Skip to content

Kernelized mean change (CostCosine)#


Given a positive semi-definite kernel \(k(\cdot, \cdot) : \mathbb{R}^d\times \mathbb{R}^d \mapsto \mathbb{R}\) and its associated feature map \(\Phi:\mathbb{R}^d \mapsto \mathcal{H}\) (where \(\mathcal{H}\) is an appropriate Hilbert space), this cost function detects changes in the mean of the embedded signal \(\{\Phi(y_t)\}_t\) [Arlot2019]. Formally, for a signal \(\{y_t\}_t\) on an interval \(I\),

\[ c(y_{a..b}) = \sum_{t=a}^{b-1} \| \Phi(y_t) - \bar{\mu} \|_{\mathcal{H}}^2 \]

where \(\bar{\mu}_{a..b}\) is the empirical mean of the embedded sub-signal \(\{\Phi(y_t)\}_{a\leq t < b-1}\). Here the kernel is the cosine similarity:

\[ k(x, y) = \frac{\langle x\mid y\rangle}{\|x\|\|y\|} \]

where \(\langle \cdot\mid\cdot \rangle\) and \(\| \cdot \|\) are the Euclidean scalar product and norm respectively. In other words, it is equal to the L2-normalized dot product of vectors. This cost function has been used for music segmentation tasks [Cooper2002] and topic segmentation of text [Hearst1994].


Start with the usual imports and create a signal.

import numpy as np
import matplotlib.pylab as plt
import ruptures as rpt

# creation of data
n, dim = 500, 3  # number of samples, dimension
n_bkps, sigma = 3, 5  # number of change points, noise standart deviation
signal, bkps = rpt.pw_constant(n, dim, n_bkps, noise_std=sigma)

Then create a CostCosine instance and print the cost of the sub-signal signal[50:150].

c = rpt.costs.CostCosine().fit(signal)
print(c.error(50, 150))

You can also compute the sum of costs for a given list of change points.

print(c.sum_of_costs([10, 100, 200, 250, n]))

In order to use this cost class in a change point detection algorithm (inheriting from BaseEstimator), either pass a CostCosine instance (through the argument custom_cost) or set model="cosine".

c = rpt.costs.CostCosine()
algo = rpt.Dynp(custom_cost=c)
# is equivalent to
algo = rpt.Dynp(model="cosine")


[Hearst1994] Hearst, M. A. (1994). Multi-paragraph segmentation of expository text. In Proceedings of the Annual Meeting of the Association for Computational Linguistics (pp. 9–16). Las Cruces, New Mexico, USA.

[Cooper2002] Cooper, M., & Foote, J. (2002). Automatic music summarization via similarity analysis. In Proceedings of the International Conference on Music Information Retrieval (ISMIR) (pp. 81–85). Paris, France.

[Arlot2019] Arlot, S., Celisse, A., & Harchaoui, Z. (2019). A kernel multiple change-point algorithm via model selection. Journal of Machine Learning Research, 20(162), 1–56.