GreedyGaussianSegmentation
Greedy Gaussian Segmentation Estimator.
Implementation based on [1].
source code adapted based on: https://github.com/cvxgrp/GGS
paper available at: https://stanford.edu/~boyd/papers/pdf/ggs.pdf
The method approximates solutions for the problem of breaking a multivariate time series into segments, where the data in each segment could be modeled as independent samples from a multivariate Gaussian distribution. It uses a dynamic programming search algorithm with a heuristic that allows finding approximate solutions in linear time with respect to the data length and always yields a locally optimal choice.
Greedy Gaussian Segmentation (GGS) fits a segmented gaussian model (SGM) to the data by computing the approximate solution to the combinatorial problem of finding the approximate covariance-regularized maximum log-likelihood for fixed number of change points and a reagularization strength. It follows an interactive procedure where a new breakpoint is added and then adjusting all breakpoints to (approximately) maximize the objective. It is similar to the top-down search used in other change point detection problems.
Schnellstart
from sktime.detection.ggs import GreedyGaussianSegmentation
estimator = GreedyGaussianSegmentation(k_max: int=10, lamb: float=1.0, max_shuffles: int=250, verbose: bool=False, random_state: int=None)Parameter(5)
- k_maxint, default=10
- Maximum number of change points to find. The number of segments is thus k+1.
- lambfloat, default=1.0
- Regularization parameter lambda (>= 0), which controls the amount of (inverse) covariance regularization. A higher lambda favors simpler models.
- max_shufflesint, default=250
- Maximum number of shuffles
- verbosebool, default=False
If
True, verbose output is enabled.- random_stateint or np.random.RandomState, default=None
Either random seed or an instance of
np.random.RandomState
Referenzen
Hallac, D., Nystrup, P. & Boyd, S., “Greedy Gaussian segmentation of multivariate time series.”, Adv Data Anal Classif 13, 727-751 (2019). https://doi.org/10.1007/s11634-018-0335-0