This .NET C# command line application example demonstrates how to implement the chemometrics prediction API into your software. Demo data and integration tests are included to show real outputs.
Complete review of the functionality with major changes in interface communication:
LabCognition.Interface (1.1.1) has been replaced by LC.Predictor (0.1.0) and LC.Runtime (0.1.0) offering slightly enhanced functionality. Interfaces and features have been split into two new nuget packages.
| LabCognition.Interface (1.1.1) | LC.Runtime (0.1.0) | Action |
|---|---|---|
| IPredictionEngineFactory | IPredictorFactory | interface renamed and enhanced |
| InstanceFactory class | InstanceFactory class | moved |
| MsiApi class | MsiApi class | moved |
| MsiConfig class | MsiConfig class | moved and enhanced to support external configuration file |
| Plugin class | Plugin class | moved |
| KnownMsi list | KnownMsi list | Change handling by reading list from json MsiConfig.config file |
| LabCognition.Interface (1.1.1) | LC.Predictor (0.1.0) | Action |
|---|---|---|
| IPredicitonEngine | IPredictor | renamed and methods changed |
| IPredictionEngineReport | no implemention! | deprecated; use IPredictor methods |
| object[,] Predict(double[] dataX, double[] dataY, string[] constituents, string[] properties); | IPredictionResult[]? Predict(double[] dataX, double[] dataY); | use type specific prediction results and changed the result output using IPredicitonResult |
| string[] GetProperties(); and string[] GetConstituents(); | IPredictionResult[]? GetResultPreview(); | changed concept by returning an empty preview containing flat PredictionResult<T> objects of the expected results rather than lists of properties and constituents |
Integration tests clearly illustrate how to use the sample predictor application with all kinds of calibration models.
With the new implementation it is much easier to gather possible prediction results returned by a calibration model upfront using the 'GetResultPreview()' method. It returns a full prediction result with all properties you might expect in a regular prediction when using the 'Predict()' method. Only the values are empty.
This sample predictor application uses LabCognition.Interface to accomodate communication between the SamplePredictor.exe and the underlying prediction engine software, e.g. panorama.exe.
- LC.Predictor NuGet package
- LC.Runtime NuGet package
- CommandLineParser NuGet Package
- .NET Standard 2.0
-
The Quantify add-on module in panorama provides you the chemometric data modelling environment to create univariate and multivariate calibration models from your spectroscopic data files yourself. It also implements the IPredictorFactory and IPredictor interfaces to serve as the prediction engine software for the SamplePredictor application.
The source code has been created with Visual Studio 2019.
-
SamplePredictor.exe source code
-
SamplePredictor integration tests source code
-
Demo calibration data
Qualitative and quantitative calibration model files (*.calibration) are enclosed in the SamplePredictor\Data\ folder to this project. Naming convention for file is the following: <Sample spectrum to use>-<calibration model type>-<calibrated constituent(s)>.calibration
NOTE: The calibrations have been solely designed to depict their usage. They do not produce any reliable scientific outputs!
-
Demo spectrum files
Spectrum files are provided as comma separated x,y data text files (*.txt)
-
Milk.txt
Use with quanitative calibration models. It is an excerpt of a mid infrared spectrum taken from cow milk powder. The calibrated constituent(s) are the fat content ('Fat') and the lactose content ('Lac') in %.
-
SweetenerAspartame.txt
Use with qualitative calibration models. It is an excerpt of a near infrared spectrum file taken from pure Aspartame sweetener powder. Qualitative calibrations may be capable of distinguishing the constituents Aspartame, brown sugar, Saccharin and Sucralose.
-
The SamplePredictor application supports two major features:
-
Sample spectrum prediction
Performs qualitative or quantitative prediction using a sample spectrum to get the comprehensibe prediction statistics.
-
Prediction preview
Performs a dry run of the qualitative or quantitative prediction without any sample spectrum to obtain a dummy prediction statistics containing all outputs you can get.
You will need the panorama software version 8.0.0 or higher serving as the prediction engine software.
-
Download and install the panorama Software on your computer.
NOTE: panorama requires a valid license to work properly! Usually it includes a 30-days trial license which grants access to all features of the software. If you run into problems, please contact [email protected].
Follow the steps below to run the sample predictor application:
-
Open the SamplePredictor.sln file with Visual Studio 2019 (or higher).
-
The best way to get started is to have a look at the SamplePredictor.IntegrationTests project.
-
Open the file PredictionTests.cs
It contains tests for all availbale 'Predict' and 'GetResultPreview' use cases. Each test operates the SamplePredictor.exe with the appropriate calibration model and sample spectrum file. The expected text output is provided as well and compared with the actual test run's output.
-
Before you run any test, please make sure to to have the panorama.exe file path properly set! Adjust the following line:
private static string PredictorExecutablePath => @"c:\Program Files (x86)\LabCognition\Panorama\Panorama.exe";
The sample predictor executable can also be run from a DOS prompt with the following arguments:
SamplePredictor.exe [OPTIONS] [FILES]
Example:
SamplePredictor.exe -f c:\...\Panorama.exe -m c:\...\CalibrationModel.calibration c:\...\Spectrum.txt
| Option | Description | Comment |
|---|---|---|
| -f | Fully qualified file path to the binary implementing the IPredictorFactory. (See example above) | Mandatory |
| -m | Fully qualified file path to the calibration model file (*.calibration) |
The SamplePredictor converts the prediction and prediction preview outputs into some text report. Most of the source code in Program.cs belongs to report creation rather than the prediction itself.
In your application you would only require the following:
-
Create an instance of the prediction engine software (i.e. panorama.exe) using methods from LC.Runtime package.
This is done here:
var factory = InstanceFactory<IPredictorFactory>.Create(factoryPath);
-
Load a calibration model into the factory using LC.Predictor package.
IPredictor predictor = factory.ReadModelFromFile(modelPath);
-
Get a preview of all results returned when using a specific calibration model:
IPredictionResult[]? preview = predictor.GetResultPreview();
Doing this might help you to accomodate the user interface or prepare your own data structures.
-
Perform a prediction with x,y spectrum data
IPredictionResult[]? results = predictor.Predict(x, y);
Since the value type of a result is not unique for all results they are returned as IPredictionResult. The class instance behind is generic PredictionResult<T>. System.Double and System.String are currently supported types.
Use the GetValueType() method in the IPredictionResult interface to identify the value type in order to resolve the PredictionResult<T> instance behind. An example implementation for such conversion is provided in the ResultToString() method. Here is the core conversion if-statement:
Excerpt from 'public static string ResultToString(IPredictionResult result)'
if (result.GetValueType() == typeof(string))
{
value = (result as PredictionResult<string>)!.Value;
}
else if (result.GetValueType() == typeof(double))
{
value = (result as PredictionResult<double>)!.Value.ToString(CultureInfo.InvariantCulture);
}
else
{
// other value types are not supported!
return value;
}The *.calibration file contains a calibration model which must have been created with the panorama software.
For details on how to create your own calibration models and export them as *.calibration files, please have a look at the panorama calibration manual.
You may provide one or more spectrum files separated by space for prediciton.
The spectrum files must be UTF-8 encoded text files containing comman separated x,y data points in lines. One data point must be given per line followed by a carriage return line feed (\r\n). Please also have a look at the enclosed *.txt files in the SamplePredictor\Data\ folder.
Multivariate data analysis is offering powerful mathematical methods for predicting physical properties from spectroscopic data rather than gathering such information from laboratory wet chemistry sample analysis. Derived calibration models create a correlation between the physical sample properties and spectral responses at particular wavelengths. In many applications in the Chemical, Pharmaceutical industry and R&D calibration models are used for quantitative and qualitative property predictions to solve analytical problems. Spectra of your samples are taken with a spectrometer (e.g. Infrared, Raman, UV/Vis, Fluorescence, ...).