VB PLS2 Cross Validation Example

Imports System
Imports System.IO

Imports CenterSpace.NMath.Core


Namespace CenterSpace.NMath.Examples.VisualBasic

  A .NET example in Visual Basic showing Partial Least Squares (PLS) and
  cross validation.

  This example shows how to use the CrossValidation classes to find the 
  optimal number of components for a Partial Least Squares calculation.
  For each number of components we will perform a K-fold cross validation.
  In K-fold cross validation the data set is divided into k subsets, and 
  the holdout method is repeated k times. Each time, one of the k subsets 
  is used as the test set and the other k-1 subsets are put together to 
  form a training set. Then the average error across all k trials is computed.
  The optimal number of components will then be the number of components for
  which this average error is a minimum.

  Module PLS2CrossValidationExample

    Sub Main()

      Dim spectralData As New DoubleMatrix(New StreamReader("chemometricX.dat"))
      Dim ConcentrationData As New DoubleMatrix(New StreamReader("chemometricY.dat"))

      Dim numDependentVars As Integer = ConcentrationData.Cols
      Dim numIndependentVars As Integer = spectralData.Cols
      Dim numSamples As Integer = spectralData.Rows
      Dim K As Integer = 6

      The CrossValidation class needs the full set of data, a way to generate
      subsets of the data and a PLS calculator object. The subset generator is 
      specified by an instance of the ICrossValidationSubets interface.
      Dim subsetGenerator As KFoldsSubsets = New KFoldsSubsets(K)

      Construct a PLS2 cross validation object that uses SIMPLS algorithm to
      calculate the partial least squares models.
      Dim Calculator As PLS2SimplsAlgorithm = New PLS2SimplsAlgorithm()
      Dim CV As PLS2CrossValidation = New PLS2CrossValidation(Calculator, subsetGenerator)

      Now for each number of components perform cross validation and record the 
      minimum average Mean Square Error and the number of components at which
      it is achieved. 
      Dim optimalNumComponents As Integer = -1
      Dim minMse As Double = Double.MaxValue

      Console.WriteLine()
      Console.WriteLine("Components" & ControlChars.Tab & "MeanSquareError")
      Console.WriteLine("=================================")
      Console.WriteLine()
      Console.WriteLine()

      For numComponents As Integer = 1 To (numIndependentVars - 2)
        CV.DoCrossValidation(spectralData, ConcentrationData, numComponents)
        If (Not Calculator.IsGood) Then
          Console.WriteLine("Calculation with {0} components is not good. Message:", numComponents)
          Console.WriteLine(Calculator.Message)
        End If
        Dim MSE As Double = CV.AverageMeanSqrError.TwoNorm()
        Console.WriteLine(numComponents & ControlChars.Tab & ControlChars.Tab & MSE.ToString("G5"))
        If (MSE < minMse) Then
          minMse = MSE
          optimalNumComponents = numComponents
        End If
      Next

      Console.WriteLine()
      Console.WriteLine()
      Console.WriteLine("Optimal number of components = " & optimalNumComponents)
      Console.WriteLine("Minimum MSE = " & minMse.ToString("G5"))

      Console.WriteLine()
      Console.WriteLine("Press Enter Key")
      Console.Read()

    End Sub
  End Module
End Namespace