select.traces: Select traces from a data set according to several criteria
In ptw: Parametric Time Warping

Description Usage Arguments Details Value Author(s) See Also Examples

For alignment purposes, it may be useful to select traces which show clear features, and to throw away traces that contain mainly noise. This function implements three ways to achieve this: CODA, a criterion similar to varimax, and a criterion based on the highest intensity.

1 2	select.traces(X, criterion = c("coda", "var", "int"), window = 5, smoothing = c("median", "mean"))

`X`	a data matrix or an array. The first dimension signifies the traces from which a selection is to be made. If X is a matrix, the first usually corresponds to samples and the second dimension is the spectral dimension. If X is an array, the data are assumed to come from a hyphenated experiment, with the first dimension the chromatographic dimension, the second the spectral dimension and the third dimension corresponding to samples
`criterion`	either Windig's CODA algorithm, a criterion calculating the variances of the length-scaled spectra, or a criterion giving the height of the highest peak
`window, smoothing`	arguments to the coda function.

The CODA criterion in essence selects traces with no baseline and no spikes, but still containing significant intensities. The variance criterion aims at something similar: it calculates the variance (or standard deviation) of every trace after length scaling - traces with a high value show few highly structured features, whereas traces with a low value show noise or a significant baseline. The intensity criterion simply returns the intensity of the highest peak. The latter two criteria are simpler than CODA but implicitly assume that the traces have been preprocessed (i.c., spikes have been removed).

The function returns a list with components

`crit.val`	a vector containing the values of the criterion for all traces. If X is an array, the function is recursively applied to all samples (elements of the third dimension) - the results are multiplied to obtain one criterion value per trace
`trace.nrs`	the order of the traces (from large to small)

Ron Wehrens

coda

data(lcms)
ntrace <- dim(lcms)[1]
lcms.selection <- select.traces(lcms[,,1:2], criterion = "var")
good <- lcms.selection$trace.nrs[1]
bad <- lcms.selection$trace.nrs[ntrace]

par(mfrow = c(1,2))
matplot(lcms[good,,1:2], type = 'l', lty = 1)
matplot(lcms[bad,,1:2], type = 'l', lty = 1)