jsat.classifiers.trees

Class TreeNodeVisitor

java.lang.Object

jsat.classifiers.trees.TreeNodeVisitor

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable

public abstract class TreeNodeVisitor
extends java.lang.Object
implements java.io.Serializable, java.lang.Cloneable

Provides an abstracted mechanism for traversing and predicting from nodes in a tree meant for a supervised learning problem. Paths and children are used interchangeably, every node has one path to one child
Paths to children nodes can be disabled, but not removed. This is done so that the implementation does not have to worry about changes in the indices to children nodes, which would be complicated to implement. Once a path is disabled, it can not be re-enabled.
If all paths to any children have been disabled, childrenCount() may choose to return 0, otherwise - it must return the original number of paths to children nodes. The implementation for a given tree should override localClassify(jsat.classifiers.DataPoint) and localRegress(jsat.classifiers.DataPoint) if the operations are supported.

See Also:

TreeLearner, TreePruner, Serialized Form

Constructor Summary

Constructors

Constructor and Description
TreeNodeVisitor()

Method Summary

Modifier and Type	Method and Description
abstract int	childrenCount() Returns the number of children this node of the tree has, and may return a non zero value even if the node is a leaf
CategoricalResults	classify(DataPoint dp)
abstract TreeNodeVisitor	clone()
abstract void	disablePath(int child) Disables the selected path to the specified child from the current node.
abstract java.util.Collection<java.lang.Integer>	featuresUsed() Returns a collection of the indices of the features used by this node in the tree to make its decision about what branch to use next.
abstract TreeNodeVisitor	getChild(int child) Returns the node for the specific child, or null if the child index was not valid
abstract int	getPath(DataPoint dp) Returns the path down the tree the given data point would have taken, or -1 if this node was a leaf node OR if a missing value prevent traversal down the path
double	getPathWeight(int path) Returns the relative weight of each path, which should be an indication of how much of the training data went down each path.
abstract boolean	isLeaf() Returns true if the node is a leaf, meaning it has no valid paths to any children
abstract boolean	isPathDisabled(int child) Returns true if the path to the specified child is disabled, meaning it can not be traveled to.
CategoricalResults	localClassify(DataPoint dp) Returns the classification result that would have been obtained if the current node was a leaf node.
double	localRegress(DataPoint dp) Returns the regression result that would have been obtained if the current node was a leaf node.
double	regress(DataPoint dp) Performs regression on the given data point by following it down the tree until it finds the correct terminal node.
void	setPath(int child, TreeNodeVisitor node) Optional operation! This method, if supported, will set the path so that the child is set to the given value.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TreeNodeVisitor

public TreeNodeVisitor()

Method Detail

childrenCount

public abstract int childrenCount()

Returns the number of children this node of the tree has, and may return a non zero value even if the node is a leaf

Returns:

the number of children this node has

isLeaf

public abstract boolean isLeaf()

Returns true if the node is a leaf, meaning it has no valid paths to any children

Returns:

true if the node is purely a leaf node

getChild

public abstract TreeNodeVisitor getChild(int child)

Returns the node for the specific child, or null if the child index was not valid

Parameters:

child - the index of the child node to obtain

Returns:

the node for the child

disablePath

public abstract void disablePath(int child)

Disables the selected path to the specified child from the current node. All child indices will not be effected by this operation.

Parameters:

child - the index of the child to disable the path too

setPath

public void setPath(int child,
                    TreeNodeVisitor node)

Optional operation!
This method, if supported, will set the path so that the child is set to the given value.
The implementation may choose to throw an exception if the NodeVisitor is not of the same implementing class.

Parameters:

child - the child path

node - the node to make the child

isPathDisabled

public abstract boolean isPathDisabled(int child)

Returns true if the path to the specified child is disabled, meaning it can not be traveled to. It will also return true for an invalid child path, since a non existent node can not be reached.

Parameters:

child - the child index to check the path for

Returns:

true if the path is unavailable, false if the path is good.

localClassify

public CategoricalResults localClassify(DataPoint dp)

Returns the classification result that would have been obtained if the current node was a leaf node.

Parameters:

dp - the data point to localClassify

Returns:

the classification result

Throws:

java.lang.UnsupportedOperationException - if the tree node does not support or was not trained for classification

getPath

public abstract int getPath(DataPoint dp)

Returns the path down the tree the given data point would have taken, or -1 if this node was a leaf node OR if a missing value prevent traversal down the path

Parameters:

dp - the data point to send down the tree

Returns:

the path that would be taken

getPathWeight

public double getPathWeight(int path)

Returns the relative weight of each path, which should be an indication of how much of the training data went down each path. By default, returns 1.0/childrenCount(). The result should sum to one

Parameters:

path - the path to select

Returns:

the fraction of data estimated to travel the specified path, with respect to data that reaches this node.

classify

public CategoricalResults classify(DataPoint dp)

localRegress

public double localRegress(DataPoint dp)

Returns the regression result that would have been obtained if the current node was a leaf node.

Parameters:

dp - the data point to regress

Returns:

the classification result

Throws:

java.lang.UnsupportedOperationException - if the tree node does not support or was not trained for classification

regress

public double regress(DataPoint dp)

Performs regression on the given data point by following it down the tree until it finds the correct terminal node.

Parameters:

dp - the data point to regress

Returns:

the regression result from the tree starting from the current node

featuresUsed

public abstract java.util.Collection<java.lang.Integer> featuresUsed()

Returns a collection of the indices of the features used by this node in the tree to make its decision about what branch to use next. Numeric features start from zero, and categorical features start from the number of numeric features.

Returns:

the integers indicating which features were used for this node in the tree.

clone

public abstract TreeNodeVisitor clone()

Overrides:

clone in class java.lang.Object