Signal-integral matrix

Builds a per-spectrum peak-area matrix. Each row is a spectrum, each column is a chemical-shift datapoint. For each peak, the column is picked from lcpar$pcisn (post-snap) when available, else lcpar$pcial (post-CluPA), else lcpar$pcide (post-decon). Peaks with pcisn = NA (snapped beyond maxCombine) are skipped. Collisions on the same column have their A * pi summed.

si_mat() is intentionally a dumb peak-list rasterizer: all alignment (continuous shift via CluPA) and reference snapping must have happened upstream — typically inside align(). To build a feature matrix where every spectrum shares the same column grid, run align(x, maxShift, maxCombine) first.

Usage

si_mat(x, drop_zero = FALSE, igrs = list(), peakPos = NULL, ...)

Arguments

x

A decons2 or aligns object.

drop_zero

Drop columns whose entries are all zero?

igrs

List of two-element ppm intervals to zero out before returning.

peakPos

Optional integer column indices. When supplied (predict mode) the matrix is subset to those columns; when NULL and used as a feat_fun the non-zero columns are kept and attached as attr(., "peakPos").

...

Ignored (protocol compatibility with other feat_funs).

Value

A numeric matrix with one row per spectrum and length(x[[1]]$cs) columns (the full cs grid). Column names are ppm values; row names are spectrum names.

Author

2024-2026 Tobias Schmidt: initial version.

Examples

decons <- deconvolute(sim[1:2], sfr=c(3.55, 3.35), verbose=FALSE)
aligned <- align(decons, maxShift=50, maxCombine=20, verbose=FALSE)
X <- si_mat(aligned)