public static void Main() {
modshogun.init_shogun_with_defaults();
int gamma = 3;
double[,] traindata_real = Load.load_numbers("../data/fm_train_real.dat");
double[,] testdata_real = Load.load_numbers("../data/fm_test_real.dat");
double[] trainlab = Load.load_labels("../data/label_train_twoclass.dat");
RealFeatures feats_train = new RealFeatures();
feats_train.set_feature_matrix(traindata_real);
RealFeatures feats_test = new RealFeatures();
feats_test.set_feature_matrix(testdata_real);
BinaryLabels labels = new BinaryLabels(trainlab);
LDA lda = new LDA(gamma, feats_train, labels);
lda.train();
Console.WriteLine(lda.get_bias());
//Console.WriteLine(lda.get_w().toString());
foreach(double item in lda.get_w()) {
Console.Write(item);
}
lda.set_features(feats_test);
double[] out_labels = LabelsFactory.to_binary(lda.apply()).get_labels();
foreach(double item in out_labels) {
Console.Write(item);
}
}
public void TestLDA()
{
SignalBus signals = new SignalBus();
Operation op = new LDA(null, signals, null);
signals.Reset();
op.Step0();
Assert.IsTrue(signals.MI);
Assert.IsTrue(signals.CO);
signals.Reset();
op.Step1();
Assert.IsTrue(signals.RO);
Assert.IsTrue(signals.II);
Assert.IsTrue(signals.CE);
signals.Reset();
op.Step2();
Assert.IsTrue(signals.IO);
Assert.IsTrue(signals.MI);
signals.Reset();
op.Step3();
Assert.IsTrue(signals.RO);
Assert.IsTrue(signals.AI);
}
public void CreateAndDispose()
{
using (var lda = new LDA())
{
GC.KeepAlive(lda);
}
}
static void Main(string[] argv)
{
modshogun.init_shogun_with_defaults();
int gamma = 3;
DoubleMatrix traindata_real = Load.load_numbers("../data/fm_train_real.dat");
DoubleMatrix testdata_real = Load.load_numbers("../data/fm_test_real.dat");
DoubleMatrix trainlab = Load.load_labels("../data/label_train_twoclass.dat");
RealFeatures feats_train = new RealFeatures();
feats_train.set_feature_matrix(traindata_real);
RealFeatures feats_test = new RealFeatures();
feats_test.set_feature_matrix(testdata_real);
Labels labels = new Labels(trainlab);
LDA lda = new LDA(gamma, feats_train, labels);
lda.train();
Console.WriteLine(lda.get_bias());
Console.WriteLine(lda.get_w().ToString());
lda.set_features(feats_test);
DoubleMatrix out_labels = lda.apply().get_labels();
Console.WriteLine(out_labels.ToString());
modshogun.exit_shogun();
}
public void Test_LDA_Performance2000_var(int rep)
{
int errorCode = 0;
try {
LDA lda = new LDA();
ILArray <double> X = ILMath.horzcat(ILMath.randn(2, 2000) * 2.0, ILMath.randn(2, 2000) * -2.0);
ILLogicalArray labels = ILMath.tological(ILMath.horzcat(ILMath.ones(1, 2000), ILMath.zeros(1, 2000)));
labels = labels.Concat(ILMath.tological(ILMath.zeros(1, 2000).Concat(ILMath.ones(1, 2000), 1)), 0);
ILPerformer timer = new ILPerformer();
LDA.Hyperplane C;
int oldRefMin = ILNumerics.Settings.ILSettings.MinimumRefDimensions;
ILNumerics.Settings.ILSettings.MinimumRefDimensions = 2;
timer.Tic();
for (int i = 0; i < rep; i++)
{
C = lda.TrainLDA(X, labels, 0.4);
}
timer.Toc();
Info("Test_LDA_Performance: with reference - data: 2x2000 run " + rep.ToString() + " times in: " + timer.Duration + "ms");
ILNumerics.Settings.ILSettings.MinimumRefDimensions = 3;
timer.Tic();
for (int i = 0; i < rep; i++)
{
C = lda.TrainLDA(X, labels, 0.4);
}
timer.Toc();
ILNumerics.Settings.ILSettings.MinimumRefDimensions = oldRefMin;
Info("Test_LDA_Performance: without reference - data: 2x2000 run " + rep.ToString() + " times in: " + timer.Duration + "ms");
Success();
}catch (Exception e) {
Error(errorCode, e.Message);
}
}
static async Task Main(string[] args)
{
Console.OutputEncoding = Encoding.UTF8;
ApplicationLogging.SetLoggerFactory(LoggerFactory.Create(lb => lb.AddConsole()));
//Need to register the languages we want to use first
Catalyst.Models.English.Register();
//Configures the model storage to use the local folder ./catalyst-models/
Storage.Current = new DiskStorage("catalyst-models");
//Download the Reuters corpus if necessary
var(train, test) = await Corpus.Reuters.GetAsync();
//Parse the documents using the English pipeline, as the text data is untokenized so far
var nlp = Pipeline.For(Language.English);
var trainDocs = nlp.Process(train).ToArray();
var testDocs = nlp.Process(test).ToArray();
//Train an LDA topic model on the trainind dateset
using (var lda = new LDA(Language.English, 0, "reuters-lda"))
{
lda.Data.NumberOfTopics = 20; //Arbitrary number of topics
lda.Train(trainDocs, Environment.ProcessorCount);
await lda.StoreAsync();
}
using (var lda = await LDA.FromStoreAsync(Language.English, 0, "reuters-lda"))
{
foreach (var doc in testDocs)
{
if (lda.TryPredict(doc, out var topics))
{
var docTopics = string.Join("\n", topics.Select(t => lda.TryDescribeTopic(t.TopicID, out var td) ? $"[{t.Score:n3}] => {td.ToString()}" : ""));
Console.WriteLine("------------------------------------------");
Console.WriteLine(doc.Value);
Console.WriteLine("------------------------------------------");
Console.WriteLine(docTopics);
Console.WriteLine("------------------------------------------\n\n");
}
}
}
}
public static void execute(CPU cpu, MEM memory)
{
switch (memory.getMem(cpu.PC))
{
case 0xA9:
LDA.LDA_IM(cpu, memory);
return;
case 0xA5:
case 0xB5:
case 0xAD:
case 0xBD:
case 0xB9:
case 0xA1:
case 0xB1:
return;
}
}
public static void Main()
{
modshogun.init_shogun_with_defaults();
int gamma = 3;
double[,] traindata_real = Load.load_numbers("../data/fm_train_real.dat");
double[,] testdata_real = Load.load_numbers("../data/fm_test_real.dat");
double[] trainlab = Load.load_labels("../data/label_train_twoclass.dat");
RealFeatures feats_train = new RealFeatures();
feats_train.set_feature_matrix(traindata_real);
RealFeatures feats_test = new RealFeatures();
feats_test.set_feature_matrix(testdata_real);
BinaryLabels labels = new BinaryLabels(trainlab);
LDA lda = new LDA(gamma, feats_train, labels);
lda.train();
Console.WriteLine(lda.get_bias());
//Console.WriteLine(lda.get_w().toString());
foreach (double item in lda.get_w())
{
Console.Write(item);
}
lda.set_features(feats_test);
double[] out_labels = BinaryLabels.obtain_from_generic(lda.apply()).get_labels();
foreach (double item in out_labels)
{
Console.Write(item);
}
modshogun.exit_shogun();
}
//public void Test_SimpleAsyncAlgorithmSample() {
// int errorCode = 0;
// try {
// SimpleAsyncSample sampleAlg = new SimpleAsyncSample(null, 2.0,50,ILMath.randn(3,2));
// sampleAlg.RunAsync();
// while (sampleAlg.State == ILAlgorithmRunningState.Running) {
// Info(sampleAlg.Progress * 100 + "\r");
// System.Threading.Thread.Sleep(300);
// }
// Info(sampleAlg.Result.result.ToString());
// //sampleAlg.Result;
// Success();
// } catch(Exception e) {
// Error(errorCode,e.Message);
// }
//}
public void Test_LDA()
{
int errorCode = 0;
try {
LDA lda = new LDA();
lda.StateChanged += new ILAlgorithmStateChangedEventHandler(lda_StateChanged);
lda.ProgressChanged += new ILAlgorithmStateChangedEventHandler(lda_ProgressChanged);
ILArray <double> X = new ILArray <double>(new double[] { -2, -2, -3, -3, 2, 2, 3, 3 }, 2, 4);
ILLogicalArray labels = new ILLogicalArray(new byte[8] {
0, 1, 0, 1, 1, 0, 1, 0
}, 2, 4);
LDA.Hyperplane C = lda.TrainLDA(X, labels, 0.4);
if (Object.ReferenceEquals(C, null))
{
throw new Exception("LDA: result is null!");
}
if (!(C.w is ILArray <double>) || C.w.Dimensions[0] != 2)
{
throw new Exception("LDA: Results C[0] should be ILArray<double> 2x1");
}
if (!(C.b is ILArray <double>) || !C.b.IsScalar)
{
throw new Exception("LDA: Results C[1] should be ILArray<double> 2x1");
}
if (ILMath.abs(C.w[0] - -9.3750) > 1e-8)
{
throw new Exception("LDA: invalid result: C.w(1) should be : -9.3750");
}
if (ILMath.abs(C.w[1] - -9.3750) > 1e-8)
{
throw new Exception("LDA: invalid result: C.w(2) should be : -9.3750");
}
if (ILMath.abs(C.b.GetValue(0)) > 1e-8)
{
throw new Exception("LDA: invalid result: C.b should be : 0.0!");
}
Success();
} catch (Exception e) {
Error(errorCode, e.Message);
}
}
// ReSharper disable once InconsistentNaming
public void LDASample()
{
double[,] d = { { 2.95, 6.63 }, { 2.53, 7.79 }, { 3.57, 5.65 }, { 3.16, 5.47 }, { 2.58, 4.46 }, { 2.16, 6.22 }, { 3.27, 3.52 } };
int[] c = { 0, 0, 0, 0, 1, 1, 1 };
using (var data = new Mat(d.GetLength(0), d.GetLength(1), MatType.CV_64FC1, d))
using (var classes = new Mat(c.Length, 1, MatType.CV_32SC1, c))
using (var lda = new LDA(data, classes))
{
using (var eigenvectors = lda.Eigenvectors())
{
Assert.Equal(2, eigenvectors.Rows);
Assert.Equal(1, eigenvectors.Cols);
Assert.Equal(-1.5836, eigenvectors.Get <double>(0), 4);
Assert.Equal(-0.659729, eigenvectors.Get <double>(1), 4);
}
using (var eigenvalues = lda.Eigenvalues())
{
Assert.Equal(1, eigenvalues.Rows);
Assert.Equal(1, eigenvalues.Cols);
Assert.Equal(3.1447, eigenvalues.Get <double>(0), 4);
}
using (var project = lda.Project(data))
{
Assert.Equal(d.GetLength(0), project.Rows);
Assert.Equal(1, project.Cols);
Assert.Equal(-9.04562, project.Get <double>(0), 5);
Assert.Equal(-9.14579, project.Get <double>(1), 5);
Assert.Equal(-9.38091, project.Get <double>(2), 5);
Assert.Equal(-8.61289, project.Get <double>(3), 5);
Assert.Equal(-7.02807, project.Get <double>(4), 5);
Assert.Equal(-7.52409, project.Get <double>(5), 5);
Assert.Equal(-7.50061, project.Get <double>(6), 5);
}
}
}
public void Test_LDA_Performance200k_10()
{
int errorCode = 0;
try {
LDA lda = new LDA();
ILArray <double> X = ILMath.horzcat(ILMath.randn(2, 200000) * 2.0, ILMath.randn(2, 200000) * -2.0);
ILLogicalArray labels = ILMath.tological(ILMath.horzcat(ILMath.ones(1, 100000), ILMath.zeros(1, 100000)));
labels = labels.Concat(ILMath.tological(ILMath.zeros(1, 100000).Concat(ILMath.ones(1, 100000), 1)), 0);
ILPerformer timer = new ILPerformer();
timer.Tic(); LDA.Hyperplane C;
for (int i = 0; i < 10; i++)
{
C = lda.TrainLDA(X, labels, 0.4);
}
timer.Toc();
Info("Test_LDA_Performance2: data: 2x200000 run 10 times in: " + timer.Duration + "ms");
Success();
}catch (Exception e) {
Error(errorCode, e.Message);
}
}
private Guid trainLDA(ImageDatabase imdb, FrcContext db)
{
if (!imdb.isSameImageSize || !imdb.isSameTotalImageForUser)
{
throw new NotImplementedException();
}
// 1) Находим средний образ для всей базы исходных данных:
var averageMatrix = Matrix <double> .Build.Dense(imdb.ImageHeight, imdb.ImageWidth);
foreach (var userMatrixList in trainImageLists)
{
foreach (var userImage in userMatrixList)
{
averageMatrix = averageMatrix + userImage;
}
}
var Xaverage = averageMatrix / (totalUserForTrain * totalTrainImageForUser);
// 2) Находим средний образ для каждого класса:
List <Matrix <double> > XclassAverageList = new List <Matrix <double> >();
foreach (var userMatrixList in trainImageLists)
{
var averageClassMatrix = Matrix <double> .Build.Dense(imdb.ImageHeight, imdb.ImageWidth);
foreach (var userImage in userMatrixList)
{
averageClassMatrix = averageClassMatrix + userImage;
}
averageClassMatrix = averageClassMatrix / totalTrainImageForUser;
XclassAverageList.Add(averageClassMatrix);
}
// 3) Вычислим матрицу внутриклассовой(w) ковариации относительно строк(R):
var SwR = Matrix <double> .Build.Dense(imdb.ImageHeight, imdb.ImageHeight);
for (int i = 0; i < trainImageLists.Count; i++)
{
foreach (var userImage in trainImageLists[i])
{
SwR = SwR + (userImage - XclassAverageList[i]) * (userImage - XclassAverageList[i]).Transpose();
}
}
// 4) Вычислим матрицу межклассовой(b) ковариации относительно строк(R):
var SbR = Matrix <double> .Build.Dense(imdb.ImageHeight, imdb.ImageHeight);
foreach (var XclassAverage in XclassAverageList)
{
SbR = SbR + (XclassAverage - Xaverage) * (XclassAverage - Xaverage).Transpose();
}
// 5) Вычислим матрицу внутриклассовой (w) ковариации относительно столбцов (C):
var SwC = Matrix <double> .Build.Dense(imdb.ImageWidth, imdb.ImageWidth);
for (int i = 0; i < trainImageLists.Count; i++)
{
foreach (var userImage in trainImageLists[i])
{
SwC = SwC + (userImage - XclassAverageList[i]).Transpose() * (userImage - XclassAverageList[i]);
}
}
// 6) Вычислим матрицу межклассовой (b) ковариации относительно столбцов (C):
var SbC = Matrix <double> .Build.Dense(imdb.ImageWidth, imdb.ImageWidth);
foreach (var XclassAverage in XclassAverageList)
{
SbC = SbC + (XclassAverage - Xaverage).Transpose() * (XclassAverage - Xaverage);
}
// 7) Сформируем общую матрицу рассеяния относительно строк:
var StotalR = SwR.Inverse() * SbR;
// Регуляризация (стр. 349):
StotalR = StotalR + Constants.SMALL_VALUE * Matrix <double> .Build.Dense(imdb.ImageHeight, imdb.ImageHeight);
// 8) Сформируем общую матрицу рассеяния относительно столбцов:
var StotalC = SwC.Inverse() * SbC;
// Регуляризация (стр. 349):
StotalC = StotalC + Constants.SMALL_VALUE * Matrix <double> .Build.Dense(imdb.ImageWidth, imdb.ImageWidth);
// 9) Решим задачу на собственные значения:
var ReigResult = StotalR.Evd();
var CeigResult = StotalC.Evd();
// 10) Пропускаем сортировку по eigenvalues и ограничение по 95%. Берём только то, что написано в мнемоническом описании:
// 11) Подготовка матриц для преобразования Карунена-Лоева:
List <Vector <double> > rVectorList = new List <Vector <double> >();
for (int i = 0; i < md.trainMartixRightDimension; i++)
{
rVectorList.Add(ReigResult.EigenVectors.Column(i));
}
// to do: значения матриц R как-то оказываются слева. Нужно переименовывать везде, а лучше ещё раз уточнить формулы:
var eigMatrixLeft = Matrix <double> .Build.DenseOfColumnVectors(rVectorList.ToArray()).Transpose();
List <Vector <double> > cVectorList = new List <Vector <double> >();
for (int i = 0; i < md.trainMartixLeftDimension; i++)
{
cVectorList.Add(CeigResult.EigenVectors.Column(i));
}
var eigMatrixRight = Matrix <double> .Build.DenseOfColumnVectors(cVectorList.ToArray());
var averageImageMatrixString = MatrixHelper.convertToMatrixString(Xaverage);
var leftMatrixString = MatrixHelper.convertToMatrixString(eigMatrixLeft);
var rightMatrixString = MatrixHelper.convertToMatrixString(eigMatrixRight);
db.MatrixStrings.Add(averageImageMatrixString);
db.MatrixStrings.Add(leftMatrixString);
db.MatrixStrings.Add(rightMatrixString);
db.SaveChanges();
var ldaEntity = new LDA
{
AverageImageMatrixId = averageImageMatrixString.MatrixStringId,
LDAId = Guid.NewGuid(),
LeftMatrixId = leftMatrixString.MatrixStringId,
RightMatrixId = rightMatrixString.MatrixStringId,
};
db.LDAs.Add(ldaEntity);
db.SaveChanges();
var frs = new Entities.FaceRecognitionSystem
{
FaceRecognitionSystemId = Guid.NewGuid(),
MnemonicDescription = md.originalDescription,
Type = "LDA",
TypeSystemId = ldaEntity.LDAId,
InputImageHeight = imdb.ImageHeight,
InputImageWidth = imdb.ImageWidth,
CreatedDT = DateTime.UtcNow,
};
db.FaceRecognitionSystems.Add(frs);
db.SaveChanges();
return(frs.FaceRecognitionSystemId);
}
public static async Task LDAllocation()
{
Storage.Current = new OnlineRepositoryStorage(new DiskStorage("catalyst-models"));
//var train = await Corpus.Reuters.GetAsync();
string connectionStr = "Data Source = (localdb)\\MSSQLLocalDB; Initial Catalog = master; Integrated Security = True; Connect Timeout = 30; Encrypt = False; TrustServerCertificate = False; ApplicationIntent = ReadWrite; MultiSubnetFailover = False";
SqlConnection connection = null;
SqlCommand command = null;
SqlCommand command1 = null;
string sqlQuery = "SELECT * FROM TestApplication.dbo.Files where test = 0";
string sqlQuery1 = "SELECT * FROM TestApplication.dbo.Files where test = 1";
var nlp = Pipeline.For(Language.English);
try
{
connection = new SqlConnection(connectionStr);
command = new SqlCommand(sqlQuery, connection);
command1 = new SqlCommand(sqlQuery1, connection);
connection.Open();
SqlDataReader reader = command.ExecuteReader();
System.Collections.Generic.List <Document> train = new System.Collections.Generic.List <Document>();
System.Collections.Generic.List <Document> test = new System.Collections.Generic.List <Document>();
System.Text.RegularExpressions.Regex reg = new System.Text.RegularExpressions.Regex("/[^/]*/");
while (reader.Read())
{
NLPFile flTrain = new NLPFile((IDataRecord)reader);
//Console.WriteLine(flTrain.FetchData());
var doc = new Document(flTrain.FetchData(), Language.English);
string label = reg.Matches(((string)reader[1]))[2].Value;
label = label.Trim('/');
doc.Labels.Add(label);
train.Add(doc);
var trainDocs = nlp.Process(train).ToArray();
using (var lda = new LDA(Language.English, 0, "reuters-lda"))
{
lda.Data.NumberOfTopics = 20; //Arbitrary number of topics
lda.Train(trainDocs, Environment.ProcessorCount);
await lda.StoreAsync();
}
}
reader.Close();
SqlDataReader reader1 = command1.ExecuteReader();
while (reader1.Read())
{
NLPFile flTest = new NLPFile((IDataRecord)reader1);
//Console.WriteLine(flTest.FetchData());
var doc1 = new Document(flTest.FetchData(), Language.English);
string label = reg.Matches(((string)reader[1]))[2].Value;
label = label.Trim('/');
doc1.Labels.Add(label);
test.Add(doc1);
var testDocs = nlp.Process(test).ToArray();
using (var lda = await LDA.FromStoreAsync(Language.English, 0, "reuters-lda"))
{
foreach (var doc in testDocs)
{
if (lda.TryPredict(doc, out var topics))
{
var docTopics = string.Join("\n", topics.Select(t => lda.TryDescribeTopic(t.TopicID, out var td) ? $"[{t.Score:n3}] => {td.ToString()}" : ""));
Console.WriteLine("------------------------------------------");
Console.WriteLine(doc.Value);
Console.WriteLine("------------------------------------------");
Console.WriteLine(docTopics);
Console.WriteLine("------------------------------------------\n\n");
}
}
}
}
reader1.Close();
}
catch (Exception e)
{
Console.WriteLine("Error: " + e.Message);
}
finally
{
if (command != null || command1 != null)
{
command.Dispose();
command1.Dispose();
}
if (connection != null)
{
connection.Close();
connection.Dispose();
}
}
}
public static double[] CalculateBiases <L>(this Sphere <L> sphere, Vector vector, RHCLib.DistanceDelegate measure, IList <L> labels)
{
SortedDictionary <L, double?> dictClasses = new SortedDictionary <L, double?>();
foreach (L label in labels)
{
dictClasses.Add(label, null);
}
double fProportion = 1.0;
int nCount = dictClasses.Count;
bool bFirstSphere = true;
SphereEx <L> sphLDA;
RHCLib.Sphere <L> sphereIteration = sphere.Recognize(vector, measure, ParallelStrategy.SingleThreaded);
while (sphereIteration != null && fProportion > 0.0 && nCount > 0)
{
if (bFirstSphere)
{
if (dictClasses.ContainsKey(sphereIteration.Label))
{
if ((sphLDA = sphereIteration as SphereEx <L>) != null && sphLDA.DiscriminantEx != null)
{
#region LDA (Uses LDAEx)
// You have four cases you need to watch for...
// /|\ /|\ 1.0
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \/ | \
// |---------M---------D--------M---------| 0.5
// Equation: 1 - (h)(x_i) <-- 1 = fProportion because haven't gotten out yet
Func <double, double, double> slope = (x1, x2) =>
{
return(0.5 / Math.Abs(x2 - x1));
};
//double[][] data = LDA.MatrixFromVector(vector.Features);
//double[][] wTx = LDA.MatrixProduct(sphLDA.Discriminant.Transposed, data); // Project the data
double proj = Accord.Math.Matrix.Dot(sphLDA.DiscriminantEx.ProjectionVector, vector.Features);
if (proj <= sphLDA.DiscriminantEx.ProjectedLeftMean)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.DiscriminantEx.ProjectedLeftMean, sphLDA.DiscriminantEx.ProjectedSetMean - sphLDA.Radius) * (sphLDA.DiscriminantEx.ProjectedLeftMean - proj));
}
else if (proj > sphLDA.DiscriminantEx.ProjectedLeftMean && proj <= sphLDA.DiscriminantEx.ProjectedSetMean)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.DiscriminantEx.ProjectedSetMean, sphLDA.DiscriminantEx.ProjectedLeftMean) * (proj - sphLDA.DiscriminantEx.ProjectedLeftMean));
}
else if (proj > sphLDA.DiscriminantEx.ProjectedSetMean && proj <= sphLDA.DiscriminantEx.ProjectedRightMean)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.DiscriminantEx.ProjectedRightMean, sphLDA.DiscriminantEx.ProjectedSetMean) * (sphLDA.DiscriminantEx.ProjectedRightMean - proj));
}
else
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.DiscriminantEx.ProjectedSetMean + sphLDA.Radius, sphLDA.DiscriminantEx.ProjectedRightMean) * (proj - sphLDA.DiscriminantEx.ProjectedRightMean));
}
#endregion
}
else if (sphLDA != null && sphLDA.Discriminant != null)
{
#region Old LDA
// You have four cases you need to watch for...
// /|\ /|\ 1.0
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \ / | \
// / | \/ | \
// |---------M---------D--------M---------| 0.5
// Equation: 1 - h(x_i)
Func <double, double, double> slope = (x1, x2) =>
{
return(0.5 / Math.Abs(x2 - x1));
};
double[][] data = LDA.MatrixFromVector(vector.Features);
double[][] wTx = LDA.MatrixProduct(sphLDA.Discriminant.Transposed, data); // Project the data
if (wTx[0][0] <= sphLDA.Discriminant.ProjectedMeanLeft)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.Discriminant.ProjectedMeanLeft, 0.0) * (sphLDA.Discriminant.ProjectedMeanLeft - wTx[0][0]));
}
else if (wTx[0][0] > sphLDA.Discriminant.ProjectedMeanLeft && wTx[0][0] <= sphLDA.Discriminant.DecisionPoint)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.Discriminant.DecisionPoint, sphLDA.Discriminant.ProjectedMeanLeft) * (wTx[0][0] - sphLDA.Discriminant.ProjectedMeanLeft));
}
else if (wTx[0][0] > sphLDA.Discriminant.DecisionPoint && wTx[0][0] <= sphLDA.Discriminant.ProjectMeanRight)
{
dictClasses[sphLDA.Label] = fProportion - (slope(sphLDA.Discriminant.ProjectMeanRight, sphLDA.Discriminant.DecisionPoint) * (sphLDA.Discriminant.ProjectMeanRight - wTx[0][0]));
}
else
{
dictClasses[sphLDA.Label] = fProportion - (slope(2 * sphLDA.Radius, sphLDA.Discriminant.ProjectMeanRight) * (wTx[0][0] - sphLDA.Discriminant.ProjectMeanRight));
}
#endregion
}
else
{
#region Linear
dictClasses[sphereIteration.Label] = fProportion - (measure(sphereIteration, vector) * (0.5 / sphereIteration.Radius));
#endregion
}
fProportion -= dictClasses[sphereIteration.Label].Value;
bFirstSphere = false;
nCount--;
}
}
else
{
if (dictClasses.ContainsKey(sphereIteration.Label) && !dictClasses[sphereIteration.Label].HasValue)
{
dictClasses[sphereIteration.Label] = fProportion;
nCount--;
}
#region Linear
// Impossible to have LDA in node that's not a leaf.
fProportion -= fProportion * ((sphereIteration.Radius - measure(sphereIteration, vector)) / sphereIteration.Radius);
#endregion
}
sphereIteration = sphereIteration.Parent;
}
double fSum = dictClasses.Values.Sum(v => v.HasValue ? v.Value : 0.0);
return(dictClasses.Values.Select(v => v.HasValue ? v.Value / fSum : 0.0).ToArray());
}