public override IModelLikelihood <int, int> GenerateModelLikelihood(IDataSet <int, int> training_set)
{
//TODO this asumes our label is finite and what not, maybe a bit silly
double[] priors = new double[training_set.DataContext.GetLabelDescriptor(0).ValueCount];
int[] instance_labels = training_set.GetLabelDataColumn(0);
DictionaryCount <int> count_map = new DictionaryCount <int>(instance_labels);
foreach (int key in count_map.Keys)
{
priors[key] = ((double)count_map.Get(key) / ((double)count_map.TotalCount));
}
IDictionary <int[], int[]> occurences = new DictionaryArrayKey <int, int[]>();
for (int instance_index = 0; instance_index < training_set.InstanceCount; instance_index++)
{
int [] instance_features = training_set.GetInstanceFeatureData(instance_index);
if (!occurences.ContainsKey(instance_features))
{
occurences[instance_features] = new int[priors.Length + 1]; //index 0 is for total
}
occurences[instance_features][0]++;
occurences[instance_features][instance_labels[instance_index] + 1]++;
}
return(new ModelLikelihoodNominalJointTable <int>(training_set.DataContext, priors, occurences));
}
public IModelDiscreteIterative <double, int> GenerateModelDiscreteIncremental(IDataSet <double, int> training_set)
{
ModelNearestNeighborListDefault model = new ModelNearestNeighborListDefault(training_set.DataContext);
IList <Tuple <double[], int> > training_instances = new List <Tuple <double[], int> >();
for (int instance_index = 0; instance_index < training_set.InstanceCount; instance_index++)
{
training_instances.Add(new Tuple <double[], int>(training_set.GetInstanceFeatureData(instance_index), training_set.GetInstanceLabelData(instance_index)[0]));
}
model.Add(training_instances);
return(model);
}
public IModelDiscreteIterative <int, int> GenerateModelDiscreteIncremental(IDataSet <int, int> training_set)
{
//TODO check data context
ModelNearestNeighborList <int, int, int> model = new ModelNearestNeighborList <int, int, int>(training_set.DataContext, new FunctionDistanceHamming());
IList <Tuple <int[], int> > training_instances = new List <Tuple <int[], int> >();
for (int instance_index = 0; instance_index < training_set.InstanceCount; instance_index++)
{
training_instances.Add(new Tuple <int[], int>(training_set.GetInstanceFeatureData(instance_index), training_set.GetInstanceLabelData(instance_index)[0]));
}
model.Add(training_instances);
return(model);
}
public ReportLikelihood <DomainType, LabelType> GenerateAndTestLikelihood(IDataSet <DomainType, LabelType> training_set, IDataSet <DomainType, LabelType> test_set)
{
IModelLikelihood <DomainType, LabelType> model = GenerateModelLikelihood(training_set);
double[][] likelihoods = new double[test_set.InstanceCount][];
LabelType[] labels = new LabelType[test_set.InstanceCount];
for (int instance_index = 0; instance_index < test_set.InstanceCount; instance_index++)
{
likelihoods[instance_index] = model.GetLikelihoods(test_set.GetInstanceFeatureData(instance_index));
}
return(new ReportLikelihood <DomainType, LabelType>(model, likelihoods, labels));
}
public IModelLabelIterative <double, double> GenerateModelLabelIterative(IDataSet <double, double> training_set)
{
//ModelNearestNeighborKDTreeDefault model = new ModelNearestNeighborKDTreeDefault(training_set.DataContext);
ModelNearestNeighborKDTree <double, double, double> model = new ModelNearestNeighborKDTree <double, double, double>("ModelKNNIneterval", training_set.DataContext, new AlgebraRealFloat64(), new FunctionDistanceEuclidean());
IList <Tuple <double[], double> > training_instances = new List <Tuple <double[], double> >();
for (int instance_index = 0; instance_index < training_set.InstanceCount; instance_index++)
{
training_instances.Add(new Tuple <double[], double>(training_set.GetInstanceFeatureData(instance_index), training_set.GetInstanceLabelData(instance_index)[0]));
}
model.Add(training_instances);
return(model);
}
public IModelDiscreteIterative <DomainType, LabelType> GenerateModelDiscrete(IDataSet <DomainType, LabelType> training_set)
{
ModelNearestNeighborKDTree <DomainType, DistanceType, LabelType> model = new ModelNearestNeighborKDTree <DomainType, DistanceType, LabelType>(
this.DataContext,
this.kdtree.Copy(),
this.DistanceFunction,
this.voting_system,
this.neighbor_count);
IList <Tuple <DomainType[], LabelType> > training_instances = new List <Tuple <DomainType[], LabelType> >();
for (int instance_index = 0; instance_index < training_set.InstanceCount; instance_index++)
{
training_instances.Add(new Tuple <DomainType[], LabelType>(training_set.GetInstanceFeatureData(instance_index), training_set.GetInstanceLabelData(instance_index)[0]));
}
model.Add(training_instances);
return(model);
}
private static void ProccessDataSetNominalPCA(IDataSet <int, int> training_set, IDataSet <int, int> test_set)
{
int cluster_count = 10;
int instance_count = training_set.InstanceCount;
ITemplateClusteringCentroid <int, IDataSet <int> > template_cluster = new TemplateClusteringKMeansNominal(cluster_count);
TemplateDimensionReductionPCADefault template_pca = new TemplateDimensionReductionPCADefault(3);
IDataSet <int> training_set_unlabeled = training_set;
IClusteringCentroid <int> clustering = template_cluster.Cluster(training_set_unlabeled);
double[][] transformed_data1 = new double[instance_count][];
for (int instance_index = 0; instance_index < instance_count; instance_index++)
{
transformed_data1[instance_index] = clustering.Transform(training_set_unlabeled.GetInstanceFeatureData(instance_index));
}
IDataSet <double> data_set_transformed1 = new DataSet <double, int>(transformed_data1);
ITransform transform = template_pca.GenerateTransform(data_set_transformed1);
double[][] transformed_data2 = new double[instance_count][];
int[][] label_data = training_set.LabelData;
Color[] instance_colors = new Color[instance_count];
Color[] label_colors = ToolsColor.GetPallete(training_set.DataContext.GetLabelDescriptor(0).ValueCount);
for (int instance_index = 0; instance_index < instance_count; instance_index++)
{
transformed_data2[instance_index] = transform.TransformForward(data_set_transformed1.GetInstanceFeatureData(instance_index));
instance_colors[instance_index] = label_colors[label_data[instance_index][0]];
}
RendererPoints <Matrix <double> > renderer = new RendererPoints <Matrix <double> >(new AlgebraLinearReal64MathNet(), 1000, 1000, (AngleRadian)0, (AngleRadian)0, 10);
BitmapFast bitmap = renderer.Render(new Tuple <double[][], Color[]>(transformed_data2, instance_colors));
bitmap.Bitmap.Save("test.png");
}
