Package network.aika.neuron.activation.search

Class SearchNode

java.lang.Object

network.aika.neuron.activation.search.SearchNode

All Implemented Interfaces:

Comparable<SearchNode>

public class SearchNode
extends Object
implements Comparable<SearchNode>

The SearchNode class represents a node in the binary search tree that is used to find the optimal interpretation for a given document. Each search node possess a refinement (simply a set of interpretation nodes). The two options that this search node examines are that the refinement will either part of the final interpretation or not. During each search step the activation values in all the neuron activations adjusted such that they reflect the interpretation of the current search path. When the search reaches the maximum depth of the search tree and no further refinements exists, a weight is computed evaluating the current search path. The search path with the highest weight is used to determine the final interpretation.

Before the search is started a set of initial refinements is generated from the conflicts within the document. In other words, if there are no conflicts in a given document, then no search is needed. In this case the final interpretation will simply be the set of all interpretation nodes. The initial refinements are then expanded, meaning all interpretation nodes that are consistent with this refinement are added to the refinement. The initial refinements are then propagated along the search path as refinement candidates.

Author:

Lukas Molzberger

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	SearchNode.DebugState
static class	SearchNode.TimeoutException

Field Summary

Fields

Modifier and Type	Field	Description
static boolean	COMPUTE_SOFT_MAX
Branch	excluded
static int	MAX_SEARCH_STEPS
static boolean	OPTIMIZE_SEARCH
Branch	selected

Constructor Summary

Constructors

Constructor	Description
SearchNode(Document doc, Decision d, SearchNode p, int level)

Method Summary

Modifier and Type	Method	Description
void	changeState(Activation.Mode m)
int	compareTo(SearchNode sn)
static void	computeCachedFactor(SearchNode sn)
void	dumpDebugState()
boolean	followPath()
double	getAccumulatedWeight()
Activation	getActivation()
SearchNode	getAlternative()
Branch	getBranch(Decision d)
Decision	getDecision()
int	getId()
Map<Activation,Option>	getModifiedActivations()
double	getWeightSum()
static void	invalidateCachedDecision(Activation act)
static void	search(Document doc, SearchNode root, long v, Long timeoutInMilliSeconds)	Searches for the best interpretation for the given document.
void	setWeight(double w)
String	toString()
void	updateActivations(Document doc)

Methods inherited from class java.lang.Object

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

Field Detail

MAX_SEARCH_STEPS

public static int MAX_SEARCH_STEPS

OPTIMIZE_SEARCH

public static boolean OPTIMIZE_SEARCH

COMPUTE_SOFT_MAX

public static boolean COMPUTE_SOFT_MAX

selected

public Branch selected

excluded

public Branch excluded

Constructor Detail

SearchNode

public SearchNode(Document doc,
                  Decision d,
                  SearchNode p,
                  int level)

Method Detail

getBranch

public Branch getBranch(Decision d)

getAlternative

public SearchNode getAlternative()

updateActivations

public void updateActivations(Document doc)
                       throws Activation.OscillatingActivationsException

Throws:

Activation.OscillatingActivationsException

followPath

public boolean followPath()

getId

public int getId()

getModifiedActivations

public Map<Activation,Option> getModifiedActivations()

getAccumulatedWeight

public double getAccumulatedWeight()

getActivation

public Activation getActivation()

search

public static void search(Document doc,
                          SearchNode root,
                          long v,
                          Long timeoutInMilliSeconds)
                   throws SearchNode.TimeoutException,
                          Activation.RecursiveDepthExceededException,
                          Activation.OscillatingActivationsException

Searches for the best interpretation for the given document.

This implementation of the algorithm is iterative to prevent stack overflow errors from happening. Depending on the document the search tree might be getting very deep.

Parameters:

doc -

root -

Throws:

SearchNode.TimeoutException

Activation.RecursiveDepthExceededException

Activation.OscillatingActivationsException

setWeight

public void setWeight(double w)

getWeightSum

public double getWeightSum()

invalidateCachedDecision

public static void invalidateCachedDecision(Activation act)

computeCachedFactor

public static void computeCachedFactor(SearchNode sn)

changeState

public void changeState(Activation.Mode m)

compareTo

public int compareTo(SearchNode sn)

Specified by:

compareTo in interface Comparable<SearchNode>

getDecision

public Decision getDecision()

dumpDebugState

public void dumpDebugState()

toString

public String toString()

Overrides:

toString in class Object