com.mwt.explorers

Class VariableActionSoftmaxExplorer<T extends VariableActionContext>

java.lang.Object

com.mwt.explorers.SoftmaxExplorer<T>

com.mwt.explorers.VariableActionSoftmaxExplorer<T>

All Implemented Interfaces:

ConsumeScorer<T>, Explorer<T>

public class VariableActionSoftmaxExplorer<T extends VariableActionContext>
extends SoftmaxExplorer<T>

In some cases, different actions have a different scores, and you would prefer to choose actions with large scores. Softmax allows you to do that.

Constructor Summary

Constructors

Constructor and Description
VariableActionSoftmaxExplorer(Scorer<T> defaultScorer, float lambda) Constructor

Method Summary

Modifier and Type	Method and Description
protected int	getNumActions(T context)

Methods inherited from class com.mwt.explorers.SoftmaxExplorer

chooseAction, enableExplore, updateScorer

Methods inherited from class java.lang.Object

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

Constructor Detail

VariableActionSoftmaxExplorer

public VariableActionSoftmaxExplorer(Scorer<T> defaultScorer,
                                     float lambda)

Constructor

Parameters:

defaultScorer - A function which outputs a score for each action.

lambda - lambda = 0 implies uniform distribution. Large lambda is equivalent to a max.

Method Detail

getNumActions

protected int getNumActions(T context)

Overrides:

getNumActions in class SoftmaxExplorer<T extends VariableActionContext>