/// <summary>
/// Constructs a multi-component sparse Bayes Point Machine using shared variables for chunking data
/// </summary>
/// <param name="nClass">Number of components (classes)</param>
/// <param name="featureCount">Number of features</param>
/// <param name="noisePrec">Noise precision</param>
/// <param name="trainChunkSize">Chunk size for training</param>
/// <param name="testChunkSize">Chunk size for testing</param>
public BPMSparse_Shared(int nClass, int featureCount, double noisePrec, int trainChunkSize, int testChunkSize)
{
nComponents = nClass;
nFeatures = featureCount;
NoisePrec = noisePrec;
this.trainChunkSize = trainChunkSize;
this.testChunkSize = testChunkSize;
feature = new Range(nFeatures).Named("feature");
w = new SharedVariableArray<double>[nComponents];
IDistribution<double[]> wPrior0 = Distribution<double>.Array(nFeatures,
delegate(int index) { return Gaussian.PointMass(0); });
IDistribution<double[]> wPrior = Distribution<double>.Array(nFeatures,
delegate(int index) { return Gaussian.FromMeanAndPrecision(0.0, 1.0); });
for (int c = 0; c < nComponents; c++)
{
w[c] = (c == 0)
? SharedVariable<double>.Random(feature, (DistributionStructArray<Gaussian,double>)wPrior0).Named("w_" + c)
: SharedVariable<double>.Random(feature, (DistributionStructArray<Gaussian,double>)wPrior).Named("w_" + c);
}
trainModel = SpecifyTrainModel("_train", trainChunkSize);
testModel = SpecifyTestModel("_test", testChunkSize);
}
/// <summary>
/// Specify the training model
/// </summary>
/// <param name="s">The name of the training model</param>
/// <param name="nChunks">The number of chunks</param>
/// <returns>A <see cref="BPMVarsModelForTrain"/> instance</returns>
private BPMModelVarsForTrain SpecifyTrainModel(string s, int nChunks)
{
BPMDataVars[] dataVars = new BPMDataVars[nComponents];
// The model identifier for the shared variables
Model model = new Model(nChunks).Named("model" + s);
// The weight vector within a submodel
VariableArray<double>[] wModel = new VariableArray<double>[nComponents];
for (int c = 0; c < nComponents; c++)
{
// Get a copy of the shared weight vector variable for the submodel
wModel[c] = w[c].GetCopyFor(model).Named("wModel_" + c + s);
}
for (int c = 0; c < nComponents; c++)
{
Variable<int> nItem = Variable.New<int>().Named("nItem_" + c + s);
Range item = new Range(nItem).Named("item_" + c + s);
VariableArray<int> xValueCount = Variable.Array<int>(item).Named("xCount_" + c + s);
Range itemFeature = new Range(xValueCount[item]).Named("itemFeature_" + c + s);
VariableArray<VariableArray<double>, double[][]> xValues = Variable.Array(Variable.Array<double>(itemFeature), item).Named("xValues_" + c + s);
VariableArray<VariableArray<int>, int[][]> xIndices = Variable.Array(Variable.Array<int>(itemFeature), item).Named("xIndices_" + c + s);
using (Variable.ForEach(item))
{
// The score for this item across all components
Variable<double>[] score =BPMUtils.ComputeClassScores(wModel, xValues[item], xIndices[item], itemFeature, NoisePrec);
BPMUtils.ConstrainArgMax(c, score);
}
dataVars[c] = new BPMDataVars(nItem, item, xIndices, xValueCount, xValues);
}
BPMModelVarsForTrain bpmVar = new BPMModelVarsForTrain();
bpmVar.ie = new InferenceEngine();
bpmVar.dataVars = new BPMDataVars[nComponents];
bpmVar.model = model;
bpmVar.dataVars = dataVars;
return bpmVar;
}
