/// <summary>
/// Constructs a Sparse Bayes point machine instance from number of components and number of features
/// </summary>
/// <param name="nClass">Number of components (classes)</param>
/// <param name="nFeatures">Number of features</param>
/// <param name="noisePrec">noisePrec</param>
public BPMSparse(int nClass, int nFeatures, double noisePrec)
{
this.nClass = nClass;
this.nFeatures = nFeatures;
NoisePrec = noisePrec;
feature = new Range(nFeatures).Named("feature");
trainModel = SpecifyTrainModel("_train");
testModel = SpecifyTestModel("_test");
}
/// <summary>
/// Specifies the training model
/// </summary>
/// <param name="s">The name of the training model</param>
/// <returns>A <see cref="BPMSparseVarsForTrain"/> instance</returns>
private BPMSparseVarsForTrain SpecifyTrainModel(string s)
{
// Place to maintain variables for each component
BPMDataVars[] dataVars= new BPMDataVars[nClass];
// The weight vector for each component
VariableArray<double>[] w = new VariableArray<double>[nClass];
// The prior weight distributions for each component
VariableArray<Gaussian>[] wInit = new VariableArray<Gaussian>[nClass];
for (int c = 0; c < nClass; c++) {
// Weight variables are an array over feature range
w[c] = Variable.Array<double>(feature).Named("w_" + c + s);
// The prior weight distributions will be set by the calling program
wInit[c] = Variable.Array<Gaussian>(feature).Named("wInit_" + c + s);
// The weight for this component and feature
w[c][feature] = Variable<double>.Random(wInit[c][feature]);
}
// Loop over the components
for (int c = 0; c < nClass; c++) {
Variable<int> nItem = Variable.New<int>().Named("nItem_" + c + s);
Range item = new Range(nItem).Named("item_" + c + s);
// Array of feature counts per item
VariableArray<int> xValueCount = Variable.Array<int>(item).Named("xCount_" + c + s);
// Range over features - size is based on variable feature count
Range itemFeature = new Range(xValueCount[item]).Named("itemFeature_" + c + s);
// Jagged array of values - each item is an array of data values whose indices
// are given by the corresping xIndices[item]
VariableArray<VariableArray<double>, double[][]> xValues = Variable.Array(Variable.Array<double>(itemFeature), item).Named("xValues_" + c + s);
// Jagged array of indices for the items
VariableArray<VariableArray<int>, int[][]> xIndices = Variable.Array(Variable.Array<int>(itemFeature), item).Named("xIndices_" + c + s);
// Loop over the data items
using (Variable.ForEach(item)) {
// The score for this item across all components
Variable<double>[] score = BPMUtils.ComputeClassScores(w, xValues[item], xIndices[item], itemFeature, NoisePrec);
// The constraint imposed by the observed component
BPMUtils.ConstrainArgMax(c, score);
}
// Store the data information
dataVars[c] = new BPMDataVars(nItem, item, xIndices, xValueCount, xValues);
}
// Store the variables
BPMSparseVarsForTrain bpmVar = new BPMSparseVarsForTrain();
bpmVar.dataVars = dataVars;
bpmVar.ie = new InferenceEngine();
bpmVar.ie.ModelName = "BPMSparse_train";
bpmVar.w = w;
bpmVar.wInit = wInit;
return bpmVar;
}
