Penalties#
Penalties govern the trade-off between the number of change points and the fit of the
model. They are used by all current detectors in skchange
. Utility functions are
provided for helping to create commonly used penalties.
Constant penalties#
The penalty for each additional change point in the model is constant.
|
Create a Bayesian Information Criterion (BIC) penalty. |
|
Create a chi-square penalty. |
Linear penalties#
The penalty for each additional change point in the model is linear in the number of variables affected by the change. Only relevant for multivariate data.Some detectors use such penalties to identify the variables responsible for the change or anomaly. Penalised scores using linear penalties are faster to compute than non-linear penalties.
|
Create a linear penalty. |
Create a linear chi-square penalty. |
Nonlinear penalties#
The penalty for each additional change point in the model is non-linear in the number of variables affected by the change. Only relevant for multivariate data. Some detectors use such penalties to identify the variables responsible for the change or anomaly.
Create a nonlinear chi-square penalty. |