Description
Parameter-Free Domain-Agnostic Season Length Detection in Time Series.
Description
Spectral and Average Autocorrelation Zero Distance Density ('sazed') is a method for estimating the season length of a seasonal time series. 'sazed' is aimed at practitioners, as it employs only domain-agnostic preprocessing and does not depend on parameter tuning or empirical constants. The computation of 'sazed' relies on the efficient autocorrelation computation methods suggested by Thibauld Nion (2012, URL: <https://etudes.tibonihoo.net/literate_musing/autocorrelations.html>) and by Bob Carpenter (2012, URL: <https://lingpipe-blog.com/2012/06/08/autocorrelation-fft-kiss-eigen/>).
README.md
SAZED
Overview
SAZED (Spectral and Average Autocorrelation Zero Distance Density) is a package and ensemble method for estimating the season length of a seasonal time series. SAZED is aimed at practitioners, as SAZED employs only domain-agnostic pre-processing and does not depend on parameter tuning or empirical constants.
Installation
This package requires the Debian packages libxml2-dev, libfftw3-dev and zlib1g-dev. Once you have those, then simply execute:
install.packages("sazedR")
Usage
Estimate the season length of a seasonal time series with the variations of the ensemble method SAZED:
library(sazedR)
season_length <- 26
y <- sin(1:400*2*pi/season_length)
sazed(y)
All components of the SAZED ensemble are also available separately. For more information on them, see the package manual and examples therein.