org.ai_heuristic.algo.classify

Class InformationGain

java.lang.Object

org.ai_heuristic.functs.Function

org.ai_heuristic.algo.classify.InformationGain

All Implemented Interfaces:

FunctionDef

public class InformationGain
extends Function

This class implements the Information Gain evaluation. The algorithm determines if the entropy of one column is improved if it is split or re-ordered, based on the information re-grouping of another column. Data is entered in tabular format and it should be in the original form, of columns of Object-based values, typically String-based, but an appropriate evaluator can compare other types as well by also setting the valueType for the algorithm. The data column that the classification is being performed for is of type Integer and can be set and retrieved using AiHeuristicConst.COLINDEX as the key and retrieved from dataset.getParams().

str1a str2a str3a
str1b str2b str3a
str1a str2c str3b
str1b str2c str3a

The tabular dataset can be created by adding columns of type MetricValue to a MetricDataset and passing this as the parameter. For example:
ArrayList dataPoints = new ArrayList();
dataPoints.add("Sunny");
dataPoints.add("Sunny");
dataPoints.add("Overcast");
dataPoints.add("Rain");
dataPoints.add("Rain");
MetricValue metricValue = MetricValue.toValue("Outlook", valueType, dataPoints);
allColumns.add(metricValue);
...
//repeat for the other data columns
...
MetricDataset evalDataset = MetricDataset.toDataset(null, valueType, allColumns);
//add column to evaluate for
evalDataset.getParams().put(AiHeuristicConst.COLINDEX, new Integer(4));
...
//initialise and run the algorithm
InformationGain infoGain = new InformationGain(valueType);
ArrayList resultList = (ArrayList)infoGain.evaluate(evalDataset).getValue();

The algorithm automatically counts the number of similar entries for each column variable and uses this number to determine the best entropy result. The information gain ordering is then returned as a result in the form of a ArrayList of MetricValue, where the name is the column variable name and the value is the Double value evaluation. Algorithms in the ai_heuristic package try to maximise and so a larger final value means a better split.

Some help from the tutorial by Mohammad A Rahman at http://www.codeproject.com/Articles/259241/ID3-Decision-Tree-Algorithm-Part-1#

Author:

Kieran Greer

Field Summary

Fields inherited from class org.ai_heuristic.functs.Function

config, mathCompare, valueType

Constructor Summary

Constructors

Constructor and Description
InformationGain() Create a new instance of InformationGain.
InformationGain(java.lang.String thisValueType) Create a new instance of InformationGain.

Method Summary

Modifier and Type	Method and Description
ReplySet	evaluate(MetricDataset dataset) Evaluate the resulting IG after splitting the datasets over each attribute (column variable), apart from the decision attribute or column that the IG is being calculated for.
protected void	initialise() Initialise the function values, setting the config parameters or other.

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

Constructor Detail

InformationGain

public InformationGain()
                throws java.lang.Exception

Create a new instance of InformationGain.

Throws:

java.lang.Exception - any error.

InformationGain

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

Create a new instance of InformationGain.

Parameters:

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

Throws:

java.lang.Exception - any error.

Method Detail

initialise

protected void initialise()

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

Specified by:

initialise in class Function

evaluate

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

Evaluate the resulting IG after splitting the datasets over each attribute (column variable), apart from the decision attribute or column that the IG is being calculated for. The Entropy of each dataset in the list is calculated and used to order the results based on the information gain. Each non-decision column is considered in turn and the one that gives the best IG is returned as the result. The input should be the raw data table, where this method will count the number of times a String occurs, group with the same value in other rows and use this as the numerical count for that attribute value, instead of the actual String value.

Specified by:

evaluate in interface FunctionDef

Overrides:

evaluate in class Function

Parameters:

dataset - tabular list of values. This should store a list of MetricValue entities that each store a set of values for a single attribute column entry. Each MetricValue in the list should be assigned the entity name. Note that in addition to this, the config of the dataset needs to set the data column that the classification is being performed for. This is of type Integer and can be set and retrieved using AiHeuristicConst.COLINDEX as the key.

Returns:

the IG results in the table order as a ArrayList of MetricValue entities. Each entry is assigned the attribute entity name and the value is the information gain, as a Double. For the decision column, the value is null, with only the entity name.

Throws:

java.lang.Exception - any error.