public void sparse_features_simple()
{
var instance1 = new[] { 20f, 30f, 0f, 0f };
var instance2 = new[] { 20f, 30f, 0f, 0f };
var instance3 = new[] { 0f, 0f, 10f, 20f };
var instance4 = new[] { 0f, 0f, 10f, 20f };
var inputFeaturesTypes = new InputFeatureTypes[] {
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal
};
var ordinalSparseSet = new InstanceRepresentation(
inputFeaturesTypes,
sparse: true);
ordinalSparseSet.AddInstance(instance1);
ordinalSparseSet.AddInstance(instance2);
ordinalSparseSet.AddInstance(instance3);
ordinalSparseSet.AddInstance(instance4);
var clustering = new MiniBatchClustering(2, 3, 10);
clustering.Train(ordinalSparseSet);
var categories = clustering.Cluster(ordinalSparseSet);
}
public void dense_features_simple()
{
var instance1 = new[] { 20f, 30f, 0f, 0f };
var instance2 = new[] { 20f, 30f, 0f, 0f };
var instance3 = new[] { 0f, 0f, 10f, 20f };
var instance4 = new[] { 0f, 0f, 10f, 20f };
var inputFeaturesTypes = new InputFeatureTypes[] {
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal
};
var ordinalDenseSet = new InstanceRepresentation(inputFeaturesTypes, sparse: false);
ordinalDenseSet.AddInstance(instance1);
ordinalDenseSet.AddInstance(instance2);
ordinalDenseSet.AddInstance(instance3);
ordinalDenseSet.AddInstance(instance4);
var som = new SelfOrganizingMap(10, new ushort[] { 20, 20 }, 100, 3);
var categories = som.Train(ordinalDenseSet);
}
public void sparse_features_medium()
{
// Cluster generator settings
var maxRadius = 100;
var minRadius = 10;
var clusterCount = 10;
// Data size
var featureDim = 100;
var obsCount = 2000;
var inputFeaturesTypes = new InputFeatureTypes[featureDim];
for (var i = 0; i < featureDim; i++)
{
inputFeaturesTypes[i] = InputFeatureTypes.Ordinal;
}
var sparseOrdinalSet = new InstanceRepresentation(
inputFeaturesTypes,
sparse: true);
var clusterGenerator = new SyntheticDataGenerator(maxRadius, minRadius, obsCount, clusterCount, featureDim);
var trueClusterLabels = new List <int>();
using (var obsGetter = clusterGenerator.GenerateClusterObservations().GetEnumerator())
{
var isNextObservation = obsGetter.MoveNext();
while (isNextObservation)
{
var cluster = obsGetter.Current.Item1;
var obs = obsGetter.Current.Item2;
trueClusterLabels.Add(cluster);
sparseOrdinalSet.AddInstance(obs);
isNextObservation = obsGetter.MoveNext();
}
}
var clustering = new MiniBatchClustering(clusterCount, 100, 2000);
clustering.Train(sparseOrdinalSet);
var categories = clustering.Cluster(sparseOrdinalSet);
var metricsGenerator = new MetricsGenerator();
metricsGenerator.Add(Metrics.Purity);
for (var i = 0; i < categories.Length; i++)
{
metricsGenerator.AddResult(categories[i], trueClusterLabels[i]);
metricsGenerator.UpdateMetrics();
}
var purity = metricsGenerator.GetMetric(Metrics.Purity);
Assert.True(purity + Epsilon - 0.6 > 0);
}
/// <summary>
/// Directly train and cluster the provided dataset.
/// </summary>
public int[] Cluster(InstanceRepresentation set)
{
if (!_isInitialized)
{
throw new InvalidOperationException("The centroids are not trained.");
}
var instancesClusters = new int[set.Instances.Count];
for (var i = 0; i < set.Instances.Count; i++)
{
instancesClusters[i] = Instances.MinEucDistanceIndex(set.Instances[i], _centroids);
}
return(instancesClusters);
}
public int[] Train(InstanceRepresentation set)
{
set.Standardize();
var instances = set.Instances.ToArray();
_weights = new float[_gridDimensions[0] * _gridDimensions[1], set.FeauturesCount];
FeaturesCount = set.FeauturesCount;
// Intialize weights with gaussians
var n = _gridDimensions[0] * _gridDimensions[1];
if (!_usePlusPlusInit || n >= instances.Length)
{
for (ushort i = 0; i < n; i++)
{
for (ushort j = 0; j < FeaturesCount; j++)
{
_weights[i, j] = (float)GaussHelper.InvPhi(_random.NextDouble());
}
}
}
else
{
_weights = PlusPlusInitializer.InitializeCentroids(n, instances, _random);
}
for (var i = 0; i < _iterationsCount; i++)
{
var instance = instances[_random.Next(instances.Length)];
// Best Matching Unit
var bmuIndex = Instances.MinEucDistanceIndex(instance, _weights);
BMUCoordinates = ToCoordinates(bmuIndex);
UpdateHexagonWeights((ushort)NeighbourhoodRadius(i), LearningRate(i), BMUCoordinates, instance.GetValues());
}
var instancesClusters = new int[instances.Length];
for (var i = 0; i < instances.Length; i++)
{
instancesClusters[i] = Instances.MinEucDistanceIndex(instances[i], _weights);
}
return(instancesClusters);
}
public void standardize_dense_instance()
{
var instance1 = new[] { -1f, 5f, 0f, 1f };
var instance2 = new[] { 0f, 6f, 1f, 1f };
var instance3 = new[] { 3f, 7f, 1f, 1f };
var instance4 = new[] { 4f, 0f, 1f, 0f };
var feature1Sigma = Math.Sqrt(new[] { -1f, 0f, 3f, 4f }.Variance());
var feature2Sigma = Math.Sqrt(new[] { 5f, 6f, 7f, 0f }.Variance());
var feature1Mean = new[] { -1f, 0f, 3f, 4f }.Mean();
var feature2Mean = new[] { 5f, 6f, 7f, 0f }.Mean();
var inputFeaturesTypes = new InputFeatureTypes[] {
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Flags,
InputFeatureTypes.Flags
};
var mixedDenseSet = new InstanceRepresentation(
inputFeaturesTypes,
sparse: false);
mixedDenseSet.AddInstance(instance1);
mixedDenseSet.AddInstance(instance2);
mixedDenseSet.AddInstance(instance3);
mixedDenseSet.AddInstance(instance4);
mixedDenseSet.Standardize();
var value1 = mixedDenseSet.Instances[0].GetValues();
Assert.True(Math.Abs((-1f - feature1Mean) / feature1Sigma - value1[0]) < Epsilon);
Assert.True((Math.Abs((5f - feature2Mean) / feature2Sigma - value1[1]) < Epsilon));
Assert.Equal(-1f, value1[2]);
Assert.Equal(1f, value1[3]);
}
public void rescale_sparse_instance()
{
var instance1 = new[] { 1f, 5f, 0f, 1f };
var instance2 = new[] { 0f, 6f, 1f, 1f };
var instance3 = new[] { 3f, 7f, 1f, 1f };
var instance4 = new[] { 4f, 0f, 1f, 0f };
var feature1Max = 4f;
var feature2Max = 7f;
var feature1Min = 0f;
var feature2Min = 0f;
var inputFeaturesTypes = new InputFeatureTypes[] {
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Flags,
InputFeatureTypes.Flags
};
var mixedSparseSet = new InstanceRepresentation(
inputFeaturesTypes,
sparse: true);
mixedSparseSet.AddInstance(instance1);
mixedSparseSet.AddInstance(instance2);
mixedSparseSet.AddInstance(instance3);
mixedSparseSet.AddInstance(instance4);
mixedSparseSet.Rescale();
var value1 = mixedSparseSet.Instances[0].GetValues();
Assert.True(Math.Abs((1f - feature1Min) / feature1Max - value1[0]) < Epsilon);
Assert.True((Math.Abs((5f - feature2Min) / feature2Max - value1[1]) < Epsilon));
Assert.Equal(0, value1[2]);
Assert.Equal(1f, value1[3]);
}
/// <summary>
/// Trains the centroids.
/// </summary>
public void Train(InstanceRepresentation set)
{
var instances = set.Instances.ToArray();
// Initializing the centroids:
if (_usePlusPlusInit)
{
_centroids = PlusPlusInitializer.InitializeCentroids(K, instances, _random);
}
else
{
_centroids = Instances.ConvertToArray(instances.SampleNoReplacement(K, _random));
}
_isInitialized = true;
FeaturesCount = set.FeauturesCount;
for (var i = 0; i < _iterationsCount; i++)
{
var miniBatch = instances.SampleReplacement(_minibatchSize, _random);
MiniBatchUpdate(miniBatch, set.IsSparseDataset);
}
}
public void dense_features_simple()
{
var instance1 = new[] { 20f, 30f, 0f, 0f };
var instance2 = new[] { 20f, 30f, 0f, 0f };
var instance3 = new[] { 0f, 0f, 10f, 20f };
var instance4 = new[] { 0f, 0f, 10f, 20f };
var inputFeaturesTypes = new InputFeatureTypes[] {
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal,
InputFeatureTypes.Ordinal
};
var ordinalDenseSet = new InstanceRepresentation(inputFeaturesTypes, sparse: false);
ordinalDenseSet.AddInstance(instance1);
ordinalDenseSet.AddInstance(instance2);
ordinalDenseSet.AddInstance(instance3);
ordinalDenseSet.AddInstance(instance4);
Assert.Equal(instance2, ordinalDenseSet.Instances[1].GetValues());
Assert.Equal(4, ordinalDenseSet.FeauturesCount);
}
