CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter > Class Template Reference

Class used in building regression trees. More...

#include <probabilisticregressiontreenode.h>

Collaboration diagram for CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >:

[legend]


Public Member Functions
	BinaryProbabilisticRegressionTreeNode (int NodeId, int CsDim, T_Regressor &regressor, T_Distribution *ParentDistribution=0)
	Constructor for a leaf.
	BinaryProbabilisticRegressionTreeNode (int NodeId, const Vector< int > &DDomainSize, int CsplitDim, int RegDim, T_Regressor &regressor, T_Distribution *ParentDistribution=0)
	Constructor for an intermediate node.
	~BinaryProbabilisticRegressionTreeNode (void)
int	GetNodeId (void)
void	ComputeSizesTree (int &nodes, int &term_nodes)
void	StartLearningEpoch (void)
void	LearnSample (const int Dvars, const double Cvars, double probability, double threshold)
bool	StopLearningEpoch (int splitType, int emRestarts, int emMaxIterations, double convergenceLim, int min_no_datapoints)
	Returns true if more learning has to be donne in future.
double	Infer (const int Dvars, const double Cvars, int maxNodeId, double threshold)
	Does the inference.
void	InitializePruningStatistics (void)
void	UpdatePruningStatistics (const int Dvars, const double Cvars, double y, double probability)
void	FinalizePruningStatistics (void)
double	PruneSubtree (void)
	Returns the optimal cost for this subtree and cuts the subtree to optimal size.
void	SaveToStream (ostream &out)
Protected Types
enum	state { stable, em, split, regression }
	the state of the node. At creation em. At load stable More...
Protected Member Functions
void	RandomDistributions (void)
double	EMStep (double &Likelihood)
	Perform an EM step on the data in buff and return the convergence factor and compute the likelihood.
Protected Attributes
int	nodeId
	unique identifier of the cluster for a regression tree.
int	emTrials
	how many times EM was attempted
int	csDim
	the number of continuous split variables
int	regDim
	the number of regressors
enum CLUS::BinaryProbabilisticRegressionTreeNode::state	State
	the state of the node. At creation em. At load stable
BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter > *	Children [2]
	the children of this node
T_Splitter	Splitter
	splitter for split criterion
T_Regressor	Regressor
	regressor for the node
T_Distribution *	Distributions [2]
	distributions for the children for the learning process
T_Distribution *	parentDistribution
	the distribution of the parent. Usefull in the learning process to pick initial values for the new distributions
DynamicBuffer *	buffer
	keep the EM samples here temporarily
double	pruningCost
	sum of squared differences between prediction and true value
int	pruningSamples
	number of samples in this node for pruning

Detailed Description

template<class T_Distribution, class T_Regressor, class T_Splitter>
class CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >

Class used in building regression trees.

Every class is a node in the tree, there can be 0 or 2 children NodeId determines the position of the node in the tree. The nodeId od children is alwais 2*nodeId and 2*nodeId+1. The nodeId starts from 1 for the root.

T_Distribution is the distribution that approximates the data T_Regressor is the regressor that commes with T_Distribution T_Spliter is the class that in a final scan can compute the splitting predicate

Definition at line 60 of file probabilisticregressiontreenode.h.

Member Enumeration Documentation

template<class T_Distribution, class T_Regressor, class T_Splitter>

enum CLUS::BinaryProbabilisticRegressionTreeNode::state [protected]

the state of the node. At creation em. At load stable

Enumeration values:

stable

em

split

regression

Definition at line 77 of file probabilisticregressiontreenode.h.

Constructor & Destructor Documentation

template<class T_Distribution, class T_Regressor, class T_Splitter>

CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode ( int NodeId,

int CsDim,

T_Regressor & regressor,

T_Distribution * ParentDistribution = 0

) [inline]

Constructor for a leaf.

Definition at line 146 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode ( int NodeId,

const Vector< int > & DDomainSize,

int CsplitDim,

int RegDim,

T_Regressor & regressor,

T_Distribution * ParentDistribution = 0

) [inline]

Constructor for an intermediate node.

Definition at line 157 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::~BinaryProbabilisticRegressionTreeNode ( void ) [inline]

Definition at line 175 of file probabilisticregressiontreenode.h.

Member Function Documentation

template<class T_Distribution, class T_Regressor, class T_Splitter>

double CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::EMStep ( double & Likelihood ) [inline, protected]

Perform an EM step on the data in buff and return the convergence factor and compute the likelihood.

Definition at line 122 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::GetNodeId ( void ) [inline]

Definition at line 192 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

double CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::Infer ( const int * Dvars,

const double * Cvars,

int maxNodeId,

double threshold

) [inline]

Does the inference.

Definition at line 482 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

void CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample ( const int * Dvars,

const double * Cvars,

double probability,

double threshold

) [inline]

Todo:
put some cutoff here to use less data as you go down

Definition at line 236 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

double CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::PruneSubtree ( void ) [inline]

Returns the optimal cost for this subtree and cuts the subtree to optimal size.

Definition at line 531 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

void CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::RandomDistributions ( void ) [inline, protected]

Definition at line 103 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

bool CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch ( int splitType,

int emRestarts,

int emMaxIterations,

double convergenceLim,

int min_no_datapoints

) [inline]

Returns true if more learning has to be donne in future.

Definition at line 293 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

void CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::UpdatePruningStatistics ( const int * Dvars,

const double * Cvars,

double y,

double probability

) [inline]

Definition at line 510 of file probabilisticregressiontreenode.h.

Field Documentation

template<class T_Distribution, class T_Regressor, class T_Splitter>

DynamicBuffer* CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::buffer [protected]

keep the EM samples here temporarily

Definition at line 95 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::EMStep(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StartLearningEpoch(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::csDim [protected]

the number of continuous split variables

Definition at line 71 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::Infer(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::UpdatePruningStatistics().

template<class T_Distribution, class T_Regressor, class T_Splitter>

T_Distribution* CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::Distributions[2] [protected]

distributions for the children for the learning process

Definition at line 89 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::EMStep(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::RandomDistributions(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::emTrials [protected]

how many times EM was attempted

Definition at line 68 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::nodeId [protected]

unique identifier of the cluster for a regression tree.
the id of the children is always 2*Id, 2*Id+1
Definition at line 65 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::GetNodeId(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::Infer(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::PruneSubtree(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::SaveToStream(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

T_Distribution* CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::parentDistribution [protected]

the distribution of the parent. Usefull in the learning process to pick initial values for the new distributions

Definition at line 92 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

double CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::pruningCost [protected]

sum of squared differences between prediction and true value

Definition at line 98 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::InitializePruningStatistics(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::PruneSubtree(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::UpdatePruningStatistics().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::pruningSamples [protected]

number of samples in this node for pruning

Definition at line 101 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::InitializePruningStatistics(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::UpdatePruningStatistics().

template<class T_Distribution, class T_Regressor, class T_Splitter>

int CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::regDim [protected]

the number of regressors

Definition at line 74 of file probabilisticregressiontreenode.h.
Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::EMStep(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StartLearningEpoch().

template<class T_Distribution, class T_Regressor, class T_Splitter>

T_Regressor CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::Regressor [protected]

regressor for the node

Definition at line 86 of file probabilisticregressiontreenode.h.

template<class T_Distribution, class T_Regressor, class T_Splitter>

enum CLUS::BinaryProbabilisticRegressionTreeNode::state CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::State [protected]

the state of the node. At creation em. At load stable

Referenced by CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::BinaryProbabilisticRegressionTreeNode(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::LearnSample(), CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StartLearningEpoch(), and CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter >::StopLearningEpoch().

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

probabilisticregressiontreenode.h

Generated on Mon Jul 21 16:57:46 2003 for SECRET by

1.3.2

CLUS::BinaryProbabilisticRegressionTreeNode< T_Distribution, T_Regressor, T_Splitter > Class Template Reference

Public Member Functions

Protected Types

Protected Member Functions