public ReportDiscrete(IModelDiscrete <DomainType, LabelType> model, int[,] confusion_matrix_instances)
{
this.Model = model;
this.confusion_matrix_instances = ToolsCollection.Copy(confusion_matrix_instances);
this.confusion_matrix_rates = new double[confusion_matrix_instances.GetLength(0), confusion_matrix_instances.GetLength(1)];
int instance_count = 0;
this.CorrectLabelCount = 0;
for (int index_0 = 0; index_0 < confusion_matrix_instances.GetLength(0); index_0++)
{
for (int index_1 = 0; index_1 < confusion_matrix_instances.GetLength(1); index_1++)
{
instance_count += confusion_matrix_instances[index_0, index_1];
if (index_0 == index_1)
{
CorrectLabelCount += confusion_matrix_instances[index_0, index_1];
}
}
}
CorrectLabelRate = CorrectLabelCount / (double)instance_count;
for (int index_0 = 0; index_0 < confusion_matrix_instances.GetLength(0); index_0++)
{
for (int index_1 = 0; index_1 < confusion_matrix_instances.GetLength(1); index_1++)
{
confusion_matrix_rates[index_0, index_1] = (double)confusion_matrix_instances[index_0, index_1] / (double)instance_count;
}
}
}
public ReportDiscrete <DomainType, LabelType> GenerateAndTestDiscrete(IDataSet <DomainType, LabelType> training_set, IDataSet <DomainType, LabelType> test_set)
{
//TODO check data contexts (they should match)
IModelDiscrete <DomainType, LabelType> model = GenerateModelDiscrete(training_set);
int label_value_count = test_set.DataContext.GetLabelDescriptor(0).ValueCount;
int[,] confusion_matrix_instances = new int[label_value_count, label_value_count];
for (int instance_index = 0; instance_index < test_set.InstanceCount; instance_index++)
{
//int true_label_index = test_set.DataContext.GetLabelValueIndex(test_set.GetInstanceLabelData(instance_index));
//int model_label_index = model.GetLabelIndex(test_set.GetInstanceFeatureData(instance_index));
//confusion_matrix_instances[true_label_index, model_label_index]++;
////
throw new NotImplementedException();
}
return(new ReportDiscrete <DomainType, LabelType>(model, confusion_matrix_instances));
}
public Tuple <IModelDiscrete <DomainType, LabelType>, IList <int> > SelectFeatureSet(ITemplateModelDiscrete <DomainType, LabelType> template, IDataSet <DomainType, LabelType> data_set)
{
ISet <ISet <int> > black_list = new HashSet <ISet <int> >();
ISet <int> full_set = new HashSet <int>();
for (int feature_index = 0; feature_index < data_set.FeatureCount; feature_index++)
{
full_set.Add(feature_index);
}
ISet <int> current_set = new HashSet <int>();
double best_score = feature_set_evaluator.Evaluate(template, data_set, current_set);
bool improvement = true;
while (improvement)
{
List <ISet <int> > options = new List <ISet <int> >();
options.AddRange(RemoveFeature(current_set, black_list));
options.AddRange(AddFeature(current_set, black_list, full_set));
improvement = false;
foreach (ISet <int> option in options)
{
double score = feature_set_evaluator.Evaluate(template, data_set, option);
//System.out.print("scorein " + score);
//CollectionTools.print(option);
if (best_score.CompareTo(score) == -1)
{
//System.out.print("New best scorein " + score);
//CollectionTools.print(option);
best_score = score;
current_set = option;
improvement = true;
}
}
}
List <int> current_list = new List <int>(current_set);
current_list.Sort();
IModelDiscrete <DomainType, LabelType> model = template.GenerateModelDiscrete(data_set.SelectFeatures(current_list));
return(new Tuple <IModelDiscrete <DomainType, LabelType>, IList <int> >(model, current_list));
}
public static void WriteModelDiscrete <DomainType, LabelType>(BinaryWriter writer, IModelDiscrete <DomainType, LabelType> model)
{
string model_type = model.ModelType;
writer.Write(model_type);
switch (model_type)
{
case "ModelDiscreteC45Default":
ModelDiscreteC45Default.Write(writer, (ModelDiscreteC45Default)model);
break;
case "ModelNearestNeighborListDefault":
ModelNearestNeighborListDefault.Write(writer, (ModelNearestNeighborListDefault)model);
break;
case "ModelNearestNeighborKDTreeDefault":
ModelNearestNeighborKDTreeDefault.Write(writer, (ModelNearestNeighborKDTreeDefault)model);
break;
case "ModelLibSVMCSVC":
ModelLibSVMCSVC.Write(writer, (ModelLibSVMCSVC)model);
break;
case "ModelNaiveBayesIntervalDefault":
ModelNaiveBayesInterval.Write(writer, (ModelNaiveBayesInterval)model);
break;
default:
throw new Exception("Unknown discrete model type: " + model_type);
}
}
public ModelDiscreteSelecting(IDataContext data_context, IModelDiscrete <DomainType, LabelType> model, IList <int> selected_feature_indexes)
: base(data_context, "ModelDiscreteSelecting")
{
this.model = model;
this.selected_feature_indexes = selected_feature_indexes;
}
public static ReportDiscrete <DomainType, LabelType> Read(IModelDiscrete <DomainType, LabelType> model, BinaryReader reader)
{
return(new ReportDiscrete <DomainType, LabelType>(model, reader.ReadInt32Array2D()));
}
public ReportDiscrete(IModelDiscrete <DomainType, LabelType> model, BinaryReader reader)
: this(model, reader.ReadInt32Array2D())
{
}