Alligator - Concepts
Concepts
Covered on this page are:
- Sample sets
- Quantitative models
- Qualitative models
- Model sets
- Transfer models
- Filters
- Grouping
- Network tools
- Serial numbers
- Model protection and encryption
- User groups
Two settings pages define the behaviour of the Alligator:
- program settings
- project settings
Both can be reached with a right mouse click on "Load Proejct". The project settings are obviously only accessible when a project has been loaded.
In the program settings information like Alligator license, network information (NET+ version only) and point sizes and stroke thickness in plots.
The project settings are manifold:
While most options are obvious for "grouping", "Check serial numbers" see explanations below.
"Automatically add other models in a group to the selected model" will ease to select all models in a group by selecting only one model from the group. This applies to prediction and validation only. An alternative would be to add models to a model set, in this case all models in the model set are always used simultaneously.
Sample set are collections of samples with sample names, spectra, constituent value and further sample information. Sample names may occur multiple in one sample set. Sample set names need to be unique in a project. White spaces in sample set name are trimmed.
Constituent names in a sample set need to be unique.
There are different quantitative models in Alligator: MLR, PLS, LOCAL, SVM and ANN. Depending on the actual license installed some quantitative model types may not be available in calibration development (grayed out).
Model names will typically be constituent names. In prediction the model name is used to identify the column of predictions. In validation the model names and constituent names are used to match predicted and reference values to compute the validation statisics. While in a single sample set no repeated constituent names are allowed, model names may be repeated in a project.
In addition PCA models as qualitative models are available. The prediction output of PCA models is limited to distance of the sample to the model in terms of global H values, Hotteling's T² and neighborhood H distance.
Furthermore calculated models may be used to process quantitative results in simple algebraic formulae. If clauses are supported.
The above models may be combined into a model set. These model sets assure all that all models in the model set are used simultaneously in prediction and validation. Model sets are the primary target to export e.g for usage in BlueScan.
Transfer models are used to modify spectra in the meaning of spectral standardization. Transfer models of different complexity are available. Transfer model based on a single transfer sample may only correct a systematic part for each data point. Transfer models based on multiple samples may also correct a slope or quadratic part at each data point.
The PDS algorithm may also move the x axis in the transfer model.
Filters are used in online measurements to separate samples with proper spectra - passing the filter - and samples with invalid spectra before they are averaged.
Grouping of sample sets and models is supported throughout Alligator to ease data overview. Gouping may be turend on and off in the project settings.
Grouping on:
Grouping off:
Modifying grouping for a project requires the project to be loaded newly.
Network tools are used when working with multiple instruments.
The primary tools are to create and apply transfer models and to trim models to a different wavelength range.
The other tools are plotting spectra and prediction of multiple sample sets simultaneously.
The networking tools are available in the NET version (or NET+) of Alligator only.
Transfer models and serial numbers are used in Alligator to verify that an appropriate model is used in prediction or validation. In the project settings (right mouse click on "Load Project", menuitem "project settings") for a certain project Alligator can be configured to check serial numbers before prediction.
Alligator will verify that the target number of a sample set and the model(s) / model set used for prediction are identical. If not prediciton will be aborted. A transfer model, if created properly, can modify the spectra to match spectra from another instrument and thereby grant for proper prediction.
An solution for a mismatch of terget numbers may be cured by the following setup:
| spectra | transfer model | model(s) | |||
|---|---|---|---|---|---|
| target number | 123456 | source number | 123456 | ||
| target number | 654321 | target number | 654321 | ||
The original target number of s spectrum is the serial number of the insturment.
An empty terget number of a prediction model will indicate that the model may be used on any spectrum, no serial numbers will be checked.
An empty source number on a transfer model will indicate that the transfer model can be used for any spectrum.
Checking of serial numbers may as well be turned off generally in the project settings.
Model protection and encryption
Any project - as a collection of spectra, samples, reference data and models - has an owner. By default the PC which the project is created on is the owner of a project. The owner might change the ownership of a project.
Any quantitative or qualitative model created in Alligator gets an owner. The owner of a newly created model is is the owner of the project. This ownerwhip is saved with the model in the project as well as in the exported file. Only the owner might grant a license - see below - to use the model on any other PC.
Any quantitative or qualitative model created in Alligator has a license. The default license is identical to the owner of the model. During prediction or validation the license is checked to verify the PC is eligible to use the model.
This guarantees that the model is only used on the PC is was created on.
When exporting a model a dialog will ask whether the license for the exported copy of the model shall be changed:
This new license is written to the file being created during this export. The exported copy of the model(s) may then be used on the PC(s) it was licensed to.
As can be seen a free model may be created which can be used on any PC.
Independent from the license a model may be encryted. While an unencryted model is an ASCII file - all information in the model file (b vector, scores, weights, ...) is human readable - an encrypted model file is AES 1024 encrypted. The encryption for the model or model set output is pre-selected by default. SVM and ANN models do not exist in an unencrypted format.
When moving a project (.Alligator file) to another PC the models in the project are still owned and licensed to the original PC solely. The project ownership and thereby the ownership and licenses to all modéls in the project may and should be modified. Only this will guarantee that the models in the transferred project file will work properly.
User groups
When collaborating with several users of Alligator / BlueScan a user might define user groups for a project. The information is stored within each project. A user group is a list of users.
A user group might be selected e.g. for licensing a model during export to file - see above. A user group might also be selected when exporting a project or when changing the ownership of a project. If a user group is selected as the owner each user in the user group gets full access to the models and to grant further ownership of the project.