Estimate trend and seasonal effects in a timeseries
Project description
Robustly estimate and remove trend and periodicity in a timeseries.
Seasonal can recover sharp trend and period estimates from noisy timeseries data with only a few periods. It is intended for estimating season, trend, and level when initializing structural timeseries models like Holt-Winters. Input samples are assumed evenly-spaced from a continuous-time signal with noise but no anomalies.
In this package, trend removal is in service of isolating and estimating periodic (non-trend) variation. “trend” is in the sense of Cleveland’s STL decomposition – a lowpass smoothing of the data, rather than a single linear trend (though you can opt for this). Detrending is accomplishd by a coarse fitted spline or a median filter.
The seasonal estimate will be a list of period-over-period averages at each seasonal offset. You may specify a period length, or have it estimated from the data. The latter is an interesting capability of this package.
See README.md for details on installation, API, theory, and examples.
Dependencies
package: numpy, scipy extras: pandas, matplotlib
Project details
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 seasonal-0.3.0-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | b795eee6c3a35c1edbb8cd20b671852ef2e2d9da913624b9079bbdf5feacd483 |
|
MD5 | b64e9b04e7457bdb4d74df432b02d907 |
|
BLAKE2b-256 | cf3637f0161e1bb478cd7f7d5ee2e71209cabaf1f04e98ca9f5ec588158b18a2 |