public static Dictionary <int, GaussianClusterCollection> GetIndexClustersMap(List <double[]> continuousData, IDictionary <int, int> continuesIndexKMap)
{
Dictionary <int, GaussianClusterCollection> indexClusterMap = new Dictionary <int, GaussianClusterCollection>();
Accord.Math.Random.Generator.Seed = 0;
foreach (int i in continuesIndexKMap.Keys)
{
// Prepare data to be analyzed in the KMeans algorithm
double[][] data = new double[continuousData.Count][];
for (int j = 0; j < continuousData.Count; j++)
{
data[j] = new[] { continuousData[j][i] };
}
// Grab k from optimum map.
int k = continuesIndexKMap[i];
KMeans kMeans = new KMeans(k);
var cluster = kMeans.Learn(data);
// Learn gaussian mixture model based on k-model
GaussianMixtureModel gaussianMixtureModel = new GaussianMixtureModel(kMeans);
indexClusterMap[i] = gaussianMixtureModel.Learn(data);
}
return(indexClusterMap);
}
public void GaussianMixtureModelExample()
{
// Test Samples
double[][] samples =
{
new double[] { 0, 1 },
new double[] { 1, 2 },
new double[] { 1, 1 },
new double[] { 0, 7 },
new double[] { 1, 1 },
new double[] { 6, 2 },
new double[] { 6, 5 },
new double[] { 5, 1 },
new double[] { 7, 1 },
new double[] { 5, 1 }
};
double[] sample = samples[0];
// Create a new Gaussian mixture with 2 components
var gmm = new GaussianMixtureModel(components: 2);
// Compute the model (estimate)
var clusters = gmm.Learn(samples);
double error = gmm.LogLikelihood;
// Classify a single sample
int c0 = clusters.Decide(samples[0]);
int c1 = clusters.Decide(samples[1]);
int c7 = clusters.Decide(samples[7]);
int c8 = clusters.Decide(samples[8]);
Assert.AreEqual(c0, c1);
Assert.AreEqual(c7, c8);
Assert.AreNotEqual(c0, c8);
// Extract the multivariate Normal distribution from it
MultivariateMixture <MultivariateNormalDistribution> mixture =
gmm.ToMixtureDistribution();
Assert.AreEqual(2, mixture.Dimension);
Assert.AreEqual(2, mixture.Components.Length);
Assert.AreEqual(2, mixture.Coefficients.Length);
}
private static void gmm(double[][] inputs)
{
var gmm = new GaussianMixtureModel(components: 3)
{
Tolerance = 1e-6
};
var model = gmm.Learn(inputs);
int[] prediction = model.Decide(inputs);
double[][] prob = model.Probabilities(inputs);
// Compute the clustering distortion
double distortion = model.Distortion(inputs, prediction);
// Plot the results
ScatterplotBox.Show("Free GMM's answer", inputs, prediction)
.Hold();
}
/// <summary>
/// Estimates Gaussian distributions from the data.
/// </summary>
///
private void btnCompute_Click(object sender, EventArgs e)
{
// Create a new Gaussian Mixture Model
var gmm = new GaussianMixtureModel(k);
// If available, initialize with k-means
if (kmeans != null)
{
gmm.Initialize(kmeans);
}
// Compute the model
GaussianClusterCollection clustering = gmm.Learn(observations);
// Classify all instances in mixture data
int[] classifications = clustering.Decide(observations);
// Draw the classifications
updateGraph(classifications);
}
private static void gmmSharedCovariances(double[][] inputs)
{
var gmm = new GaussianMixtureModel(components: 3)
{
Tolerance = 1e-6,
Options = new NormalOptions()
{
Shared = true
}
};
var model = gmm.Learn(inputs);
int[] prediction = model.Decide(inputs);
// Compute the clustering distortion
double distortion = model.Distortion(inputs, prediction);
// Plot the results
ScatterplotBox.Show("Shared GMM's answer", inputs, prediction)
.Hold();
}
public void GaussianMixtureModelTest4()
{
// Suppose we have a weighted data set. Those are the input points:
double[][] points =
{
new double[] { 0 }, new double[] { 3 }, new double[] { 1 },
new double[] { 7 }, new double[] { 3 }, new double[] { 5 },
new double[] { 1 }, new double[] { 2 }, new double[] { -1 },
new double[] { 2 }, new double[] { 7 }, new double[] { 6 },
new double[] { 8 }, new double[] { 6 } // (14 points)
};
// And those are their respective unnormalized weights:
double[] weights = { 10, 1, 1, 2, 2, 1, 1, 1, 8, 1, 2, 5, 1, 1 }; // (14 weights)
// with weights
{
Accord.Math.Tools.SetupGenerator(0);
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
var options = new GaussianMixtureModelOptions();
options.Weights = weights;
options.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Compute(points, options);
int a = 1;
int b = 0;
double tol = 1e-6;
if (!gmm.Gaussians[a].Mean[0].IsRelativelyEqual(6.41922, 1e-4))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.4192285647145395, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(0.2888226129013588, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.32321638614859777, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.67678361385140218, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion);
}
// with weights, new style
{
Accord.Math.Tools.SetupGenerator(0);
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
gmm.UseLogarithm = false;
gmm.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Learn(points, weights);
int a = 1;
int b = 0;
double tol = 1e-6;
if (!gmm.Gaussians[a].Mean[0].IsRelativelyEqual(6.41922, 1e-4))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.4192285647145395, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(0.2888226129013588, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.32321638614859777, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.67678361385140218, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion);
}
// without weights
{
Accord.Math.Random.Generator.Seed = 0;
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
var options = new GaussianMixtureModelOptions();
options.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Compute(points, options);
int a = 1;
int b = 0;
double tol = 1e-2;
if (!6.5149525060859848.IsRelativelyEqual(gmm.Gaussians[a].Mean[0], tol))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.5149525060859848, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(1.4191977895308987, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.4195042394315267, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.58049576056847307, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion, 1e-8);
}
}
public void GaussianMixtureModelTest4()
{
// Suppose we have a weighted data set. Those are the input points:
double[][] points =
{
new double[] { 0 }, new double[] { 3 }, new double[] { 1 },
new double[] { 7 }, new double[] { 3 }, new double[] { 5 },
new double[] { 1 }, new double[] { 2 }, new double[] { -1 },
new double[] { 2 }, new double[] { 7 }, new double[] { 6 },
new double[] { 8 }, new double[] { 6 } // (14 points)
};
// And those are their respective unnormalized weights:
double[] weights = { 10, 1, 1, 2, 2, 1, 1, 1, 8, 1, 2, 5, 1, 1 }; // (14 weights)
// with weights
{
Accord.Math.Tools.SetupGenerator(0);
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
var options = new GaussianMixtureModelOptions();
options.Weights = weights;
options.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Compute(points, options);
int a = 1;
int b = 0;
double tol = 1e-6;
if (!gmm.Gaussians[a].Mean[0].IsRelativelyEqual(6.41922, 1e-4))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.4192285647145395, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(0.2888226129013588, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.32321638614859777, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.67678361385140218, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion);
}
// with weights, new style
{
Accord.Math.Tools.SetupGenerator(0);
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
gmm.UseLogarithm = false;
gmm.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Learn(points, weights);
int a = 1;
int b = 0;
double tol = 1e-6;
if (!gmm.Gaussians[a].Mean[0].IsRelativelyEqual(6.41922, 1e-4))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.4192285647145395, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(0.2888226129013588, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.32321638614859777, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.67678361385140218, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion);
}
// without weights
{
Accord.Math.Random.Generator.Seed = 0;
// If we need the GaussianMixtureModel functionality, we can
// use the estimated mixture to initialize a new model:
GaussianMixtureModel gmm = new GaussianMixtureModel(2);
gmm.Initializations = 1;
var options = new GaussianMixtureModelOptions();
options.ParallelOptions.MaxDegreeOfParallelism = 1;
gmm.Compute(points, options);
int a = 1;
int b = 0;
double tol = 1e-2;
if (!6.5149525060859848.IsRelativelyEqual(gmm.Gaussians[a].Mean[0], tol))
{
var t = a;
a = b;
b = t;
}
Assert.AreEqual(6.5149525060859848, gmm.Gaussians[a].Mean[0], tol);
Assert.AreEqual(1.4191977895308987, gmm.Gaussians[b].Mean[0], tol);
Assert.AreEqual(0.4195042394315267, gmm.Gaussians[a].Proportion, tol);
Assert.AreEqual(0.58049576056847307, gmm.Gaussians[b].Proportion, tol);
Assert.AreEqual(1, gmm.Gaussians[0].Proportion + gmm.Gaussians[1].Proportion, 1e-8);
}
}
/// <summary>
/// Generates both training set consisting of feasible and infeasible examples
/// </summary>
public void GenerateTrainingData()
{
Feasibles = GenerateFeasibleExamples(GlobalVariables.FeasibleExamplesCount);
// Fill training data
TrainingData = Feasibles.AddColumnWithValues(1.0);
// Fill output List
var tempArr = new double[TrainingData.Count];
Output.AddRange(tempArr.Select(x => 1));
_dal.TrainingFeasibleExamples = Feasibles.ToArray();
GaussianMixtureModel gmm = new GaussianMixtureModel(GlobalVariables.Components);
gmm = new GaussianMixtureModel(GlobalVariables.Components)
{
Initializations = 100,
MaxIterations = 10000,
ParallelOptions = new System.Threading.Tasks.ParallelOptions()
{
MaxDegreeOfParallelism = 1
},
Tolerance = 10E-11,
Options = new Accord.Statistics.Distributions.Fitting.NormalOptions()
{
Regularization = double.Epsilon
}
};
// Estimate the Gaussian Mixture
gmm.Learn(_dal.TrainingFeasibleExamples);
var iterations = gmm.Iterations;
var distribution = gmm.ToMixtureDistribution();
// Get minimal probability of probability density function from distribution (percentile 0)
var minimalProbability = _dal.TrainingFeasibleExamples
.Select(item => distribution.ProbabilityDensityFunction(item))
.Min();
// Rescale data range for infeasible example creation
NewBoundries = BoundryRescaler.Rescale(Feasibles);
// Generate infeasible examples
var infeasibles = new List <double[]>();
while (infeasibles.Count < GlobalVariables.InfeasibleExamplesCount)
{
// Generate points within new boundry
var x = GenerateLimitedInputs(GlobalVariables.Dimensions, NewBoundries);
// Calculate probability density function value for given input
var probability = distribution.ProbabilityDensityFunction(x);
// Check if the value is smaller than smallest probability of all feasible examples
if (probability > minimalProbability)
{
continue;
}
infeasibles.Add(x);
TrainingData.Add(x.ExtendArrayWithValue(0.0));
Output.Add(0);
}
_dal.TrainingInfeasibleExamples = infeasibles.ToArray();
_dal.TrainingData = TrainingData.ToArray();
}
/// <summary>
/// Estimates Gaussian distributions from the data.
/// </summary>
///
private void btnCompute_Click(object sender, EventArgs e)
{
// Create a new Gaussian Mixture Model
var gmm = new GaussianMixtureModel(k);
// If available, initialize with k-means
if (kmeans != null)
gmm.Initialize(kmeans);
// Compute the model
GaussianClusterCollection clustering = gmm.Learn(observations);
// Classify all instances in mixture data
int[] classifications = clustering.Decide(observations);
// Draw the classifications
updateGraph(classifications);
}