public IEnumerable <IEmergingPattern> Mine(InstanceModel model, IEnumerable <Instance> instances, Feature classFeature)
{
EmergingPatternCreator EpCreator = new EmergingPatternCreator();
IEmergingPatternSimplifier simplifier = new EmergingPatternSimplifier(new ItemComparer());
List <Feature> featuresToConsider = model.Features.Where(f => f != classFeature).ToList();
//int featureCount = (FeatureCount != -1) ? FeatureCount : Convert.ToInt32(Math.Max((int)Math.Log(featuresToConsider.Count, 2) + 1, 0.63* featuresToConsider.Count));
int featureCount = (FeatureCount != -1) ? FeatureCount : (int)Math.Log(featuresToConsider.Count, 2) + 1;
var resultPatterns = new List <IEmergingPattern>();
var instanceCount = instances.Count();
for (int i = 0; i < TreeCount; i++)
{
unsupervisedDecisionTreeBuilder.OnSelectingFeaturesToConsider =
(features, level) => _sampler.SampleWithoutRepetition(featuresToConsider, featureCount);
DecisionTree tree = unsupervisedDecisionTreeBuilder.Build(model, instances, classFeature);
DecisionTreeClassifier treeClassifier = new DecisionTreeClassifier(tree);
if (treeClassifier.DecisionTree.Leaves > 1)
{
EpCreator.ExtractPatterns(treeClassifier,
delegate(EmergingPattern p)
{
if (EPTester.Test(p.Counts, model, classFeature))
{
resultPatterns.Add(simplifier.Simplify(p));
}
},
classFeature);
}
resultPatterns.Add(null);
}
foreach (var ep in resultPatterns)
{
if (ep != null)
{
ep.Counts = new double[1];
foreach (var instance in instances)
{
if (ep.IsMatch(instance))
{
ep.Counts[0]++;
}
}
ep.Supports = new double[1];
ep.Supports[0] = ep.Counts[0] / instanceCount;
}
}
return(resultPatterns);
}
public IEnumerable <IEmergingPattern> Mine(InstanceModel model, IEnumerable <Instance> instances, Feature classFeature)
{
EmergingPatternCreator EpCreator = new EmergingPatternCreator();
IEmergingPatternSimplifier simplifier;
if (Multivariate)
{
simplifier = new EmergingPatternSimplifier(new MultivariateItemComparer());
}
else
{
simplifier = new EmergingPatternSimplifier(new ItemComparer());
}
List <IEmergingPattern> patternsList = new List <IEmergingPattern>();
if (MinePatternsWhileBuildingTree)
{
DecisionTreeBuilder.OnSplitEvaluation =
delegate(IDecisionTreeNode node, ISplitIterator iterator, List <SelectorContext> currentContext)
{
IChildSelector currentSelector = null;
for (int i = 0; i < iterator.CurrentDistribution.Length; i++)
{
double[] distribution = iterator.CurrentDistribution[i];
if (EPTester.Test(distribution, model, classFeature))
{
if (currentSelector == null)
{
currentSelector = iterator.CreateCurrentChildSelector();
}
EmergingPattern ep = EpCreator.ExtractPattern(currentContext, model, classFeature,
currentSelector, i);
ep.Counts = (double[])distribution.Clone();
patternsList.Add(simplifier.Simplify(ep));
}
}
};
DoMine(model, instances, classFeature, EpCreator, null);
}
else
{
DoMine(model, instances, classFeature, EpCreator, p =>
{
if (EPTester.Test(p.Counts, model, classFeature))
{
patternsList.Add(simplifier.Simplify(p));
}
}
);
}
return(patternsList);
}
public IEnumerable <IEmergingPattern> MineTest(InstanceModel model, IEnumerable <Instance> instances, Feature classFeature)
{
EmergingPatternCreator EpCreator = new EmergingPatternCreator();
IEmergingPatternSimplifier simplifier = new EmergingPatternSimplifier(new ItemComparer());
List <Feature> featuresToConsider = model.Features.Where(f => f != classFeature).ToList();
int featureCount = (FeatureCount != -1) ? FeatureCount : (int)Math.Max(Math.Log(featuresToConsider.Count, 2) + 1, 0.63 * featuresToConsider.Count);
var resultPatterns = new List <IEmergingPattern>();
featureUseCount = new Dictionary <Feature, int>();
foreach (var feature in featuresToConsider)
{
featureUseCount.Add(feature, 0);
}
allFeaturesUseCount = 0;
var instanceCount = instances.Count();
for (int i = 0; i < TreeCount; i++)
{
cumulativeProbabilities = new List <double>();
double max = 0;
for (int j = 0; j < featuresToConsider.Count; j++)
{
if (featureUseCount[featuresToConsider[j]] > max)
{
max = featureUseCount[featuresToConsider[j]];
}
}
double sum = 0;
for (int j = 0; j < featuresToConsider.Count; j++)
{
cumulativeProbabilities.Add(allFeaturesUseCount == 0
? 1.0 / featuresToConsider.Count
: 1.0 * (max - featureUseCount[featuresToConsider[j]]) / max);
//cumulativeProbabilities.Add(allFeaturesUseCount == 0
// ? 1.0 / featuresToConsider.Count
// : 1.0 * (featureUseCount[featuresToConsider[j]]) / allFeaturesUseCount);
sum += cumulativeProbabilities[j];
if (j > 0)
{
cumulativeProbabilities[j] += cumulativeProbabilities[j - 1];
}
if (sum != cumulativeProbabilities[j])
{
throw new Exception("Error computing cumalitive probabilities!");
}
}
for (int j = 0; j < featuresToConsider.Count; j++)
{
cumulativeProbabilities[j] /= sum;
}
unsupervisedDecisionTreeBuilder.OnSelectingFeaturesToConsider =
(features, level) => SampleWithDistribution(featuresToConsider, featureCount);
DecisionTree tree = unsupervisedDecisionTreeBuilder.Build(model, instances, classFeature);
DecisionTreeClassifier treeClassifier = new DecisionTreeClassifier(tree);
if (treeClassifier.DecisionTree.Leaves > 1)
{
EpCreator.ExtractPatterns(treeClassifier,
delegate(EmergingPattern p)
{
if (EPTester.Test(p.Counts, model, classFeature))
{
foreach (Item item in p.Items)
{
featureUseCount[item.Feature]++;
allFeaturesUseCount++;
}
resultPatterns.Add(simplifier.Simplify(p));
}
},
classFeature);
}
resultPatterns.Add(null);
}
foreach (var ep in resultPatterns)
{
if (ep != null)
{
ep.Counts = new double[1];
foreach (var instance in instances)
{
if (ep.IsMatch(instance))
{
ep.Counts[0]++;
}
}
ep.Supports = new double[1];
ep.Supports[0] = ep.Counts[0] / instanceCount;
}
}
return(resultPatterns);
}