This module takes a time series and returns: (a) the underlaying linear trend and (b) the times where there is a change in the trend
Project description
ChangePointDetector This module takes a time series and returns: (a) the piecewise underlaying linear trend, and (b) the times where there is a change in the underlying trend
We use a Kalman Filter to find the piecewise underlying linear trend, with a state space representation of seasonality and an underlying linear trend. We initialise the parameters of the Kalman filter using a Least Squares estimator out of "statsmodels'
For the change detector on the underlying trend we use another Kalman filter, this time a single period autoregression. We then consider the Malhalanobis distance between that filter output, using a Gumbel distribution to decide where the increase in distance likely indicates a change in trend. This implements the approach described by Lee & Roberts at https://www.robots.ox.ac.uk/~sjrob/Pubs/LeeRoberts_EVT.pdf
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for ChangePointDetectorEVT-0.0.1.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | f8cd173622e8316667d34a7c084159ae00c98263c5923ac066d827be5e0e56d8 |
|
MD5 | f1f227b3955ea3f476ced355bdf9f2bc |
|
BLAKE2b-256 | a1e49da4ddd164c2116eca42b99d3bbf99ba115eb2ba9ba0af1157ca17ba4671 |
Hashes for ChangePointDetectorEVT-0.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2869e83f6d9dcc863c169d9d5d8f9ae84717a2dd6bdfc13e67b67b0614f214ce |
|
MD5 | 8d092b49d96a610b5c3b523ba48cc038 |
|
BLAKE2b-256 | 6a8bef07af74de4b2cf557b2e22d67fb2d51a6511efd931a1b3fefc743e9872a |