag : int, optional

Separation between compared matrices. Defaults to max(n_windows // 3, 1).

rank : int, default 5

Number of dominant joint directions.

scale : bool, default True

Min-max scale every channel independently to [1, 2].

method : str, default "ika"

One of "ika", "rsvd", "weighted", or "symmetric".

lanczos_rank : int, optional

Krylov approximation size for IKA.

random_rank : int, optional

Sampled dimension for randomized methods.

feedback_noise_level : float, default 1e-3

Noise added to the recycled future direction.

scoring_step : int, default 1

Distance between evaluated positions.

use_fast_hankel : bool, default False

Select fast block-Hankel products. The current implementation uses the fast representation internally for scoring regardless of this value.

Methods

transform(time_series)

time_series : numpy.ndarray, shape (n_samples, n_channels)

Multivariate signal with samples on axis 0.

Returns

score : numpy.ndarray, shape (n_samples,)

Joint structural-change score.

Minimal Example

import numpy as np
from changepoynt.algorithms.msst import MSST

t = np.linspace(0, 12 * np.pi, 400)
signal = np.column_stack((np.sin(t), np.cos(t)))
signal[200:, 1] = np.cos(2 * t[200:])
score = MSST(window_length=40, method="rsvd").transform(signal)

References: Ide and Inoue (2005), Ide and Tsuda (2007).