void Forest::TrainLarge(ClassificationData& trainingData, bool allNodesStoreLabelDistribution, bool refineWithAllTrainingData, int verbosityLevel)
{
// get available labels from training data
labels = trainingData.GetAvailableLabels();
if(verbosityLevel > 0)
{
std::cout << "Train forest with " << nTrees << " trees..." << std::endl;
if(verbosityLevel > 1)
{
// List all labels the forest is trained for
std::cout << "Used label IDs:" << std::endl;
for(unsigned int i=0; i<labels.size(); ++i)
{
std::cout << labels[i] << std::endl;
}
}
}
// train every tree independently
for(int i=0; i < nTrees; ++i)
{
if(verbosityLevel > 0)
{
std::cout << "Tree " << i+1 << "/" << nTrees << std::endl;
}
if(verbosityLevel > 1)
{
std::cout << "- Bag training data..." << std::endl;
}
// storage for the indices of the used data points for each tree (bagging)
std::vector<unsigned int> dataPointIndices = trainingData.NewBag(baggingRatio);
if(verbosityLevel > 1)
{
std::cout << "- Train tree with " << dataPointIndices.size() << " datapoints..." << std::endl;
}
// create and train tree
Tree t(&randomGenerator);
t.TrainParallel(trainingData, dataPointIndices, maxDepth, testedSplittingFunctions, minInformationGain, minPointsForSplit, allNodesStoreLabelDistribution, verbosityLevel);
trees.push_back(t);
}
if(refineWithAllTrainingData)
{
if(verbosityLevel > 0)
{
std::cout << "Refine all trees with all available training data..." << std::endl;
}
RefineLeafNodes(trainingData, verbosityLevel);
}
splitNodesStoreLabelDistribution = allNodesStoreLabelDistribution;
if(verbosityLevel > 0)
{
std::cout << "### TRAINING DONE ###" << std::endl;
}
}
void Forest::Train(ClassificationData& trainingData, int verbosityLevel)
{
// get available labels from training data
labels = trainingData.GetAvailableLabels();
if(verbosityLevel > 0)
{
std::cout << "Train forest with " << nTrees << " trees..." << std::endl;
if(verbosityLevel > 1)
{
// List all labels the forest is trained for
std::cout << "Used label IDs:" << std::endl;
for(unsigned int i=0; i<labels.size(); ++i)
{
std::cout << labels[i] << std::endl;
}
}
}
// create trees
for(int i=0; i<nTrees; i++)
trees.push_back(Tree(&randomGenerator));
// random generator for bagging of training data
boost::uniform_int<int> intDist(0, trainingData.GetCount()-1);
// how many data points for every tree?
int nDataPoints = floor(trainingData.GetCount() * baggingRatio);
// train every tree independently
#pragma omp parallel for
for(int i=0; i < nTrees; i++)
{
// storage for the indices of the used data points for each tree (bagging)
std::vector<unsigned int> dataPointIndices;
#pragma omp critical
{
if(verbosityLevel > 0)
{
std::cout << "Tree " << i+1 << "/" << nTrees << std::endl;
}
if(verbosityLevel > 1)
{
std::cout << "- Bag training data..." << std::endl;
std::cout << "- Train tree with " << nDataPoints << " datapoints..." << std::endl;
}
}
// refill index array for tree
dataPointIndices.clear();
// randomly select training points for each tree (bagging)
for(int j=0; j < nDataPoints; j++)
dataPointIndices.push_back(intDist(randomGenerator));
trees[i].Train(trainingData, dataPointIndices, maxDepth, testedSplittingFunctions, minInformationGain, minPointsForSplit, verbosityLevel);
}
if(verbosityLevel > 0)
{
std::cout << "### TRAINING DONE ###" << std::endl;
}
}