RegressionTreeNode* RegressionTree::buildTree(const RegressionData &trainingData,RegressionTreeNode *parent,Vector< UINT > features,UINT nodeID){
const UINT M = trainingData.getNumSamples();
const UINT N = trainingData.getNumInputDimensions();
const UINT T = trainingData.getNumTargetDimensions();
VectorFloat regressionData(T);
//Update the nodeID
//Get the depth
UINT depth = 0;
if( parent != NULL )
depth = parent->getDepth() + 1;
//If there are no training data then return NULL
if( trainingData.getNumSamples() == 0 )
return NULL;
//Create the new node
RegressionTreeNode *node = new RegressionTreeNode;
if( node == NULL )
return NULL;
//Set the parent
node->initNode( parent, depth, nodeID );
//If there are no features left then create a leaf node and return
if( features.size() == 0 || M < minNumSamplesPerNode || depth >= maxDepth ){
//Flag that this is a leaf node
node->setIsLeafNode( true );
//Compute the regression data that will be stored at this node
computeNodeRegressionData( trainingData, regressionData );
//Set the node
node->set( trainingData.getNumSamples(), 0, 0, regressionData );
Regressifier::trainingLog << "Reached leaf node. Depth: " << depth << " NumSamples: " << trainingData.getNumSamples() << std::endl;
return node;
}
//Compute the best spilt point
UINT featureIndex = 0;
Float threshold = 0;
Float minError = 0;
if( !computeBestSpilt( trainingData, features, featureIndex, threshold, minError ) ){
delete node;
return NULL;
}
Regressifier::trainingLog << "Depth: " << depth << " FeatureIndex: " << featureIndex << " Threshold: " << threshold << " MinError: " << minError << std::endl;
//If the minError is below the minRMSError then create a leaf node and return
if( minError <= minRMSErrorPerNode ){
//Compute the regression data that will be stored at this node
computeNodeRegressionData( trainingData, regressionData );
//Set the node
node->set( trainingData.getNumSamples(), featureIndex, threshold, regressionData );
Regressifier::trainingLog << "Reached leaf node. Depth: " << depth << " NumSamples: " << M << std::endl;
return node;
}
//Set the node
node->set( trainingData.getNumSamples(), featureIndex, threshold, regressionData );
//Remove the selected feature so we will not use it again
if( removeFeaturesAtEachSpilt ){
for(UINT i=0; i<features.getSize(); i++){
if( features[i] == featureIndex ){
features.erase( features.begin()+i );
break;
}
}
}
//Split the data
RegressionData lhs(N,T);
RegressionData rhs(N,T);
for(UINT i=0; i<M; i++){
if( node->predict( trainingData[i].getInputVector() ) ){
rhs.addSample(trainingData[i].getInputVector(), trainingData[i].getTargetVector());
}else lhs.addSample(trainingData[i].getInputVector(), trainingData[i].getTargetVector());
}
//Run the recursive tree building on the children
node->setLeftChild( buildTree( lhs, node, features, nodeID ) );
node->setRightChild( buildTree( rhs, node, features, nodeID ) );
return node;
}
SEXP regression_Laplace(SEXP Rlocations, SEXP Robservations, SEXP Rmesh, SEXP Rorder, SEXP Rlambda,
SEXP Rcovariates, SEXP RBCIndices, SEXP RBCValues, SEXP DOF, SEXP RGCVmethod, SEXP Rnrealizations, SEXP RRNGstate, SEXP Rsolver, SEXP Rnprocessors, SEXP Rhosts)
{
//Set data
RegressionData regressionData(Rlocations, Robservations, Rorder, Rlambda, Rcovariates, RBCIndices, RBCValues, DOF, RGCVmethod, Rnrealizations, RRNGstate, Rsolver, Rnprocessors, Rhosts);
//std::cout<< "Data loaded"<<std::endl;
SEXP result = NILSXP;
if(regressionData.getOrder()==1)
{
MeshHandler<1> mesh(Rmesh);
//std::cout<< "Mesh loaded"<<std::endl;
MixedFERegression<RegressionData, IntegratorTriangleP2,1> regression(mesh,regressionData);
regression.smoothLaplace();
const std::vector<VectorXr>& solution = regression.getSolution();
const std::vector<Real>& dof = regression.getDOF();
const std::vector<Real>& var = regression.getVar();
//Copy result in R memory
result = PROTECT(Rf_allocVector(VECSXP, 4));
SET_VECTOR_ELT(result, 0, Rf_allocMatrix(REALSXP, solution[0].size(), solution.size()));
SET_VECTOR_ELT(result, 1, Rf_allocVector(REALSXP, solution.size()));
SET_VECTOR_ELT(result, 2, Rf_allocVector(REALSXP, solution.size()));
SET_VECTOR_ELT(result, 3, Rf_allocVector(STRSXP, 1));
Real *rans = REAL(VECTOR_ELT(result, 0));
for(UInt j = 0; j < solution.size(); j++)
{
for(UInt i = 0; i < solution[0].size(); i++)
rans[i + solution[0].size()*j] = solution[j][i];
}
Real *rans2 = REAL(VECTOR_ELT(result, 1));
for(UInt i = 0; i < solution.size(); i++)
{
rans2[i] = dof[i];
}
Real *rans3 = REAL(VECTOR_ELT(result, 2));
for(UInt i = 0; i < solution.size(); i++)
{
rans3[i] = var[i];
}
std::string RNGstate = regression.getFinalRNGstate();
SET_STRING_ELT(VECTOR_ELT(result, 3), 0, Rf_mkChar(RNGstate.c_str()));
UNPROTECT(1);
}
else if(regressionData.getOrder()==2)
{
MeshHandler<2> mesh(Rmesh);
//std::cout<< "Mesh loaded"<<std::endl;
MixedFERegression<RegressionData, IntegratorTriangleP4,2> regression(mesh,regressionData);
regression.smoothLaplace();
const std::vector<VectorXr>& solution = regression.getSolution();
const std::vector<Real>& dof = regression.getDOF();
const std::vector<Real>& var = regression.getVar();
//Copy result in R memory
result = PROTECT(Rf_allocVector(VECSXP, 4));
SET_VECTOR_ELT(result, 0, Rf_allocMatrix(REALSXP, solution[0].size(), solution.size()));
SET_VECTOR_ELT(result, 1, Rf_allocVector(REALSXP, solution.size()));
SET_VECTOR_ELT(result, 2, Rf_allocVector(REALSXP, solution.size()));
SET_VECTOR_ELT(result, 3, Rf_allocVector(STRSXP, 1));
Real *rans = REAL(VECTOR_ELT(result, 0));
for(UInt j = 0; j < solution.size(); j++)
{
for(UInt i = 0; i < solution[0].size(); i++)
rans[i + solution[0].size()*j] = solution[j][i];
}
Real *rans2 = REAL(VECTOR_ELT(result, 1));
for(UInt i = 0; i < solution.size(); i++)
{
rans2[i] = dof[i];
}
Real *rans3 = REAL(VECTOR_ELT(result, 2));
for(UInt i = 0; i < solution.size(); i++)
{
rans3[i] = var[i];
}
std::string RNGstate = regression.getFinalRNGstate();
SET_STRING_ELT(VECTOR_ELT(result, 3), 0, Rf_mkChar(RNGstate.c_str()));
UNPROTECT(1);
}
return(result);
}
SEXP regression_Laplace(SEXP Rlocations, SEXP Robservations, SEXP Rmesh, SEXP Rorder, SEXP Rlambda,
SEXP Rcovariates, SEXP RBCIndices, SEXP RBCValues, SEXP DOF)
{
//Set data
RegressionData regressionData(Rlocations, Robservations, Rorder, Rlambda, Rcovariates, RBCIndices, RBCValues, DOF);
//std::cout<< "Data loaded"<<std::endl;
SEXP result;
if(regressionData.getOrder()==1)
{
MeshHandler<1> mesh(Rmesh);
//std::cout<< "Mesh loaded"<<std::endl;
MixedFERegression<RegressionData, IntegratorTriangleP2,1> regression(mesh,regressionData);
regression.smoothLaplace();
const std::vector<VectorXr>& solution = regression.getSolution();
const std::vector<Real>& dof = regression.getDOF();
//Copy result in R memory
result = PROTECT(allocVector(VECSXP, 2));
SET_VECTOR_ELT(result, 0, allocMatrix(REALSXP, solution[0].size(), solution.size()));
SET_VECTOR_ELT(result, 1, allocVector(REALSXP, solution.size()));
Real *rans = REAL(VECTOR_ELT(result, 0));
for(int j = 0; j < solution.size(); j++)
{
for(int i = 0; i < solution[0].size(); i++)
rans[i + solution[0].size()*j] = solution[j][i];
}
Real *rans2 = REAL(VECTOR_ELT(result, 1));
for(int i = 0; i < solution.size(); i++)
{
rans2[i] = dof[i];
}
UNPROTECT(1);
}
else if(regressionData.getOrder()==2)
{
MeshHandler<2> mesh(Rmesh);
//std::cout<< "Mesh loaded"<<std::endl;
MixedFERegression<RegressionData, IntegratorTriangleP4,2> regression(mesh,regressionData);
regression.smoothLaplace();
const std::vector<VectorXr>& solution = regression.getSolution();
const std::vector<Real>& dof = regression.getDOF();
//Copy result in R memory
result = PROTECT(allocVector(VECSXP, 2));
SET_VECTOR_ELT(result, 0, allocMatrix(REALSXP, solution[0].size(), solution.size()));
SET_VECTOR_ELT(result, 1, allocVector(REALSXP, solution.size()));
Real *rans = REAL(VECTOR_ELT(result, 0));
for(int j = 0; j < solution.size(); j++)
{
for(int i = 0; i < solution[0].size(); i++)
rans[i + solution[0].size()*j] = solution[j][i];
}
Real *rans2 = REAL(VECTOR_ELT(result, 1));
for(int i = 0; i < solution.size(); i++)
{
rans2[i] = dof[i];
}
UNPROTECT(1);
}
return(result);
}
