bool RandomForests::train(LabelledClassificationData trainingData){
//Clear any previous model
clear();
const unsigned int M = trainingData.getNumSamples();
const unsigned int N = trainingData.getNumDimensions();
const unsigned int K = trainingData.getNumClasses();
if( M == 0 ){
errorLog << "train(LabelledClassificationData labelledTrainingData) - Training data has zero samples!" << endl;
return false;
}
numInputDimensions = N;
numClasses = K;
classLabels = trainingData.getClassLabels();
ranges = trainingData.getRanges();
//Scale the training data if needed
if( useScaling ){
//Scale the training data between 0 and 1
trainingData.scale(0, 1);
}
//Train the random forest
forestSize = 10;
Random random;
DecisionTree tree;
tree.enableScaling( false ); //We have already scaled the training data so we do not need to scale it again
tree.setTrainingMode( DecisionTree::BEST_RANDOM_SPLIT );
tree.setNumSplittingSteps( numRandomSplits );
tree.setMinNumSamplesPerNode( minNumSamplesPerNode );
tree.setMaxDepth( maxDepth );
for(UINT i=0; i<forestSize; i++){
LabelledClassificationData data = trainingData.getBootstrappedDataset();
if( !tree.train( data ) ){
errorLog << "train(LabelledClassificationData labelledTrainingData) - Failed to train tree at forest index: " << i << endl;
return false;
}
//Deep copy the tree into the forest
forest.push_back( tree.deepCopyTree() );
}
//Flag that the algorithm has been trained
trained = true;
return trained;
}
bool BAG::train(LabelledClassificationData trainingData){
const unsigned int M = trainingData.getNumSamples();
const unsigned int N = trainingData.getNumDimensions();
const unsigned int K = trainingData.getNumClasses();
trained = false;
classLabels.clear();
if( M == 0 ){
errorLog << "train(LabelledClassificationData trainingData) - Training data has zero samples!" << endl;
return false;
}
numFeatures = N;
numClasses = K;
classLabels.resize(K);
ranges = trainingData.getRanges();
UINT ensembleSize = (UINT)ensemble.size();
if( ensembleSize == 0 ){
errorLog << "train(LabelledClassificationData trainingData) - The ensemble size is zero! You need to add some classifiers to the ensemble first." << endl;
return false;
}
for(UINT i=0; i<ensembleSize; i++){
if( ensemble[i] == NULL ){
errorLog << "train(LabelledClassificationData trainingData) - The classifier at ensemble index " << i << " has not been set!" << endl;
return false;
}
}
//Train the ensemble
for(UINT i=0; i<ensembleSize; i++){
LabelledClassificationData boostedDataset = trainingData.getBootstrappedDataset();
//Train the classifier with the bootstrapped dataset
if( !ensemble[i]->train( boostedDataset ) ){
errorLog << "train(LabelledClassificationData trainingData) - The classifier at ensemble index " << i << " failed training!" << endl;
return false;
}
}
//Set the class labels
classLabels = trainingData.getClassLabels();
//Flag that the algorithm has been trained
trained = true;
return trained;
}
bool Softmax::train(LabelledClassificationData trainingData){
//Clear any previous model
clear();
const unsigned int M = trainingData.getNumSamples();
const unsigned int N = trainingData.getNumDimensions();
const unsigned int K = trainingData.getNumClasses();
if( M == 0 ){
errorLog << "train(LabelledClassificationData labelledTrainingData) - Training data has zero samples!" << endl;
return false;
}
numFeatures = N;
numClasses = K;
models.resize(K);
classLabels.resize(K);
ranges = trainingData.getRanges();
//Scale the training data if needed
if( useScaling ){
//Scale the training data between 0 and 1
trainingData.scale(0, 1);
}
//Train a regression model for each class in the training data
for(UINT k=0; k<numClasses; k++){
//Set the class label
classLabels[k] = trainingData.getClassTracker()[k].classLabel;
//Train the model
if( !trainSoftmaxModel(classLabels[k],models[k],trainingData) ){
errorLog << "train(LabelledClassificationData labelledTrainingData) - Failed to train model for class: " << classLabels[k] << endl;
return false;
}
}
//Flag that the algorithm has been trained
trained = true;
return trained;
}
bool KNN::train_(LabelledClassificationData &trainingData,UINT K){
//Clear any previous models
clear();
if( trainingData.getNumSamples() == 0 ){
errorLog << "train(LabelledClassificationData &trainingData) - Training data has zero samples!" << endl;
return false;
}
//Set the dimensionality of the input data
this->K = K;
this->numFeatures = trainingData.getNumDimensions();
this->numClasses = trainingData.getNumClasses();
//TODO: In the future need to build a kdtree from the training data to allow better realtime prediction
this->trainingData = trainingData;
if( useScaling ){
ranges = this->trainingData.getRanges();
this->trainingData.scale(ranges, 0, 1);
}
//Set the class labels
classLabels.resize(numClasses);
for(UINT k=0; k<numClasses; k++){
classLabels[k] = trainingData.getClassTracker()[k].classLabel;
}
//Flag that the algorithm has been trained so we can compute the rejection thresholds
trained = true;
//If null rejection is enabled then compute the null rejection thresholds
if( useNullRejection ){
//Set the null rejection to false so we can compute the values for it (this will be set back to its current value later)
bool tempUseNullRejection = useNullRejection;
useNullRejection = false;
rejectionThresholds.clear();
//Compute the rejection thresholds for each of the K classes
VectorDouble counter(numClasses,0);
trainingMu.resize( numClasses, 0 );
trainingSigma.resize( numClasses, 0 );
rejectionThresholds.resize( numClasses, 0 );
//Compute Mu for each of the classes
const unsigned int numTrainingExamples = trainingData.getNumSamples();
vector< IndexedDouble > predictionResults( numTrainingExamples );
for(UINT i=0; i<numTrainingExamples; i++){
predict( trainingData[i].getSample(), K);
UINT classLabelIndex = 0;
for(UINT k=0; k<numClasses; k++){
if( predictedClassLabel == classLabels[k] ){
classLabelIndex = k;
break;
}
}
predictionResults[ i ].index = classLabelIndex;
predictionResults[ i ].value = classDistances[ classLabelIndex ];
trainingMu[ classLabelIndex ] += predictionResults[ i ].value;
counter[ classLabelIndex ]++;
}
for(UINT j=0; j<numClasses; j++){
trainingMu[j] /= counter[j];
}
//Compute Sigma for each of the classes
for(UINT i=0; i<numTrainingExamples; i++){
trainingSigma[predictionResults[i].index] += SQR(predictionResults[i].value - trainingMu[predictionResults[i].index]);
}
for(UINT j=0; j<numClasses; j++){
double count = counter[j];
if( count > 1 ){
trainingSigma[ j ] = sqrt( trainingSigma[j] / (count-1) );
}else{
trainingSigma[ j ] = 1.0;
}
}
//Check to see if any of the mu or sigma values are zero or NaN
bool errorFound = false;
for(UINT j=0; j<numClasses; j++){
if( trainingMu[j] == 0 ){
warningLog << "TrainingMu[ " << j << " ] is zero for a K value of " << K << endl;
}
if( trainingSigma[j] == 0 ){
warningLog << "TrainingSigma[ " << j << " ] is zero for a K value of " << K << endl;
}
if( isnan( trainingMu[j] ) ){
errorLog << "TrainingMu[ " << j << " ] is NAN for a K value of " << K << endl;
errorFound = true;
}
if( isnan( trainingSigma[j] ) ){
errorLog << "TrainingSigma[ " << j << " ] is NAN for a K value of " << K << endl;
errorFound = true;
}
}
if( errorFound ){
trained = false;
return false;
}
//Recompute the rejection thresholds
recomputeNullRejectionThresholds();
//Restore the actual state of the null rejection
useNullRejection = tempUseNullRejection;
}else{
//Resize the rejection thresholds but set the values to 0
rejectionThresholds.clear();
rejectionThresholds.resize( numClasses, 0 );
}
return true;
}
bool DecisionStump::train(LabelledClassificationData &trainingData, VectorDouble &weights){
trained = false;
numInputDimensions = trainingData.getNumDimensions();
//There should only be two classes in the dataset, the positive class (classLable==1) and the negative class (classLabel==2)
if( trainingData.getNumClasses() != 2 ){
errorLog << "train(LabelledClassificationData &trainingData, VectorDouble &weights) - There should only be 2 classes in the training data, but there are : " << trainingData.getNumClasses() << endl;
return false;
}
//There should be one weight for every training sample
if( trainingData.getNumSamples() != weights.size() ){
errorLog << "train(LabelledClassificationData &trainingData, VectorDouble &weights) - There number of examples in the training data (" << trainingData.getNumSamples() << ") does not match the lenght of the weights vector (" << weights.size() << ")" << endl;
return false;
}
//Pick the training sample to use as the stump feature
const UINT M = trainingData.getNumSamples();
UINT bestFeatureIndex = 0;
vector< MinMax > ranges = trainingData.getRanges();
double minError = numeric_limits<double>::max();
double minRange = 0;
double maxRange = 0;
double step = 0;
double threshold = 0;
double bestThreshold = 0;
for(UINT n=0; n<numInputDimensions; n++){
minRange = ranges[n].minValue;
maxRange = ranges[n].maxValue;
step = (maxRange-minRange)/double(numSteps);
threshold = minRange;
while( threshold <= maxRange ){
//Compute the error using the current threshold on the current input dimension
//We need to check both sides of the threshold
double rhsError = 0;
double lhsError = 0;
for(UINT i=0; i<M; i++){
bool positiveClass = trainingData[ i ].getClassLabel() == WEAK_CLASSIFIER_POSITIVE_CLASS_LABEL;
bool rhs = trainingData[ i ][ n ] >= threshold;
bool lhs = trainingData[ i ][ n ] <= threshold;
if( (rhs && !positiveClass) || (!rhs && positiveClass) ) rhsError += weights[ i ];
if( (lhs && !positiveClass) || (!lhs && positiveClass) ) lhsError += weights[ i ];
}
//Check to see if either the rhsError or lhsError beats the minError, if so then store the results
if( rhsError < minError ){
minError = rhsError;
bestFeatureIndex = n;
bestThreshold = threshold;
direction = 1; //1 means rhs
}
if( lhsError < minError ){
minError = lhsError;
bestFeatureIndex = n;
bestThreshold = threshold;
direction = 0; //0 means lhs
}
//Update the threshold
threshold += step;
}
}
decisionFeatureIndex = bestFeatureIndex;
decisionValue = bestThreshold;
trained = true;
//cout << "Best Feature Index: " << decisionFeatureIndex << " Value: " << decisionValue << " Direction: " << direction << " Error: " << minError << endl;
return true;
}
bool ANBC::train(LabelledClassificationData &labelledTrainingData,double gamma) {
const unsigned int M = labelledTrainingData.getNumSamples();
const unsigned int N = labelledTrainingData.getNumDimensions();
const unsigned int K = labelledTrainingData.getNumClasses();
trained = false;
models.clear();
classLabels.clear();
if( M == 0 ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - Training data has zero samples!" << endl;
return false;
}
if( weightsDataSet ) {
if( weightsData.getNumDimensions() != N ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - The number of dimensions in the weights data (" << weightsData.getNumDimensions() << ") is not equal to the number of dimensions of the training data (" << N << ")" << endl;
return false;
}
}
numFeatures = N;
numClasses = K;
models.resize(K);
classLabels.resize(K);
ranges = labelledTrainingData.getRanges();
//Train each of the models
for(UINT k=0; k<numClasses; k++) {
//Get the class label for the kth class
UINT classLabel = labelledTrainingData.getClassTracker()[k].classLabel;
//Set the kth class label
classLabels[k] = classLabel;
//Get the weights for this class
VectorDouble weights(numFeatures);
if( weightsDataSet ) {
bool weightsFound = false;
for(UINT i=0; i<weightsData.getNumSamples(); i++) {
if( weightsData[i].getClassLabel() == classLabel ) {
weights = weightsData[i].getSample();
weightsFound = true;
break;
}
}
if( !weightsFound ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - Failed to find the weights for class " << classLabel << endl;
return false;
}
} else {
//If the weights data has not been set then all the weights are 1
for(UINT j=0; j<numFeatures; j++) weights[j] = 1.0;
}
//Get all the training data for this class
LabelledClassificationData classData = labelledTrainingData.getClassData(classLabel);
MatrixDouble data(classData.getNumSamples(),N);
//Copy the training data into a matrix, scaling the training data if needed
for(UINT i=0; i<data.getNumRows(); i++) {
for(UINT j=0; j<data.getNumCols(); j++) {
if( useScaling ) {
data[i][j] = scale(classData[i][j],ranges[j].minValue,ranges[j].maxValue,MIN_SCALE_VALUE,MAX_SCALE_VALUE);
} else data[i][j] = classData[i][j];
}
}
//Train the model for this class
models[k].gamma = gamma;
if( !models[k].train(classLabel,data,weights) ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - Failed to train model for class: " << classLabel << endl;
//Try and work out why the training failed
if( models[k].N == 0 ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - N == 0!" << endl;
models.clear();
return false;
}
for(UINT j=0; j<numFeatures; j++) {
if( models[k].mu[j] == 0 ) {
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - The mean of column " << j+1 << " is zero! Check the training data" << endl;
models.clear();
return false;
}
}
models.clear();
return false;
}
}
//Store the null rejection thresholds
nullRejectionThresholds.resize(numClasses);
for(UINT k=0; k<numClasses; k++) {
nullRejectionThresholds[k] = models[k].threshold;
}
//Flag that the models have been trained
trained = true;
return trained;
}
bool MinDist::train(LabelledClassificationData &labelledTrainingData,double gamma){
const unsigned int M = labelledTrainingData.getNumSamples();
const unsigned int N = labelledTrainingData.getNumDimensions();
const unsigned int K = labelledTrainingData.getNumClasses();
trained = false;
models.clear();
classLabels.clear();
if( M == 0 ){
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - Training data has zero samples!" << endl;
return false;
}
if( M <= numClusters ){
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - There are not enough training samples for the number of clusters. Either reduce the number of clusters or increase the number of training samples!" << endl;
return false;
}
numFeatures = N;
numClasses = K;
models.resize(K);
classLabels.resize(K);
ranges = labelledTrainingData.getRanges();
//Train each of the models
for(UINT k=0; k<numClasses; k++){
//Get the class label for the kth class
UINT classLabel = labelledTrainingData.getClassTracker()[k].classLabel;
//Set the kth class label
classLabels[k] = classLabel;
//Get all the training data for this class
LabelledClassificationData classData = labelledTrainingData.getClassData(classLabel);
MatrixDouble data(classData.getNumSamples(),N);
//Copy the training data into a matrix, scaling the training data if needed
for(UINT i=0; i<data.getNumRows(); i++){
for(UINT j=0; j<data.getNumCols(); j++){
if( useScaling ){
data[i][j] = scale(classData[i][j],ranges[j].minValue,ranges[j].maxValue,0,1);
}else data[i][j] = classData[i][j];
}
}
//Train the model for this class
models[k].setGamma( gamma );
if( !models[k].train(classLabel,data,numClusters) ){
errorLog << "train(LabelledClassificationData &labelledTrainingData,double gamma) - Failed to train model for class: " << classLabel;
errorLog << ". This is might be because this class does not have enough training samples! You should reduce the number of clusters or increase the number of training samples for this class." << endl;
models.clear();
return false;
}
}
trained = true;
return true;
}
bool GMM::train(LabelledClassificationData trainingData){
//Clear any old models
models.clear();
trained = false;
numFeatures = 0;
numClasses = 0;
if( trainingData.getNumSamples() == 0 ){
errorLog << "train(LabelledClassificationData &trainingData) - Training data is empty!" << endl;
return false;
}
//Set the number of features and number of classes and resize the models buffer
numFeatures = trainingData.getNumDimensions();
numClasses = trainingData.getNumClasses();
models.resize(numClasses);
if( numFeatures >= 6 ){
warningLog << "train(LabelledClassificationData &trainingData) - The number of features in your training data is high (" << numFeatures << "). The GMMClassifier does not work well with high dimensional data, you might get better results from one of the other classifiers." << endl;
}
//Get the ranges of the training data if the training data is going to be scaled
if( useScaling ){
ranges = trainingData.getRanges();
}
//Fit a Mixture Model to each class (independently)
for(UINT k=0; k<numClasses; k++){
UINT classLabel = trainingData.getClassTracker()[k].classLabel;
LabelledClassificationData classData = trainingData.getClassData( classLabel );
//Scale the training data if needed
if( useScaling ){
if( !classData.scale(ranges,GMM_MIN_SCALE_VALUE, GMM_MAX_SCALE_VALUE) ){
errorLog << "train(LabelledClassificationData &trainingData) - Failed to scale training data!" << endl;
return false;
}
}
//Convert the labelled data to unlabelled data
UnlabelledClassificationData unlabelledData = classData.reformatAsUnlabelledClassificationData();
//Train the Mixture Model for this class
GaussianMixtureModels gaussianMixtureModel;
gaussianMixtureModel.setMinChange( minChange );
gaussianMixtureModel.setMaxIter( maxIter );
if( !gaussianMixtureModel.train(unlabelledData, numMixtureModels) ){
errorLog << "train(LabelledClassificationData &trainingData) - Failed to train Mixture Model for class " << classLabel << endl;
return false;
}
//Setup the model container
models[k].resize( numMixtureModels );
models[k].setClassLabel( classLabel );
//Store the mixture model in the container
for(UINT j=0; j<numMixtureModels; j++){
models[k][j].mu = gaussianMixtureModel.getMu().getRowVector(j);
models[k][j].sigma = gaussianMixtureModel.getSigma()[j];
//Compute the determinant and invSigma for the realtime prediction
LUDecomposition ludcmp(models[k][j].sigma);
if( !ludcmp.inverse( models[k][j].invSigma ) ){
models.clear();
errorLog << "train(LabelledClassificationData &trainingData) - Failed to invert Matrix for class " << classLabel << "!" << endl;
return false;
}
models[k][j].det = ludcmp.det();
}
//Compute the normalize factor
models[k].recomputeNormalizationFactor();
//Compute the rejection thresholds
double mu = 0;
double sigma = 0;
VectorDouble predictionResults(classData.getNumSamples(),0);
for(UINT i=0; i<classData.getNumSamples(); i++){
vector< double > sample = classData[i].getSample();
predictionResults[i] = models[k].computeMixtureLikelihood( sample );
mu += predictionResults[i];
}
//Update mu
mu /= double( classData.getNumSamples() );
//Calculate the standard deviation
for(UINT i=0; i<classData.getNumSamples(); i++)
sigma += SQR( (predictionResults[i]-mu) );
sigma = sqrt( sigma / (double(classData.getNumSamples())-1.0) );
sigma = 0.2;
//Set the models training mu and sigma
models[k].setTrainingMuAndSigma(mu,sigma);
if( !models[k].recomputeNullRejectionThreshold(nullRejectionCoeff) && useNullRejection ){
warningLog << "train(LabelledClassificationData &trainingData) - Failed to recompute rejection threshold for class " << classLabel << " - the nullRjectionCoeff value is too high!" << endl;
}
//cout << "Training Mu: " << mu << " TrainingSigma: " << sigma << " RejectionThreshold: " << models[k].getNullRejectionThreshold() << endl;
//models[k].printModelValues();
}
//Reset the class labels
classLabels.resize(numClasses);
for(UINT k=0; k<numClasses; k++){
classLabels[k] = models[k].getClassLabel();
}
//Resize the rejection thresholds
nullRejectionThresholds.resize(numClasses);
for(UINT k=0; k<numClasses; k++){
nullRejectionThresholds[k] = models[k].getNullRejectionThreshold();
}
//Flag that the models have been trained
trained = true;
return true;
}