public ConvolutionLayerConfigurtion(int numberOfKernels,
int kernelSize,
MessageShape inputMessageShape) : base(inputMessageShape)
{
NumberOfKernels = numberOfKernels;
KernelSize = kernelSize;
}
public ActivationLayerConfiguration(
MessageShape inputMessageShape,
ActivatorType activatorType)
: base(inputMessageShape)
{
ActivatorType = activatorType;
}
public ActivationForwardLayer(
ActivatorType type,
MessageShape inputMessageShape)
: base(inputMessageShape, inputMessageShape)
{
Activator = ActivatorFactory.Produce(type);
}
public BaseLayer(
MessageShape inputMessageShape,
MessageShape outputMessageShape)
{
_outputMessageShape = outputMessageShape;
_inputMessageShape = inputMessageShape;
}
public static MessageShape BuildOutputMessageShape(MessageShape inputMessageShape, int stride)
{
int size = inputMessageShape.Size % stride == 0
? inputMessageShape.Size / stride
: (inputMessageShape.Size / stride) + 1;
return(new MessageShape(size, inputMessageShape.Depth));
}
public PoolingFullLayer(
MessageShape inputMessageShape, int stride)
: base(inputMessageShape, stride)
{
_cache = new bool[
inputMessageShape.Depth,
inputMessageShape.Size,
inputMessageShape.Size];
}
public DenseLayerConfiguration(
MessageShape inputMessageShape,
bool enableBiase,
int numberOfNeurons)
: base(inputMessageShape)
{
EnableBias = enableBiase;
NumberOfNeurons = numberOfNeurons;
}
public static MessageShape BuildOutputMessageShape(
MessageShape inputMessageShape,
int kernelSize,
int numberOfKernels)
{
int size = inputMessageShape.Size - kernelSize + 1;
return(new MessageShape(size, numberOfKernels));
}
public ActivationFullLayer(
ActivatorType type,
MessageShape inputMessageShape)
: base(type, inputMessageShape)
{
_cache = inputMessageShape.Depth == 1
? _cache = new double[inputMessageShape.Size]
: _cache = new double[inputMessageShape.Depth,
inputMessageShape.Size,
inputMessageShape.Size];
}
public ConvolutionLayerConfigurtion(int numberOfKernels,
int kernelSize,
double[][,,] weights,
double[] biases,
MessageShape inputMessageShape) : base(inputMessageShape)
{
Weights = weights;
Biases = biases;
NumberOfKernels = numberOfKernels;
KernelSize = kernelSize;
}
public DenseForwardLayer(
MessageShape inputMessageShape,
int numberOfNeurons,
bool enableBiases)
: base(inputMessageShape, new MessageShape(numberOfNeurons))
{
EnableBiases = enableBiases;
NumberOfNeurons = numberOfNeurons;
Weights = Matrix.Build.Dense(inputMessageShape.Size, NumberOfNeurons);
Biases = Vector.Build.Dense(NumberOfNeurons);
}
public DenseLayerConfiguration(MessageShape inputMessageShape,
bool enableBiase,
int numberOfNeurons,
double[,] weights,
double[] biases)
: base(inputMessageShape)
{
EnableBias = enableBiase;
Weights = weights;
Biases = biases;
NumberOfNeurons = numberOfNeurons;
}
public ConvolutionForwardLayer(
MessageShape inputMessageShape,
int kernelSize,
int numberOfKernels) :
base(inputMessageShape, BuildOutputMessageShape(inputMessageShape, kernelSize, numberOfKernels))
{
KernelSize = kernelSize;
NumberOfKernels = numberOfKernels;
Kernels = new double[numberOfKernels][, , ];
Kernels.UpdateForEach <double[, , ]>(q => new double[inputMessageShape.Depth, kernelSize, kernelSize]);
Biases = new double[numberOfKernels];
}
public ConvolutionFullLayer(
MessageShape inputMessageShape,
int kernelSize,
int numberOfkernels,
Optimizer optimizer)
: base(inputMessageShape, kernelSize, numberOfkernels)
{
_cache = new double[inputMessageShape.Depth, inputMessageShape.Size, inputMessageShape.Size];
_weightOptimizers = Helper.InitializeKernelOptimizers(
inputMessageShape.Depth, numberOfkernels, kernelSize, optimizer);
_biasOptimizers = Helper.InitializeBiasOptimizers(numberOfkernels, optimizer);
}
public DenseFullLayer(
MessageShape inputMessageShape,
int numberOfNeurons,
bool enableBiases,
Optimizer optimizer)
: base(inputMessageShape, numberOfNeurons, enableBiases)
{
_weightOptimizers = new Optimizer[inputMessageShape.Size, numberOfNeurons];
_weightOptimizers.UpdateForEach <Optimizer>((q, i) => optimizer.Clone() as Optimizer);
_biasOptimizers = new Optimizer[numberOfNeurons];
_biasOptimizers.UpdateForEach <Optimizer>((q, i) => optimizer.Clone() as Optimizer);
_cache = Vector.Build.Dense(inputMessageShape.Size);
}
static MessageItemRowNotificationHandler()
{
Shape[] shapes = new Shape[]
{
ItemShape.CreateShape(),
MessageShape.CreateShape(),
TaskShape.CreateShape()
};
ResponseShape responseShape = WellKnownShapes.ResponseShapes[WellKnownShapeName.MailListItem];
MessageItemRowNotificationHandler.defaultSubscriptionProperties = RowNotificationHandler.GetPropertyDefinitionsForResponseShape(shapes, responseShape, new PropertyDefinition[0]);
MessageItemRowNotificationHandler.normalizedSubjectPropertyDefinition = WellKnownProperties.NormalizedSubject.ToPropertyDefinition();
MessageItemRowNotificationHandler.lastVerbExecutedPropertyDefinition = WellKnownProperties.LastVerbExecuted.ToPropertyDefinition();
MessageItemRowNotificationHandler.lastVerbExecutionTimePropertyDefinition = WellKnownProperties.LastVerbExecutionTime.ToPropertyDefinition();
}
private static PropertyDefinition[] GetSubscriptionProperties(IFeaturesManager featuresManager)
{
string text = WellKnownShapeName.MailListItem.ToString();
ItemResponseShape itemResponseShape = new ItemResponseShape();
itemResponseShape.BaseShape = ShapeEnum.IdOnly;
ItemResponseShape responseShape = Global.ResponseShapeResolver.GetResponseShape <ItemResponseShape>(text, itemResponseShape, featuresManager);
if (responseShape == null)
{
ExTraceGlobals.NotificationsCallTracer.TraceError <string>((long)text.GetHashCode(), "[MessageItemRowNotificationHandler.GetSubscriptionProperties] Unable to resolve shapeName: {0} with features manager", text);
return(MessageItemRowNotificationHandler.defaultSubscriptionProperties);
}
Shape[] shapes = new Shape[]
{
ItemShape.CreateShape(),
MessageShape.CreateShape(),
TaskShape.CreateShape()
};
return(RowNotificationHandler.GetPropertyDefinitionsForResponseShape(shapes, responseShape, new PropertyDefinition[0]));
}
public static MessageShape ComputeOutputMessageShape(MessageShape shape)
{
return(new MessageShape(shape.Size * shape.Size * shape.Depth));
}
public PoolingForwardLayer(MessageShape inputMessageShape, int stride)
: base(inputMessageShape, BuildOutputMessageShape(inputMessageShape, stride))
{
Stride = stride;
}
public FlattenFullLayer(MessageShape inputMessageShape) : base(inputMessageShape)
{
}
public InputForwardLayer(
MessageShape inputMessageShape)
: base(inputMessageShape, inputMessageShape)
{
}
public LayerConfiguration(MessageShape inputMessageShape)
{
MessageShape = inputMessageShape;
}
public InputFullLayer(MessageShape inputMessageShape)
: base(inputMessageShape)
{
}
public FlattenForwardLayer(MessageShape inputMessageShape)
: base(inputMessageShape, ComputeOutputMessageShape(inputMessageShape))
{
}
public SoftMaxFullLayer(MessageShape inputMessageShape)
: base(inputMessageShape)
{
_cache = Vector.Build.Dense(inputMessageShape.Size);
}
public InputLayerConfiguration(MessageShape inputMessageShape)
: base(inputMessageShape)
{
}
public SoftmaxLayerConfiguration(MessageShape inputMessageShape)
: base(inputMessageShape)
{
}
public FlattenLayerConfiguration(MessageShape inputMessageShape)
: base(inputMessageShape)
{
}
public PoolingLayerConfiguration(int kernelSize, MessageShape inputMessageShape)
: base(inputMessageShape)
{
KernelSize = kernelSize;
}
public SoftMaxForwardLayer(
MessageShape inputMessageShape)
: base(inputMessageShape, inputMessageShape)
{
}
