BUCHI NIRCal - Manual (Page 200)

NIRCal 5.5 Software Manual
200 NIRCal 5.5 Manual, Version A
3.18.81 Spectra Residuals
Description The RMSD (Root Mean Squared Deviation) of the spectra residuum.
Use To look at residual outlier.
Method PCR / PLS / Cluster (CLU) / SIMCA
Matrices ID 41
Tip Special tip to optimize the calibration
Details The secondary PCs take no effect.
Formula
Related Topic X-PRESS , Spectra Residuum
Indicates the difference of the pretreated spectra and the reconstructed spectra with the number of
primary PCs.
For the application the allowed residual is in most cases: 2x C-Set maximum residual. The default
blow up limit of 2 can be adjusted by the user.
3.18.82 Spectra Residuals too big
Description Number of spectra with too big residual.
Use Additional information for primary PCs selection or for outlier search.
Method PCR / PLS / Cluster (CLU) / SIMCA
Matrices ID 54
Tip The number of spectra with residual too big should be 0.
Details The allowed residual is = 2* max. C-Set residual. Using this calculation only
V-Set spectra can be out of the limit.
2 is the default of residual blow up, can be edited by the user.
Related Topic Residuals , Primary PCs
In a good calibration all spectra should fall within the allowed residual limits, so it should be zero in this
window.
3.18.83 Spectra Residuum
Description Difference spectra between the pretreated spectra and the reconstructed
spectra.
Use See where (wavelength, spectra) the major differences are.
Method PCR / PLS / Cluster (CLU) / SIMCA
Matrices ID 9
Tip To see that the C-Set spectra are not overfitted open this plot twice. In the
first set the visibility to the C-Set spectra, in the other set the visibility to the
V-Set spectra. The two sets should be similar.
Details
Related Topic Loadings , Pretreated Spectra , X-PRESS , Reconstructed Spectra
The spectra residuum shows the difference spectrum for all pretreated spectra and the corresponding
reconstructed spectra calculated with the number of primary principal components. The size of these
difference spectra should be roughly the same for C- and V-Set, when the correct number of primary