bool KMeansQuantizer::train_(MatrixDouble &trainingData){
//Clear any previous model
clear();
//Train the KMeans model
KMeans kmeans;
kmeans.setNumClusters(numClusters);
kmeans.setComputeTheta( true );
kmeans.setMinChange( minChange );
kmeans.setMinNumEpochs( minNumEpochs );
kmeans.setMaxNumEpochs( maxNumEpochs );
if( !kmeans.train_(trainingData) ){
errorLog << "train_(MatrixDouble &trainingData) - Failed to train quantizer!" << endl;
return false;
}
trained = true;
initialized = true;
numInputDimensions = trainingData.getNumCols();
numOutputDimensions = 1; //This is always 1 for the KMeansQuantizer
featureVector.resize(numOutputDimensions,0);
clusters = kmeans.getClusters();
quantizationDistances.resize(numClusters,0);
return true;
}
bool DecisionTreeClusterNode::computeError( const ClassificationData &trainingData, MatrixFloat &data, const Vector< UINT > &classLabels, Vector< MinMax > ranges, Vector< UINT > groupIndex, const UINT featureIndex, Float &threshold, Float &error ){
error = 0;
threshold = 0;
const UINT M = trainingData.getNumSamples();
const UINT K = (UINT)classLabels.size();
Float giniIndexL = 0;
Float giniIndexR = 0;
Float weightL = 0;
Float weightR = 0;
VectorFloat groupCounter(2,0);
MatrixFloat classProbabilities(K,2);
//Use this data to train a KMeans cluster with 2 clusters
KMeans kmeans;
kmeans.setNumClusters( 2 );
kmeans.setComputeTheta( true );
kmeans.setMinChange( 1.0e-5 );
kmeans.setMinNumEpochs( 1 );
kmeans.setMaxNumEpochs( 100 );
//Disable the logging to clean things up
kmeans.setTrainingLoggingEnabled( false );
if( !kmeans.train_( data ) ){
errorLog << __GRT_LOG__ << " Failed to train KMeans model for feature: " << featureIndex << std::endl;
return false;
}
//Set the split threshold as the mid point between the two clusters
const MatrixFloat &clusters = kmeans.getClusters();
threshold = 0;
for(UINT i=0; i<clusters.getNumRows(); i++){
threshold += clusters[i][0];
}
threshold /= clusters.getNumRows();
//Iterate over each sample and work out if it should be in the lhs (0) or rhs (1) group based on the current threshold
groupCounter[0] = groupCounter[1] = 0;
classProbabilities.setAllValues(0);
for(UINT i=0; i<M; i++){
groupIndex[i] = trainingData[ i ][ featureIndex ] >= threshold ? 1 : 0;
groupCounter[ groupIndex[i] ]++;
classProbabilities[ getClassLabelIndexValue(trainingData[i].getClassLabel(),classLabels) ][ groupIndex[i] ]++;
}
//Compute the class probabilities for the lhs group and rhs group
for(UINT k=0; k<K; k++){
classProbabilities[k][0] = groupCounter[0]>0 ? classProbabilities[k][0]/groupCounter[0] : 0;
classProbabilities[k][1] = groupCounter[1]>0 ? classProbabilities[k][1]/groupCounter[1] : 0;
}
//Compute the Gini index for the lhs and rhs groups
giniIndexL = giniIndexR = 0;
for(UINT k=0; k<K; k++){
giniIndexL += classProbabilities[k][0] * (1.0-classProbabilities[k][0]);
giniIndexR += classProbabilities[k][1] * (1.0-classProbabilities[k][1]);
}
weightL = groupCounter[0]/M;
weightR = groupCounter[1]/M;
error = (giniIndexL*weightL) + (giniIndexR*weightR);
return true;
}
bool KMeansFeatures::train_(MatrixDouble &trainingData){
if( !initialized ){
errorLog << "train_(MatrixDouble &trainingData) - The quantizer has not been initialized!" << endl;
return false;
}
//Reset any previous model
featureDataReady = false;
const UINT M = trainingData.getNumRows();
const UINT N = trainingData.getNumCols();
numInputDimensions = N;
numOutputDimensions = numClustersPerLayer[ numClustersPerLayer.size()-1 ];
//Scale the input data if needed
ranges = trainingData.getRanges();
if( useScaling ){
for(UINT i=0; i<M; i++){
for(UINT j=0; j<N; j++){
trainingData[i][j] = scale(trainingData[i][j],ranges[j].minValue,ranges[j].maxValue,0,1.0);
}
}
}
//Train the KMeans model at each layer
const UINT K = (UINT)numClustersPerLayer.size();
for(UINT k=0; k<K; k++){
KMeans kmeans;
kmeans.setNumClusters( numClustersPerLayer[k] );
kmeans.setComputeTheta( true );
kmeans.setMinChange( minChange );
kmeans.setMinNumEpochs( minNumEpochs );
kmeans.setMaxNumEpochs( maxNumEpochs );
trainingLog << "Layer " << k+1 << "/" << K << " NumClusters: " << numClustersPerLayer[k] << endl;
if( !kmeans.train_( trainingData ) ){
errorLog << "train_(MatrixDouble &trainingData) - Failed to train kmeans model at layer: " << k << endl;
return false;
}
//Save the clusters
clusters.push_back( kmeans.getClusters() );
//Project the data through the current layer to use as training data for the next layer
if( k+1 != K ){
MatrixDouble data( M, numClustersPerLayer[k] );
VectorDouble input( trainingData.getNumCols() );
VectorDouble output( data.getNumCols() );
for(UINT i=0; i<M; i++){
//Copy the data into the sample
for(UINT j=0; j<input.size(); j++){
input[j] = trainingData[i][j];
}
//Project the sample through the current layer
if( !projectDataThroughLayer( input, output, k ) ){
errorLog << "train_(MatrixDouble &trainingData) - Failed to project sample through layer: " << k << endl;
return false;
}
//Copy the result into the training data for the next layer
for(UINT j=0; j<output.size(); j++){
data[i][j] = output[j];
}
}
//Swap the data for the next layer
trainingData = data;
}
}
//Flag that the kmeans model has been trained
trained = true;
featureVector.resize( numOutputDimensions, 0 );
return true;
}