asysm: Trend estimation with asymmetric least squares

In ptw: Parametric Time Warping

Description

Estimates a trend based on asymmetric least squares. In this case used to estimate the baseline of a given spectrum.

Usage

  asysm(y, lambda = 1e+07, p = 0.001, eps = 1e-8, maxit = 25)

Arguments

y	data: either a vector or a data matrix containing spectra as rows
lambda	smoothing parameter (generally 1e5 - 1e8)
p	asymmetry parameter
eps	numerical precision for convergence
maxit	max number of iterations. If no convergence is reached, a warning is issued.

Details

Asymmetric least squares (not to be confused with alternating least squares) assigns different weights to the data points that are above and below an iteratively estimated trendline, respectively. In this case, the asymmetry parameter p (0 <= p <= 1) is the weight for points above the trendline, whereas 1-p is the weight for points below it. Naturally, p should be small for estimating baselines. The parameter lambda controls the amount of smoothing: the larger it is, the smoother the trendline will be.

Value

An estimated baseline

Author(s)

Paul Eilers, Jan Gerretzen

References

Eilers, P.H.C. Eilers, P.H.C. (2004) "Parametric Time Warping", Analytical Chemistry, 76 (2), 404 – 411.

Boelens, H.F.M., Eilers, P.H.C., Hankemeier, T. (2005) "Sign constraints improve the detection of differences between complex spectral data sets: LC-IR as an example", Analytical Chemistry, 77, 7998 – 8007.

Examples

data(gaschrom)
plot(gaschrom[1,], type = "l", ylim = c(0, 100))
lines(asysm(gaschrom[1,]), col = 2)

Example output

ptw documentation built on Jan. 19, 2022, 5:07 p.m.