Return a series of least-squares transforms of a real-valued time series. Each transform is minimized independently for each frequency. The method used is a Lomb-Scargle transform of the real-valued (time, mag) series, starting from frequency maxfreq and descending numoctaves octaves with numcoeff coefficients per octave.
The result of the transform for each frequency is the coefficient of a sum of sine and cosine functions modified by that frequency, in the form of a complex numberâwhere the cosine coefficient is encoded in the real term, and the sine coefficient is encoded in the imaginary term. Each frequency is fit independently from the others, and to minimize very low frequency error, consider storing the mean of a dataset with a constant or near-constant offset separately, and subtracting it from the dataset.
See also: lscomplex.
Package: lssa