org.ai_heuristic.algo.classify

Class Markov

java.lang.Object

org.ai_heuristic.functs.Function

org.ai_heuristic.algo.AlgorithmLearn

org.ai_heuristic.algo.classify.Markov

All Implemented Interfaces:

FunctionDef

public class Markov
extends AlgorithmLearn

A very basic Markov model that counts the number of occurrences in the training datasets. This algorithm calculates the most likely next term based on the state graphs. Parameters can be set to allow you to define a AiHeuristicConst.BREADTH and DEPTH, to return a tree structure that retrieves the n most probable next states and repeats this d times, where each next state is determined only by the previous one. Note that this should be passed as part of the dataset, not as part of the initialisation.

Note that the model is stored as a Tree and so you should be able to save it to a file and then read it in again. The graph is simply a count of the frequency of states that follow other states.

Some help with ideas from http://herselfsai.com/2007/02/hidden-markov-models.html.

Author:

Kieran greer.

Field Summary

Fields

Modifier and Type	Field and Description
protected int	breadth Breadth to search to
protected int	depth Depth to search to
protected Tree	markovModel The Markov model

Fields inherited from class org.ai_heuristic.algo.AlgorithmLearn

evalMetric

Fields inherited from class org.ai_heuristic.functs.Function

config, mathCompare, valueType

Constructor Summary

Constructors

Constructor and Description
Markov() Create a new instance of Markov.
Markov(java.lang.String thisValueType) Create a new instance of Markov.
Markov(java.lang.String thisValueType, FunctionLearn thisEvalMetric) Create a new instance of Markov.

Method Summary

Modifier and Type	Method and Description
ReplySet	evaluate(MetricDataset dataset) Evaluate the dataset and return the result.
<T> ReplySet	evaluate(T dataset) Return a value based on the function evaluation.
Tree	getModel() Get the markov model that is learned.
protected java.util.ArrayList	getNextStates(TreeNode thisNode, int thisDepth) Evaluate the dataset and return the result.
protected void	initialise() Initialise the function values, setting the config parameters or other.
void	setModel(Tree theMarkovModel) Set the model to an already existing one.
ReplySet	train(MetricDataset dataset) Train the function based on the supervised training method, using the dataset as input.
void	trainText(java.lang.String thisDocument) Train the function based on the contents of the text document.

Methods inherited from class org.ai_heuristic.functs.Function

checkValueType, createFunction, createFunction, createFunction, getConfigParams, innerObject, isLegalNumber, setConfigParams, setEvaluator, setValueType

Methods inherited from class java.lang.Object

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

Field Detail

breadth

protected int breadth

Breadth to search to

depth

protected int depth

Depth to search to

markovModel

protected Tree markovModel

The Markov model

Constructor Detail

Markov

public Markov()
       throws java.lang.Exception

Create a new instance of Markov. The learning function evalMetric is set to the default 'count of number of occurrences' FunctionCount and the mathematical comparisons is set to the default SimpleMathCompare one.

Throws:

java.lang.Exception - any error.

Markov

public Markov(java.lang.String thisValueType)
       throws java.lang.Exception

Create a new instance of Markov. This can be used if the data type is known by the function and the processing is automatic. The learning function is set to the default 'count of number of occurrences' FunctionCount.

Parameters:

thisValueType - the type of object being evaluated. Can be null if set later or not used.

Throws:

java.lang.Exception - any error.

Markov

public Markov(java.lang.String thisValueType,
              FunctionLearn thisEvalMetric)
       throws java.lang.Exception

Create a new instance of Markov. This can be used if the data type is known by the function and the processing is automatic.

Parameters:

thisValueType - the type of object being evaluated. Can be null if set later or not used.

thisEvalMetric - the learning update function.

Throws:

java.lang.Exception - any error.

Method Detail

initialise

protected void initialise()
                   throws java.lang.Exception

Initialise the function values, setting the config parameters or other.

Specified by:

initialise in class Function

Throws:

java.lang.Exception - any error.

train

public ReplySet train(MetricDataset dataset)
               throws java.lang.Exception

Train the function based on the supervised training method, using the dataset as input.

Specified by:

train in class AlgorithmLearn

Parameters:

dataset - the dataset, or any ArrayList list of dataset, etc. These can be in the form of String-based lists (text documents) In that case the document gets parsed into individual words or terms. Alternatively, a Tree, or ArrayList of Trees can be submitted and the links between the graph nodes used instead.

Returns:

the training result, true in this case. You can use ((Boolean)MetricValue.getValue()).booleanValue() to retrieve the value.

Throws:

java.lang.Exception - any error.

trainText

public void trainText(java.lang.String thisDocument)
               throws java.lang.Exception

Train the function based on the contents of the text document.

Parameters:

thisDocument - the next document to train with. This is tokenized first.

Throws:

java.lang.Exception - any error.

evaluate

public <T> ReplySet evaluate(T dataset)
                      throws java.lang.Exception

Return a value based on the function evaluation.

Parameters:

dataset - the value to pass through the function. This can be, for example, a single String value representing a single state, or a MetricDataset. If the MetricDataset is used, it should contain a single MetricValue, where only the value (objValue) is used. You can then add additional parameters, such as breadth or depth, to return a TreeNode structure that includes more than one linked state.

Returns:

the result as a TreeNode representing a section of a state-graph. The node value is the next state and the count is the evaluation value associated with it.

Throws:

java.lang.Exception - any error.

evaluate

public ReplySet evaluate(MetricDataset dataset)
                  throws java.lang.Exception

Evaluate the dataset and return the result. If the list has only one object, then apply to the object directly and not the list.

Specified by:

evaluate in interface FunctionDef

Overrides:

evaluate in class Function

Parameters:

dataset - the dataset to evaluate.

Returns:

the result of applying the operator to the dataset values, as a TreeNode.

Throws:

java.lang.Exception - any error.

getNextStates

protected java.util.ArrayList getNextStates(TreeNode thisNode,
                                            int thisDepth)
                                     throws java.lang.Exception

Evaluate the dataset and return the result. If the list has only one object, then apply to the object directly and not the list.

Parameters:

thisNode - the node to search from.

thisDepth - the current search depth.

Returns:

a list of next best states, as defined by the breadth parameter.

Throws:

java.lang.Exception - any error.

setModel

public void setModel(Tree theMarkovModel)

Set the model to an already existing one.

Parameters:

theMarkovModel - the markov model.

getModel

public Tree getModel()

Get the markov model that is learned.

Returns:

the value of markovModel.