public static ResultObject EvaluateGreedyDR_WekaClassifier(string algorithm, string trainingSetPath, string testSetPath, Dataset trainingSet, bool useAttributes, bool useInstances)
{
Classifier classifier = WekaNETBridge.WekaClassification.GetWekaClassifier(algorithm, trainingSetPath);
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
WekaClassificationQualityEvaluator evaluator = new WekaClassificationQualityEvaluator(wekaClassification, true);
DRLocalSearch localSearch = new DRLocalSearch(evaluator);
Problem <DRComponent> problem = new Problem <DRComponent>(null, null, evaluator, localSearch);
GreedyDR greedy = new GreedyDR(maxIterations, colonySize, convergenceIterations, problem, useAttributes, useInstances, trainingSet);
greedy.OnPostAntSolutionContruction += OnPostAntSolutionContruction;
greedy.OnPostColonyIteration += OnPostColonyIteration;
WekaNETBridge.WekaClassifier wekcaClassifier = greedy.CreateWekaClassifier();
double quality = WekaNETBridge.WekaClassification.EvaluateClassifier(wekcaClassifier, testSetPath);
double attribueReduction = greedy.BestAnt.Solution.FeatureCount() / (double)trainingSet.Metadata.Attributes.Length;
double instanceReduction = greedy.BestAnt.Solution.InstanceCount() / (double)trainingSet.Size;
ResultObject result = new ResultObject()
{
Quality = quality, AttributeReduction = attribueReduction, InstanceReduciton = instanceReduction
};
return(result);
}
public static Classifier GetWekaClassifier(string algorithm, string trainingSetPath)
{
Classifier classifier = null;
switch (algorithm)
{
case "KNN":
{
classifier = new weka.classifiers.lazy.IB1();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
case "NBayes":
{
classifier = new weka.classifiers.bayes.NaiveBayes();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
case "JRip":
{
classifier = new weka.classifiers.rules.JRip();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
case "J48":
{
classifier = new weka.classifiers.trees.J48();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
case "NeuralNets":
{
classifier = new weka.classifiers.functions.MultilayerPerceptron();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
case "SVM":
{
classifier = new weka.classifiers.functions.SMO();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
classifier = wekaClassification.CreateClassifier();
}
break;
}
return(classifier);
}
public static List <ResultObject> EvaluateACOMinerDR_WekaClassifier_Multi(string algorithm, string trainingSetPath, string testSetPath, Dataset trainingSet, bool useAttributes, bool useInstances)
{
Classifier classifier = WekaNETBridge.WekaClassification.GetWekaClassifier(algorithm, trainingSetPath);
DefaultDRHeuristicCalculator calculator = new DefaultDRHeuristicCalculator();
DRComponentInvalidator invalidator = new DRComponentInvalidator();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
WekaClassificationQualityEvaluator evaluator = new WekaClassificationQualityEvaluator(wekaClassification, !useAttributes);
DRLocalSearch localSearch = new DRLocalSearch(evaluator);
Problem <DRComponent> problem = new Problem <DRComponent>(invalidator, calculator, evaluator, localSearch);
ADRMiner acominer = new ADRMiner(maxIterations, colonySize, convergenceIterations, problem, useAttributes, useInstances, false, trainingSet);
acominer.OnPostAntSolutionContruction += OnPostAntSolutionContruction;
acominer.OnPostColonyIteration += OnPostColonyIteration;
acominer.Initialize();
acominer.Work();
List <ResultObject> results = new List <ResultObject>();
foreach (string algo in WekaNETBridge.WekaClassification.GetWekaAlgorithmNames())
{
Classifier currentClassifier = WekaNETBridge.WekaClassification.GetWekaClassifier(algo, trainingSetPath);
WekaNETBridge.WekaClassifier final = evaluator.CreateWekaClassifier(currentClassifier, acominer.BestAnt.Solution);
double quality = WekaNETBridge.WekaClassification.EvaluateClassifier(final, testSetPath);
double attribueReduction = acominer.BestAnt.Solution.FeatureCount() / (double)trainingSet.Metadata.Attributes.Length;
double instanceReduction = acominer.BestAnt.Solution.InstanceCount() / (double)trainingSet.Size;
ResultObject result = new ResultObject()
{
Quality = quality, AttributeReduction = attribueReduction, InstanceReduciton = instanceReduction
};
results.Add(result);
}
return(results);
}
public static ResultObject EvaluateRandDR_WekaClassifier(string algorithm, string trainingSetPath, string testSetPath, Dataset trainingSet, bool useAttributes, bool useInstances)
{
Classifier classifier = WekaNETBridge.WekaClassification.GetWekaClassifier(algorithm, trainingSetPath);
//DefaultDRHeuristicCalculator calculator = new DefaultDRHeuristicCalculator();
DefaultHeuristicCalculator <DRComponent> calculator = new DefaultHeuristicCalculator <DRComponent>(trainingSet);
DRComponentInvalidator invalidator = new DRComponentInvalidator();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
WekaClassificationQualityEvaluator evaluator = new WekaClassificationQualityEvaluator(wekaClassification, true);
DRLocalSearch localSearch = new DRLocalSearch(evaluator);
Problem <DRComponent> problem = new Problem <DRComponent>(invalidator, calculator, evaluator, localSearch);
RandomDR random = new RandomDR(maxIterations, colonySize, 50, problem, useAttributes, useInstances, false, trainingSet);
random.OnPostAntSolutionContruction += OnPostAntSolutionContruction;
random.OnPostColonyIteration += OnPostColonyIteration;
WekaNETBridge.WekaClassifier wekcaClassifier = random.CreateWekaClassifier();
double quality = WekaNETBridge.WekaClassification.EvaluateClassifier(wekcaClassifier, testSetPath);
double attribueReduction = random.BestAnt.Solution.FeatureCount() / (double)trainingSet.Metadata.Attributes.Length;
double instanceReduction = random.BestAnt.Solution.InstanceCount() / (double)trainingSet.Size;
ResultObject result = new ResultObject()
{
Quality = quality, AttributeReduction = attribueReduction, InstanceReduciton = instanceReduction
};
return(result);
}
public static ResultObject EvaluateACOMinerDR2_WekaClassifier(string algorithm, string trainingSetPath, string testSetPath, Dataset trainingSet, bool attributeFirst)
{
Classifier classifier = WekaNETBridge.WekaClassification.GetWekaClassifier(algorithm, trainingSetPath);
ADRHeuristicCalculator calculator = new ADRHeuristicCalculator();
DRComponentInvalidator invalidator = new DRComponentInvalidator();
WekaNETBridge.WekaClassification wekaClassification = new WekaNETBridge.WekaClassification(trainingSetPath, classifier);
WekaClassificationQualityEvaluator evaluator = new WekaClassificationQualityEvaluator(wekaClassification, false);
DRLocalSearch localSearch = new DRLocalSearch(evaluator);
Problem <DRComponent> problem = new Problem <DRComponent>(invalidator, calculator, evaluator, localSearch);
ADRMiner2 acominer = new ADRMiner2(maxIterations, colonySize, convergenceIterations, problem, attributeFirst, false, trainingSet);
acominer.ACO1.OnPostAntSolutionContruction += OnPostAntSolutionContruction;
acominer.ACO1.OnPostColonyIteration += OnPostColonyIteration;
acominer.ACO2.OnPostAntSolutionContruction += OnPostAntSolutionContruction;
acominer.ACO2.OnPostColonyIteration += OnPostColonyIteration;
WekaNETBridge.WekaClassifier wekcaClassifier = acominer.CreateWekaClassifier();
double quality = WekaNETBridge.WekaClassification.EvaluateClassifier(wekcaClassifier, testSetPath);
double attribueReduction = acominer.BestSolution.AttributesToRemove().Length;
double instanceReduction = acominer.BestSolution.InstanceCount() / (double)trainingSet.Size;
ResultObject result = new ResultObject()
{
Quality = quality, AttributeReduction = attribueReduction, InstanceReduciton = instanceReduction
};
return(result);
}
//public DataMining.Data.Dataset LearningSet
//{
// get { return this._learningSet; }
// set { this._learningSet = value; }
//}
//public DataMining.Data.Dataset ValidationSet
//{
// get { return this._validationSet; }
// set { this._validationSet = value; }
//}
//public DataMining.ClassificationMeasures.IClassificationMeasure Measure
//{
// get { return this._measure; }
//}
#endregion
#region Constructor
public WekaClassificationQualityEvaluator(WekaNETBridge.WekaClassification wekaClasification, bool skipAttributesValidation)
{
this._skipAttibutesValidation = skipAttributesValidation;
this.WekaClassification = wekaClasification;
}
