Calibration Model Wizard
- Step 3 - Calibration & Validation data
Step 3 shows a first overview of all spectra and the calibration and
validation data. Additional statistical analysis is also available:
The calibration data view in the upper part of the dialog shows an average
spectrum (red line) derived from all data objects selected for calibration.
The light red shape around the average spectrum indicates the variance
of all spectra. If a particular data object is selected in the tree view
below, the corresponding spectrum graph is shown as green line in the
data view.
As in all data views in the software, the user can zoom into spectral
regions by left
clicking and dragging
the mouse. A double-click
in the data view restores the original view.
The experienced user will draw conclusions for selection of significant
spectral regions. Some additional statistical spectra can be displayed
as described below.
In the lower part of the dialog, a tree view shows all available data
objects from the selected project. By default all objects are taken into
account for calibration and none are used for validation. Corresponding
property values of the selected property are shown accordingly.
The user can change the selection for calibration and validation spectra
and also modify property values in this dialog.
Clicking the Statistical
Spectra button shows a
menu with several statistical spectra:
Show Correlation for property
Show Mean
Show Variance
Show Standard deviation
Show Calibration Spectra
Show Validation Spectra
Show Band
By checking or un-checking those entries the corresponding item is shown
in the data view or not. The mean spectrum and the band are shown by default.
The top of the menu will hold entries to show the correlation for each
selected property.
Data objects can be selected for use in calibration or validation data
sets or both.
The selection of a single data object is modified by double
clicking on Yes or No with the Left
mousebutton in the respective
calibration or validation column field. In this case selection is toggled.
What is the minimum number of calibration
spectra allowed?
For all calibration models except for the
univariate model at least two spectra
need to be part of the calibration set in order to perform a successful
calibration.
For the univariate
model one spectrum is sufficient, but additional options will be added
to be able to perform a successful calibration: The flag Data
Origin is Zero in the General Options of the Model Settings
will be automatically activated and dummy spectrum with data value of
(0,0) for all datapoints will be added to the calibration set and calibration
project.
To change selection for multiple data objects at once, please select the project
or folder node above in the tree.
Then click the Right
mouse button to show the
following context menu. Alternatively multiple objects can be selected
by holding the shift
or ctrl
button and left clicking.
Add to calibration set
Adds all selected data objects to the calibration set. In the calibration
column, all entries for those data objects will be changed to Yes.
Remove from calibration set
Removes all selected data objects from the calibration set. In the
calibration column, all entries for those data objects will be changed
to No.
Add to validation set
Adds all selected data objects to the validation set. In the validation
column, all entries for those data objects will be changed to Yes.
Remove from validation set
Removes all selected data objects from the validation set. In the validation
column, all entries for those data objects will be changed to No.
Those users, who do not have a separate or independent set of validation
spectra, may use the calibration spectra for validation purposes. In order
to define a random set of validation spectra, click the Auto
Validation button. The
following dialog is shown:
Percentage of Spectra
A user defined percentage of available data objects will be randomly assigned as validation spectra according to the following conditions:
Selection Based On
Calibration Data
Only those objects in the tree being used in calibration (Calibration column entry = Yes) are considered for validation data selection
All Data
All data suitable objects in the project are considered for validation data selection.
Why are some data objects not considered?
Only those spectra can be considered, which have a properly assigned component label. Especially for quantitative calibration modelling a numeric value must be present to be considered for use in calibration modelling.
Action
Toggle Selection
This toggles the given percentage of the current active data selection. Spectra defined as calibration spectra become validation spectra and vice versa.
Only Activate Validation
This keeps the current selection of calibration spectra and will only add the given percentage of validation data.
Cross validation is another option to perform the validation of the
calibration model. Please refer to the topic Cross Validation Setup in the chapter Chemometric
Analyis for detailed description of the Cross Validation.
Typically, all property values should be properly adjusted before starting
the calibration wizard.
How can I edit the labels of my calibration
data?
The most convenient way to add a new label,
change existing label or remove labels from multiple spectra at once is
using the Label
Editor dialog. It can be started by clicking on the Edit
Labels button on the right
side of the window. Outside of the Calibration Model Wizard it can be
started from the Tools menu.
Nevertheless, in some cases it will be necessary to change a property
value during calibration, e.g. because of typing errors.
A value can be changed by clicking
on it with the Left
mouse button. In this case,
the value becomes editable. After completion of changes, you need to click somewhere else to apply changes.
Navigation
Just click the Next
> button to proceed to the next step.
Clicking the Cancel button will
abort creating a new calibration.
