public void InputLayer_Backward()
{
var batchSize = 1;
var width = 28;
var height = 28;
var depth = 3;
var random = new Random(232);
var fanIn = width * height * depth;
var sut = new InputLayer(height, width, depth);
sut.Initialize(1, 1, 1, batchSize, Initialization.GlorotUniform, random);
var input = Matrix <float> .Build.Random(batchSize, fanIn, random.Next());
sut.Forward(input);
var delta = Matrix <float> .Build.Random(batchSize, fanIn, random.Next());
var actual = sut.Backward(delta);
var expected = delta;
MatrixAsserts.AreEqual(expected, actual);
}
public void ForwardPropagation()
{
InputLayer.Forward();
for (int i = 0; i < Layers.Count; i++)
{
Layers[i].Forward();
}
}
public void Forward(double[] input)
{
_inputLayer.SetInput(input);
_inputLayer.Forward();
foreach (var layer in _layers)
{
layer.Forward();
}
}
public void ForwardPropagation()
{
InputLayer.Forward();
foreach (MyAbstractFBLayer layer in Layers)
{
layer.Forward();
}
// Copy the end layer output to the node output
m_copyKernel.SetupExecution(LastLayer.Output.Count);
m_copyKernel.Run(LastLayer.Output.Ptr, 0, CurrentSampleOutputPtr, 0, LastLayer.Output.Count);
SamplesProcessed++;
}
public void ForwardPropagation()
{
switch (AutoencoderTask.NetworkMode)
{
case MyAutoencoderMode.TRAINING:
case MyAutoencoderMode.FORWARD_PASS:
InputLayer.Forward();
foreach (MyAbstractFBLayer layer in Layers)
{
layer.Forward();
}
// Copy the end layer output to the node output
m_copyKernel.SetupExecution(LastLayer.Output.Count);
m_copyKernel.Run(LastLayer.Output.Ptr, 0, CurrentSampleOutputPtr, 0, LastLayer.Output.Count);
SamplesProcessed++;
break;
case MyAutoencoderMode.FEATURE_ENCODING:
InputLayer.Forward();
for (int i = 0; i <= FeatureLayerPosition; i++)
{
Layers[i].Forward();
}
// Copy the featureLayer to feature output
m_copyKernel.SetupExecution(FeatureLayer.Output.Count);
m_copyKernel.Run(FeatureLayer.Output.Ptr, 0, CurrentSampleFeatureOutputPtr, 0, FeatureLayer.Output.Count);
SamplesProcessed++;
break;
case MyAutoencoderMode.FEATURE_DECODING:
// Copy the feature input to the feature layer
m_copyKernel.SetupExecution(FeatureLayer.Output.Count);
m_copyKernel.Run(CurrentSampleFeatureInputPtr, 0, FeatureLayer.Output.Ptr, 0, FeatureLayer.Output.Count);
for (int i = FeatureLayerPosition + 1; i < Layers.Count; i++)
{
Layers[i].Forward();
}
// Copy the end layer output to the node output
m_copyKernel.SetupExecution(LastLayer.Output.Count);
m_copyKernel.Run(LastLayer.Output.Ptr, 0, CurrentSampleOutputPtr, 0, LastLayer.Output.Count);
SamplesProcessed++;
break;
}
}
public void ForwardBackwardTest1()
{
Shape shape = new Shape(Shape.BWHC, -1, 20, 15, 10);
InputLayer layer = new InputLayer(shape);
for (int i = 1; i <= 3; i++)
{
Tensor x = new Tensor(null, shape.Reshape(Axis.B, i));
x.Randomize();
IList <Tensor> xs = new[] { x };
IList <Tensor> ys = layer.Forward(null, xs);
Assert.IsTrue(xs[0] == ys[0]);
}
}
public void ForwardTest2()
{
Shape shape = new Shape(new[] { 1, 20, 15, 10 });
InputLayer layer = new InputLayer(shape);
try
{
Tensor x = new Tensor(null, shape);
layer.Forward(null, new[] { x, x });
}
catch (ArgumentException e)
{
Assert.AreEqual(Properties.Resources.E_InvalidInputTensor_InvalidCount, e.Message);
throw;
}
}
public void ForwardTest1()
{
Shape shape = new Shape(new[] { 1, 20, 15, 10 });
InputLayer layer = new InputLayer(shape);
try
{
Tensor x = new Tensor(null, shape);
x.Randomize();
IList <Tensor> xs = new[] { x, x };
layer.Forward(null, xs);
}
catch (ArgumentException e)
{
Assert.AreEqual(
new ArgumentException(Properties.Resources.E_InvalidInputTensor_InvalidCount).Message,
e.Message);
throw;
}
}
public void ForwardTest3()
{
Shape shape = new Shape(Shape.BWHC, -1, 20, 15, 10);
InputLayer layer = new InputLayer(shape);
try
{
Tensor x = new Tensor(null, new[] { 1, 2, 3 });
layer.Forward(null, new[] { x });
}
catch (ArgumentException e)
{
Assert.AreEqual(
new ArgumentException(string.Format(
CultureInfo.InvariantCulture,
Properties.Resources.E_InvalidInputTensor_InvalidRank,
3,
4)).Message,
e.Message);
throw;
}
}
