Tree Class Reference

A single decision tree. More...

#include <tree.h>

List of all members.

Public Member Functions
	Tree (istream &in)
	Construct a new tree by loading it from a file.
	Tree (const InstanceSet &set, weight_list *weights, int K, uchar max_depth=16, int min_size=1, float min_gain=0, unsigned int seed=0)
	Construct a new tree by training.
int	predict (const Instance &c) const
	predict a new instance
int	predict (const InstanceSet &set, int instance_no) const
	predict an instance from a set
float	training_accuracy () const
	Return the accuracy for the training set.
float	testing_accuracy (const InstanceSet &testset) const
float	oob_accuracy () const
void	oob_cases (weight_list *counts, weight_list *correct) const
void	variable_importance (map< int, float > *scores, unsigned int *seed) const
void	print () const
void	grow ()
void	write (ostream &o) const
void	read (istream &i)

---

Detailed Description

Strategy:

• Binary tree in a fixed size array
  • #leaves is bounded by #instances!
• Sorted instances kept in an special matrix such that:
  • Each node has a contiguous group of rows in the matrix
  • Each column gives the sorted instance order with respect to a feature

Trees can only be created in two ways:

1. load from a saved model
2. grown from a certain bagging of a dataset

---

Constructor & Destructor Documentation

Tree (  const InstanceSet &  set, weight_list *  weights, int  K, uchar  max_depth = 16, int  min_size = 1, float  min_gain = 0, unsigned int  seed = 0 )

Constructor Parameters: set  training set weights  weights --- presumably from bagging process K  number of vars to try per split max_depth  (to be deprecated) min_size  minimum number of instances in a node min_gain  minimum information gain for making a split seed  random seed

---

Member Function Documentation

void grow (   )

Do the work of growing the tree Copy the data into special matrix Build the tree Delete special matrix

void read (  istream &  in  )

Read the tree from disk

void write (  ostream &  o  )  const

Save a tree to disk Important things to record: Nodes (all active nodes need to be written) WeightList ? - this seems irrelevant without the instance set Statistics?

---

The documentation for this class was generated from the following files:

• tree.h
• tree.cc

---