linear_chi2_penalty#

linear_chi2_penalty(n_samples: int, n_features: int, n_params_per_feature: int = 1) → ndarray[source][source]#

Create a linear chi-square penalty.

The penalty is a piece of the default penalty for the MVCAPA algorithm. It is described as “penalty regime 2” in the MVCAPA article [1], suitable for detecting sparse anomalies in the data. Sparse anomalies only affect a few features.

Parameters:

n_samplesint: Sample size.
n_featuresint: Number of features/columns in the data being analysed.
n_params_per_featureint, default=1: Number of model parameters per feature and segment.

Returns:

np.ndarray of shape (n_features,): The non-decreasing linear chi-square penalty values. Element i is the penalty for i+1 features being affected by a change or anomaly.

References

[1]

Fisch, A. T., Eckley, I. A., & Fearnhead, P. (2022). Subset multivariate segment and point anomaly detection. Journal of Computational and Graphical Statistics, 31(2), 574-585.

Examples

>>> linear_chi2_penalty(100, 3)
array([...])