Getting Started

library(SpectralUnmix)
demo <- coelho_demo_spectra()

Load the bundled spectra

The bundled Coelho demo provides four normalized stellar spectra on a common wavelength grid. It is the simplest fully reproducible entry point for the package.

dim(demo$matrix)

[1]    4 2250

demo$metadata

           id                         file  teff logg feh
1  cool_giant t03000_g-0.5_p00p00_sed.fits  3000 -0.5 0.0
2 solar_dwarf t05750_g+4.5_p00p04_sed.fits  5750  4.5 0.0
3      a_star t09000_g+4.5_p00p00_sed.fits  9000  4.5 0.0
4    hot_star t18000_g+4.5_p02p00_sed.fits 18000  4.5 0.2

Fit the model

fit <- spectral_unmix(
  demo$matrix,
  k = 4,
  lambda_smooth = 0,
  niter = 400,
  lr = 0.03
)

Inspect the result

print(fit)
summary(fit)
basis(fit)
coef(fit)
fit$niter_run
fit$converged

Plot the spectra

plot(fit, type = "spectra", wavelength = demo$wavelength)
plot_reconstruction(
  fit,
  demo$matrix,
  wavelength = demo$wavelength
)

Recovered-versus-observed view for the bundled Coelho sample:

What the outputs mean

fit$spectra contains the recovered component spectra.
basis(fit) returns those components as wavelength-by-component columns.
fit$reconstruction is the model approximation A %*% S.
fit$loss stores the optimization history.
fit$niter_run reports how many optimization steps were actually used.
fit$converged reports whether early stopping triggered before niter.

Larger library subset

For a denser stellar-library example, use the curated Coelho subset bundled with the package.

stellar_lib <- coelho_stellar_subset()
dim(stellar_lib$matrix)

[1]  100 2250

table(stellar_lib$metadata$type)


  a_f_star cool_dwarf cool_giant   hot_star solar_like 
        20         20         20         20         20

fit_lib <- spectral_unmix(
  stellar_lib$matrix,
  k = 5,
  lambda_smooth = 0.001,
  niter = 500,
  lr = 0.03
)

plot(fit_lib, type = "spectra", wavelength = stellar_lib$wavelength)

IFU note

The package still supports cube workflows through cube_to_matrix(), matrix_to_cube(), predict(type = "cube"), and component_map(). Those helpers are kept in the API, but the website examples stay focused on the bundled Coelho data so the documentation remains short and reproducible.