public void ConvertFromHyperstreamlineToHalfEdgeMesh()
{
var r = new Random();
Func <double> random = r.NextDouble;
var m = new NamedBoxCollection();
//Deserialize config
var config = NetworkDescriptor.Deserialize(new StringReader(SCRIPT));
//Build main roads
var builder = new NetworkBuilder();
builder.Build(config.Major(random, m), random, m, new Vector2(0, 0), new Vector2(100, 100));
builder.Reduce();
//extract regions
var regions = builder.Regions();
//Build minor roads
foreach (var region in regions)
{
builder.Build(config.Minor(random, m), random, m, region);
}
builder.Reduce();
//Build graph
Mesh <IVertexBuilder, IHalfEdgeBuilder, IFaceBuilder> mesh = new Mesh <IVertexBuilder, IHalfEdgeBuilder, IFaceBuilder>();
//Extract vertices
var result = builder.Result;
Dictionary <Vector2, Vertex> vertices = result.Vertices.ToDictionary(v => v.Position, v => v);
//Useful data
var road = 3; //HierarchicalParameters.RoadLaneWidth(Random);
var path = 1; //HierarchicalParameters.RoadSidewalkWidth(Random);
//Create blocks
var blocks = builder.Regions();
foreach (var block in blocks)
{
var face = mesh.GetOrConstructFace(block.Vertices.Select(mesh.GetOrConstructVertex).ToArray());
//Tags edges of face with road width
foreach (var primaryEdge in face.Edges.Select(e => e.IsPrimaryEdge ? e : e.Pair))
{
if (primaryEdge.Tag != null)
{
continue;
}
var start = vertices[primaryEdge.Pair.EndVertex.Position];
var end = vertices[primaryEdge.EndVertex.Position];
var edge = start.Edges.FirstOrDefault(a => Equals(a.B, end));
primaryEdge.Tag = new HalfEdgeRoadBuilder(primaryEdge, road, path, edge == null ? 1u : edge.Streamline.Width);
}
}
}
public void NetworkBuilder_GeneratesMinorRoads()
{
var c = NetworkDescriptor.Deserialize(new StringReader(@"
!Network
Aliases:
- &base-field !AddTensors
Tensors:
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 0 }, Decay: 2.5, Center: { X: 0, Y: 0 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 45 }, Decay: 2.5, Center: { X: 1, Y: 0 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 45 }, Decay: 2.5, Center: { X: 0, Y: 1 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 0 }, Decay: 2.5, Center: { X: 1, Y: 1 } }
Major:
MergeSearchAngle: 22.5
MergeDistance: 25
SegmentLength: 10
RoadWidth: !NormalValue { Min: 2, Max: 4, Vary: true }
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 50 }
TensorField: *base-field
Minor:
MergeSearchAngle: 12.5
MergeDistance: 2.5
SegmentLength: 2
RoadWidth: !NormalValue { Min: 1, Max: 2, Vary: true }
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 25 }
TensorField:
!WeightedAverage
Tensors:
0.1: *base-field
0.9: !Grid { Angle: !UniformValue { Min: 1, Max: 360, Vary: true } }
"));
Random r = new Random(12);
NetworkBuilder b = new NetworkBuilder();
var m = new NamedBoxCollection();
b.Build(c.Major(r.NextDouble, m), r.NextDouble, m, new Vector2(0, 0), new Vector2(100, 100));
b.Reduce();
var regions = b.Regions();
foreach (var region in regions)
{
b.Build(c.Minor(r.NextDouble, m), r.NextDouble, m, region);
}
Console.WriteLine(b.Result.ToSvg());
Assert.IsFalse(b.Result.Vertices.GroupBy(a => a.Position).Any(a => a.Count() > 1));
}
public void Demo()
{
var c = NetworkDescriptor.Deserialize(new StringReader(@"
!Network
Aliases:
- &base-field !AddTensors
Tensors:
- !PointDistanceDecayTensors { Tensors: !Radial { Center: { X: 1, Y: 1 } }, Decay: 1.6, Center: { X: 0.45, Y: 0.45 } }
- !Polyline
Decay: 5
Points: [ { X: 0.3, Y: 0 }, { X: 0.1, Y: 0.33 }, { X: 0.5, Y: 0.66 }, { X: 0.2, Y: 1 } ]
Major:
MergeSearchAngle: 22.5
MergeDistance: 25
SegmentLength: 10
RoadWidth: !NormalValue { Min: 2, Max: 3, Vary: false }
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 50 }
TensorField: *base-field
Minor:
MergeSearchAngle: 12.5
MergeDistance: 2.5
SegmentLength: 2
RoadWidth: 1
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 15 }
TensorField:
!WeightedAverage
Tensors:
0.9: *base-field
0.1: !Grid { Angle: !UniformValue { Min: 1, Max: 360, Vary: true } }
"));
NetworkBuilder b = new NetworkBuilder();
Random r = new Random(24);
var m = new NamedBoxCollection();
b.Build(c.Major(r.NextDouble, m), r.NextDouble, m, new Vector2(0, 0), new Vector2(350, 350));
b.Reduce();
var regions = b.Regions();
foreach (var region in regions)
{
b.Build(c.Minor(r.NextDouble, m), r.NextDouble, m, region);
}
Console.WriteLine(b.Result.ToSvg(regions));
}
public Task <TrainingResult> Train(bool liveTracking)
{
return(Task.Run(() =>
{
int inputSize = NetworkSampleInput.InputSize(NumberOfVertices);
Random random = new Random(Seed);
Network network = NetworkBuilder.Build(new NetworkDefinition(FunctionName.ReLU, inputSize, NumberOfVertices, HiddenLayers), new He(Seed));
Optimizer optimizer = new SGDMomentum(network, LearningRate, Momentum);
Trainer trainer = new Trainer(optimizer, random);
GraphPath[] trainingGraphs = GraphGenerator.Generate(random, NumberOfGraphs, NumberOfVertices).Select(Dijkstra.Find).ToArray();
NetworkSample[] trainingData = NetworkSampleGenerator.Generate(trainingGraphs).ToArray();
MeanSquareErrorMonitor monitor = new MeanSquareErrorMonitor();
trainer.Monitors.Add(monitor);
if (liveTracking)
{
trainer.Monitors.Add(new TrainingMonitor(Name));
}
trainer.Train(trainingData, Epoches, BatchSize);
lock (ThreadLock)
{
double error = MeanSquareErrorMonitor.CalculateError(network, trainingData);
Console.WriteLine($"[{Name}] Seed: {Seed,12} Momentum: {Momentum,-6} Learning Rate: {LearningRate,-6} MSE: {error}");
}
return new TrainingResult(Name, monitor.CollectedData as IEnumerable <double>, network, trainingGraphs, trainingData);
}));
}
/// <summary>
/// Use MEF to handle View/ViewModel/Model creation and binding
/// </summary>
/// <param name="e"></param>
protected override void OnStartup(StartupEventArgs e)
{
base.OnStartup(e);
var conventions = new RegistrationBuilder();
conventions.ForType <KinectManager>()
.Export <IKinectManager>()
.SetCreationPolicy(CreationPolicy.Shared);
conventions.ForType <KinectGreenScreenViewModel>()
.Export <IKinectGreenScreenViewModel>()
.SetCreationPolicy(CreationPolicy.Shared);
conventions.ForType <KinectGreenScreenViewModel>().ImportProperty(p => p.KinectManager);
Assembly assembly = Assembly.GetExecutingAssembly();
var catalog = new AssemblyCatalog(assembly, conventions);
_container = new CompositionContainer(catalog);
Container.ComposeParts();
ISourceBlock <BitmapSource> imageSource = NetworkBuilder.Build(_container.GetExportedValue <IKinectManager>());
_container.GetExportedValue <IKinectGreenScreenViewModel>().GreenScreenImageGenerator = imageSource;
}
public AgentNetwork(AgentNetworkDefinition definition, ILayerInitializer initializer)
{
InputCoder = new AgentNetworkInputCoder(definition.ViewRadius);
var lastLayer = new NetworkLayerDefinition(definition.LastLayerActivationFunction, 1);
var networkDefinition = new NetworkDefinition(InputCoder.EncodedSize, definition.Layers.Concat(new[] { lastLayer }));
Network = NetworkBuilder.Build(networkDefinition, initializer);
}
public IActionResult FindAllPath([FromBody] FindAllPathQuery findAllPathQuery)
{
try
{
var nb = new NetworkBuilder(findAllPathQuery.SourceIds, findAllPathQuery.DestinationIds, findAllPathQuery.MaxLength, false);
nb.Build();
return(Ok(new Tuple <HashSet <Node>, HashSet <Edge> >(nb.Nodes, nb.Edges)));
}
catch (Exception e)
{
Console.WriteLine(e);
return(BadRequest(e.Message));
}
}
private static void Main(string[] args)
{
var logger = new ConsoleLogger();
try
{
var dataFolder = GetDataFolder(args);
if (string.IsNullOrEmpty(dataFolder))
{
return;
}
Control.UseNativeMKL();
var(trainX, trainY, devX, devY) = LoadData(dataFolder, logger);
bool repeat;
do
{
var randomSeed = 13;
var network = NetworkBuilder.Build(28 * 28, new LayerOptions(10, new Sigmoid()), new[]
{
new LayerOptions(30, new Sigmoid()),
//new LayerOptions(30, new Sigmoid()),
//new LayerOptions(30, new Sigmoid()),
}, randomSeed);
var trainer = new SgdTrainer(30, 10, 3.0, new QuadraticCostFunction(), logger, randomSeed);
var(randomTrainX, randomTrainY) = Shuffler.Shuffle(randomSeed, trainX, trainY);
PrintDataHistograms(trainY, devY, logger);
trainer.Train(network, randomTrainX, randomTrainY, 0.95);
DisplayTestPrecision(devX, devY, network, logger);
logger.Log("Press key to exit. \"r\" to repeat...");
var answer = Console.ReadKey();
repeat = answer.KeyChar == 'r';
} while (repeat);
}
catch (Exception e)
{
logger.Log(e.Message);
}
}
public void NetworkBuilder_GeneratesRegions()
{
var c = NetworkDescriptor.Deserialize(new StringReader(@"
!Network
Major:
MergeSearchAngle: 22.5
MergeDistance: 25
SegmentLength: 10
RoadWidth: !NormalValue { Min: 1, Max: 10, Vary: true }
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 50 }
TensorField:
!AddTensors
Tensors:
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 0, Length: 1 }, Decay: 2.5, Center: { X: 0, Y: 0 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 45, Length: 1 }, Decay: 2.5, Center: { X: 1, Y: 0 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 45, Length: 1 }, Decay: 2.5, Center: { X: 0, Y: 1 } }
- !PointDistanceDecayTensors { Tensors: !Grid { Angle: 0, Length: 1 }, Decay: 2.5, Center: { X: 1, Y: 1 } }
Minor:
MergeSearchAngle: 12.5
MergeDistance: 2.5
SegmentLength: 2
RoadWidth: 1
PriorityField: !ConstantScalars { Value: 1 }
SeparationField: !ConstantScalars { Value: 15 }
TensorField: !Grid { Angle: 0, Length: 1 }
"));
NetworkBuilder b = new NetworkBuilder();
Random r = new Random(10);
var m = new NamedBoxCollection();
b.Build(c.Major(r.NextDouble, m), r.NextDouble, m, new Vector2(0, 0), new Vector2(500, 500));
b.Reduce();
var regions = b.Regions();
Console.WriteLine(b.Result.ToSvg(regions));
}
public static Network Network(FunctionName activationFunction, int inputSize, int outputSize, params int[] hiddenLayersSizes)
{
return(NetworkBuilder.Build(new NetworkDefinition(activationFunction, inputSize, outputSize, hiddenLayersSizes), new He(0)));
}
public TicTacToeValueNetwork()
{
Network = NetworkBuilder.Build(new NetworkDefinition(FunctionName.Sigmoidal, 9, 9, new int[] { 9, 9 }), new He(0));
}
public TicTacToeValueNetwork(int[] hiddenLayerSizes, FunctionName activationFunction, Random random)
{
Network = NetworkBuilder.Build(new NetworkDefinition(activationFunction, 9, 3, hiddenLayerSizes), new He(0));
InputTransform = DefaultInputTransform;
OutputTransform = TicTacToeLabeledStateLoader.OutputTransforms.ResultProbabilities;
}
protected virtual Network CreateNeuralNetwork()
{
Random random = new Random(Properties.Seed);
return(NetworkBuilder.Build(new NetworkDefinition(FunctionName.Sigmoidal, Properties.NetworkInputSize, 1, Properties.NetworkLayersSize), new He(Properties.Seed)));
}
public TicTacToePVNetwork()
{
Network = NetworkBuilder.Build(new NetworkDefinition(FunctionName.ReLU, 9, 3, 72, 72, 72, 36, 36, 36, 18, 18), new He(0));
}
private void backgroundWorker1_DoWork(object sender, DoWorkEventArgs e)
{
bitmap = new Bitmap(200, 200);
Random random = new Random(0);
Optimizer optimizer;
Network network;
if (ActivationFunction == Function.ReLU)
{
//var network = NetworkBuilder.Build(Function.ReLU, 2, 3, 64, 64, 32, 32);
//evaluator = new NetworkEvaluator(network);
//optimizer = new SGDMomentum(evaluator, 0.001, 0.04);
network = NetworkBuilder.Build(new NetworkDefinition(FunctionName.ReLU, 2, 3, 72, 72, 72, 36, 36, 36, 18, 18), new He(0));
optimizer = new SGDMomentum(network, 0.001, 0.008);
}
else
{
network = NetworkBuilder.Build(new NetworkDefinition(FunctionName.Sigmoidal, 2, 3, 32, 8), new He(0));
optimizer = new SGDMomentum(network, 0.1, 0.8);
}
var trainer = new Trainer(optimizer, new Random(0));
var mseMonitor = new MeanSquareErrorMonitor();
trainer.Monitors.Add(mseMonitor);
int width;
int height;
lock (bitmap)
{
width = bitmap.Width;
height = bitmap.Height;
}
var imagePointList = new List <double[]>();
for (int y = 0; y < height; y++)
{
for (int x = 0; x < width; x++)
{
imagePointList.Add(new double[] { (double)x / width, (double)y / height });
}
}
trainingData = GenerateTrainingData(random);
for (int epoch = 0; epoch < 15000; epoch++)
{
//if (epoch % 300 == 0)
//{
// GenerateTrainingData(random);
//}
if (epoch % 100 == 0)
{
RefreshBitmap(network, width, height, imagePointList);
}
trainer.Train(trainingData, 1, 1);
Console.WriteLine(mseMonitor.CollectedData.Last());
}
}