hex.tree.uplift

Class Divergence

java.lang.Object

water.Iced<Divergence>

hex.tree.uplift.Divergence

All Implemented Interfaces:

java.io.Externalizable, java.io.Serializable, java.lang.Cloneable, water.Freezable<Divergence>

Direct Known Subclasses:

EuclideanDistance, KLDivergence

public abstract class Divergence
extends water.Iced<Divergence>

Divergence class used to calculate gain to split the node in Uplift trees algorithms. Currently only UpliftRandomForest uses this class. Source: https://link.springer.com/content/pdf/10.1007/s10115-011-0434-0.pdf page 308

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static double	ZERO_TO_DIVIDE

Constructor Summary

Constructors

Constructor and Description
Divergence()

Method Summary

Modifier and Type	Method and Description
double	gain(double prY1CT1, double prY1CT0, double prL, double prLY1CT1, double prLY1CT0, double prR, double prRY1CT1, double prRY1CT0) Calculate overall gain as divergence between split gain and node gain.
abstract double	metric(double prCT1, double prCT0) Calculate distance divergence metric between two probabilities.
double	node(double prCT1, double prCT0) Calculate distance metric between two probabilities in the node.
abstract double	norm(double prCT1, double prCT0, double prLCT1, double prLCT0) Calculate normalization factor to normalize gain.
double	split(double prL, double prLY1CT1, double prLY1CT0, double prR, double prRY1CT1, double prRY1CT0) Calculate gain after split
double	value(double prY1CT1, double prY1CT0, double prL, double prLY1CT1, double prLY1CT0, double prR, double prRY1CT1, double prRY1CT0, double prCT1, double prCT0, double prLCT1, double prLCT0) Calculate normalized gain as result value to select best split.

Methods inherited from class water.Iced

asBytes, clone, copyOver, frozenType, read, readExternal, readJSON, reloadFromBytes, toJsonBytes, toJsonString, write, writeExternal, writeJSON

Methods inherited from class java.lang.Object

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

Field Detail

ZERO_TO_DIVIDE

public static double ZERO_TO_DIVIDE

Constructor Detail

Divergence

public Divergence()

Method Detail

metric

public abstract double metric(double prCT1,
                              double prCT0)

Calculate distance divergence metric between two probabilities.

Parameters:

prCT1 -

prCT0 -

Returns:

distance divergence metric

node

public double node(double prCT1,
                   double prCT0)

Calculate distance metric between two probabilities in the node.

Parameters:

prCT1 - probability of treatment group

prCT0 - probability of control group

Returns:

distance divergence metric in the node

split

public double split(double prL,
                    double prLY1CT1,
                    double prLY1CT0,
                    double prR,
                    double prRY1CT1,
                    double prRY1CT0)

Calculate gain after split

Parameters:

prL - probability of response in left node

prLY1CT1 - probability of response = 1 in treatment group in left node

prLY1CT0 - probability of response = 1 in control group in left node

prR - probability of response in right node

prRY1CT1 - probability of response = 1 in treatment group in right node

prRY1CT0 - probability of response = 1 in control group in right node

Returns:

gain after split

gain

public double gain(double prY1CT1,
                   double prY1CT0,
                   double prL,
                   double prLY1CT1,
                   double prLY1CT0,
                   double prR,
                   double prRY1CT1,
                   double prRY1CT0)

Calculate overall gain as divergence between split gain and node gain.

Parameters:

prY1CT1 - probability of response = 1 in treatment group before split

prY1CT0 - probability of response = 1 in control group before

prL - probability of response in left node

prLY1CT1 - probability of response = 1 in treatment group in left node

prLY1CT0 - probability of response = 1 in control group in left node

prR - probability of response in right node

prRY1CT1 - probability of response = 1 in treatment group in right node

prRY1CT0 - probability of response = 1 in control group in right node

Returns:

overall gain

norm

public abstract double norm(double prCT1,
                            double prCT0,
                            double prLCT1,
                            double prLCT0)

Calculate normalization factor to normalize gain.

Parameters:

prCT1 - probability of treatment group

prCT0 - probability of control group

prLCT1 - probability of treatment group in left node

prLCT0 - probability of control group in left node

Returns:

normalization factor

value

public double value(double prY1CT1,
                    double prY1CT0,
                    double prL,
                    double prLY1CT1,
                    double prLY1CT0,
                    double prR,
                    double prRY1CT1,
                    double prRY1CT0,
                    double prCT1,
                    double prCT0,
                    double prLCT1,
                    double prLCT0)

Calculate normalized gain as result value to select best split.

Parameters:

prY1CT1 - probability of response = 1 in treatment group before split

prY1CT0 - probability of response = 1 in control group before

prL - probability of response in left node

prLY1CT1 - probability of response = 1 in treatment group in left node

prLY1CT0 - probability of response = 1 in control group in left node

prR - probability of response in right node

prRY1CT1 - probability of response = 1 in treatment group in right node

prRY1CT0 - probability of response = 1 in control group in right node

prCT1 - probability of treatment group

prCT0 - probability of control group

prLCT1 - probability of treatment group in left node

prLCT0 - probability of control group in left node

Returns:

normalized gain