Ruptures

Ruptures [@Truong2020] detects transitions by partitioning the time series into homogeneous segments and finding breakpoints that minimize a cost function. Ammonyte exposes six search algorithms (Pelt, Binseg, BottomUp, Dynp, Window, and KernelCPD) and multiple cost functions, making it the most flexible of the three methods.

When to use

Best for: Sharp structural breaks in mean or variance, when flexibility in algorithm and cost function choice is needed.

Limitation: Like the Augmented KS Test, ruptures detects statistical breaks in amplitude. It says nothing about the dynamics, for dynamically-sensitive detection, see LERM.

Reference

Truong, C., Oudre, L., & Vayatis, N. (2020). Selective review of offline change point detection methods. Signal Processing, 167, 107299. https://doi.org/10.1016/j.sigpro.2019.107299

Tutorial

See the Ruptures Validation notebook for a worked example.