Class that does the actual training of a tree. More...
#include <random_tree.h>
Public Member Functions | |
RandomTreeTrain (int id, size_t numClasses, const TrainingConfiguration &configuration) | |
featureCount: K >= 1, the number of random splits to sample and evaluate at each interior node. | |
void | train (FeatureEvaluation &featureEvaluation, RandomSource &randomSource, const std::vector< std::pair< RandomTreePointer, Samples > > &samplesPerNode, int idNode, int currentLevel=1) const |
Train a single random tree breadth-first. |
Class that does the actual training of a tree.
Definition at line 1135 of file random_tree.h.
|
inline |
featureCount: K >= 1, the number of random splits to sample and evaluate at each interior node.
Note that K=1 yields completely randomized trees, whereas large values of K lead to aggressive optimization of the split. A common choice is round(sqrt(N)), where N is the number of sample instances in the training set passed to the Train method. minSampleCount: n_min, the minimum number of training samples necessary such that training is continued. maxTreeDepth: if < 0, no limit on the tree depth is assumed. If > 0, the tree learning is stopped when the given depth is reached.
Definition at line 1147 of file random_tree.h.