chchanges

Detect statistically meaningful changes in streams of data via online changepoint detection.

Available as a package on pypi.

pip install chchanges

For example, we can detect changes in a stream of normally distributed data, where the mean of the distribution is piecewise constant:

The stream of data

The evolving posterior distribution

The probability that a changepoint was detected

To generate these figures, experiment with chchanges/demos/mean.py

Contributing

• By adding other Posterior distributions and Hazard functions, you can fine-tune chchanges for your specific application.
• E.g. a Multivariate Student's T posterior would enable detecting changes in the correlation of multivariate data.
• Please contribute and expand the range of chchanges uses.

References:

• Ryan P. Adams, David J.C. MacKay, "Bayesian Online Changepoint Detection" (2007)
• Byrd, M Gentry et al. “Lagged Exact Bayesian Online Changepoint Detection with Parameter Estimation” (2017)

Other software implementations:

• https://github.com/dtolpin/bocd Particularly indebted to this implementation.
• https://github.com/hildensia/bayesian_changepoint_detection
• https://github.com/lnghiemum/LEXO

---

"turn and face the strange" - David Bowie