ClaSP

ClaSP frames segmentation as a classification problem: subsequences from opposite sides of a candidate split should be distinguishable when the regimes differ.

Currently unavailable

CLASP.__init__() raises NotImplementedError because its former dependency combination is not supported by the package.

Parameters

n_segments : int or "learn", default "learn"

Number of segments, or automatic selection.

n_estimators : int, default 10

Number of ensemble estimators.

window_size : int or "suss", default "suss"

Subsequence length or automatic SuSS selection.

k_neighbours : int, default 3

Number of nearest neighbors used by classification.

distance, score, validation : str

Distance, classification score, and validation strategy forwarded to the intended ClaSP backend.

early_stopping : bool, default True

Enable early stopping of recursive segmentation.

threshold : float, default 1e-15

Significance threshold.

excl_radius : int, default 5

Exclusion radius around selected split points.

random_state : int, default 2357

Random seed.

Minimal Example

from changepoynt.algorithms.clasp import CLASP

try:
    detector = CLASP()
except NotImplementedError:
    detector = None  # ClaSP is unavailable in the current release

Reference: Schafer, Ermshaus, and Leser, ClaSP - Time Series Segmentation.