search

facebookresearch / Kats

Kats, a kit to analyze time series data, a lightweight, easy-to-use, generalizable, and extendable framework to perform time series analysis, from understanding the key statistics and characteristics, detecting change points and anomalies, to forecasting future trends.
MIT License
4.88k stars 534 forks source link

What parameters can be used for specify change directions in bayesian and robust changepoint detection models? #259

Closed jscanass closed 1 year ago

jscanass commented 2 years ago

Hi there! In the detection tutorial notebook it is shown for the CUSUMDetector model a parameter, change_directions, for specifying what type of change to be detected. What is the analogous parameter in the case of BOCPDetector, and RobustStatDetector? Or in which other form I can specify an increase or decrease? A working example could be the same as the tutorial.

jeffhandl commented 1 year ago

@jscanass There is not a straightforward way to pre-specify the change direction in either of these detectors. For these detectors you will need to calculate all change points and then filter afterwards if you wish.