Dataset Reference Spectra

Reference spectra add known signatures to a dataset's Training or Validation data. They are useful only when the chosen model and scientific workflow can learn from those signatures.

Add signatures

1. Add the relevant classes on Overview.
2. Open Reference spectra and select Add reference spectra.
3. Choose the class, search or browse the spectral libraries, and select the signatures to add. New selections enter Training.
4. Select signatures to change their class, move or copy them between Training and Validation, or remove them. You can also drag signatures between the two groups.

Reference spectra do not use a Test group. Avoid placing the same signature in both Training and Validation: Clarity allows it, but the validation metrics can be overly optimistic.

Automatic splitting also cannot create independent validation from a class with only one eligible signature; the current generator can place that signature in both buckets. Add another independent signature or use a separate validation source rather than treating the duplicated result as held out.

When dataset PCA is enabled, a current generator limitation includes Validation reference spectra in the fitted PCA transform. Validation is therefore not independent in that configuration. See Advanced Dataset Settings.

The wavelength picker includes wavelength sets from reference spectra as well as images. Resolve incompatible ranges there and configure normalization and resampling under Advanced Dataset Settings.

Manage the source signatures under Spectral Libraries.