public unsafe void FullyConnectedForward()
{
FullyConnectedLayer fc = new FullyConnectedLayer(TensorInfo.Linear(231), 125, ActivationType.Sigmoid, WeightsInitializationMode.GlorotUniform, BiasInitializationMode.Gaussian);
Tensor x = CreateRandomTensor(400, fc.InputInfo.Size);
fixed(float *pw = fc.Weights, pb = fc.Biases)
{
Tensor.Reshape(pw, fc.InputInfo.Size, fc.OutputInfo.Size, out Tensor w);
Tensor.Reshape(pb, 1, fc.OutputInfo.Size, out Tensor b);
Tensor.New(x.Entities, fc.OutputInfo.Size, out Tensor y1);
CpuDnn.FullyConnectedForward(x, w, b, y1);
Gpu gpu = Gpu.Default;
using (DeviceMemory <float>
x_gpu = gpu.AllocateDevice(x),
w_gpu = gpu.AllocateDevice(w),
b_gpu = gpu.AllocateDevice(b),
y_gpu = gpu.AllocateDevice <float>(y1.Size))
{
Dnn.Get(gpu).FullyConnectedForward(x.Entities, x.Length, y1.Length, x_gpu.Ptr, w_gpu.Ptr, b_gpu.Ptr, y_gpu.Ptr);
y_gpu.CopyToHost(y1.Entities, y1.Length, out Tensor y2);
Assert.IsTrue(y1.ContentEquals(y2));
Tensor.Free(x, y1, y2);
}
}
}
public unsafe void FullyConnectedBackwardData()
{
FullyConnectedLayer fc = new FullyConnectedLayer(TensorInfo.Linear(231), 125, ActivationType.Sigmoid, WeightsInitializationMode.GlorotUniform, BiasInitializationMode.Gaussian);
Tensor dy = CreateRandomTensor(400, fc.OutputInfo.Size);
fixed(float *pw = fc.Weights, pb = fc.Biases)
{
Tensor.Reshape(pw, fc.InputInfo.Size, fc.OutputInfo.Size, out Tensor w);
Tensor.Reshape(pb, 1, fc.OutputInfo.Size, out Tensor b);
Tensor.New(dy.Entities, fc.InputInfo.Size, out Tensor dx1);
CpuDnn.FullyConnectedBackwardData(w, dy, dx1);
Gpu gpu = Gpu.Default;
using (DeviceMemory <float>
dy_gpu = gpu.AllocateDevice(dy),
w_gpu = gpu.AllocateDevice(w),
dx_gpu = gpu.AllocateDevice <float>(dx1.Size))
{
Dnn.Get(gpu).FullyConnectedBackwardData(dy.Entities, fc.InputInfo.Size, fc.OutputInfo.Size, dy_gpu.Ptr, w_gpu.Ptr, dx_gpu.Ptr);
dx_gpu.CopyToHost(dx1.Entities, dx1.Length, out Tensor dx2);
Assert.IsTrue(dx1.ContentEquals(dx2));
Tensor.Free(dy, dx1, dx2);
}
}
}
/// <inheritdoc/>
public override unsafe void Forward(Span <Tensor> inputs, out Tensor z, out Tensor a)
{
Descriptor.Set4D(DataType.FLOAT, TensorFormat.CUDNN_TENSOR_NCHW, inputs[0].Entities, inputs[0].Length, 1, 1);
fixed(Tensor *p = inputs)
{
Tensor.New(p->Entities, p->Length, out z);
using (DeviceMemory <float> y_gpu = DnnInstance.Gpu.AllocateDevice(*p))
{
// Sum the inputs
for (int i = 1; i < inputs.Length; i++)
{
using (DeviceMemory <float> x_gpu = DnnInstance.Gpu.AllocateDevice(p[i]))
DnnInstance.AddTensor(1, Descriptor, x_gpu.Ptr, 1, Descriptor, y_gpu.Ptr);
}
y_gpu.CopyToHost(p[0].Entities, p[0].Length, out z);
// Activation
DnnInstance.ActivationForward(p[0].Entities, p[0].Length, y_gpu.Ptr, y_gpu.Ptr, ActivationFunctions.Activation);
y_gpu.CopyToHost(z.Entities, z.Length, out a);
}
}
}
public void ActivationForward()
{
Tensor x = CreateRandomTensor(400, 1200);
Tensor.Like(x, out Tensor y1);
CpuDnn.ActivationForward(x, ActivationFunctions.Sigmoid, y1);
Gpu gpu = Gpu.Default;
using (DeviceMemory <float>
x_gpu = gpu.AllocateDevice(x),
y_gpu = gpu.AllocateDevice <float>(x.Size))
{
Dnn.Get(gpu).ActivationForward(x.Entities, x.Length, x_gpu.Ptr, y_gpu.Ptr, ActivationFunctions.Sigmoid);
y_gpu.CopyToHost(y1.Entities, y1.Length, out Tensor y2);
Assert.IsTrue(y1.ContentEquals(y2));
Tensor.Free(x, y1, y2);
}
}
public void ActivationBackward()
{
Tensor
x = CreateRandomTensor(400, 1200),
dy = CreateRandomTensor(400, 1200);
Tensor.Like(x, out Tensor dx1);
CpuDnn.ActivationBackward(x, dy, ActivationFunctions.SigmoidPrime, dx1);
Gpu gpu = Gpu.Default;
using (DeviceMemory <float>
x_gpu = gpu.AllocateDevice(x),
dy_gpu = gpu.AllocateDevice(dy))
{
Dnn.Get(gpu).ActivationBackward(x.Entities, x.Length, x_gpu.Ptr, dy_gpu.Ptr, ActivationFunctions.SigmoidPrime, dy_gpu.Ptr);
dy_gpu.CopyToHost(dy.Entities, dy.Length, out Tensor dx2);
Assert.IsTrue(dx1.ContentEquals(dx2));
Tensor.Free(x, dy, dx1, dx2);
}
}
public void FullyConnectedBackwardFilter()
{
FullyConnectedLayer fc = new FullyConnectedLayer(TensorInfo.Linear(231), 125, ActivationType.Sigmoid, WeightsInitializationMode.GlorotUniform, BiasInitializationMode.Gaussian);
Tensor
x = CreateRandomTensor(400, fc.InputInfo.Size),
dy = CreateRandomTensor(x.Entities, fc.OutputInfo.Size);
Tensor.New(fc.InputInfo.Size, fc.OutputInfo.Size, out Tensor dJdw1);
CpuDnn.FullyConnectedBackwardFilter(x, dy, dJdw1);
dJdw1.Reshape(1, dJdw1.Size, out dJdw1);
Gpu gpu = Gpu.Default;
using (DeviceMemory <float>
x_gpu = gpu.AllocateDevice(x),
dy_gpu = gpu.AllocateDevice(dy),
djdb_gpu = gpu.AllocateDevice <float>(fc.Weights.Length))
{
Dnn.Get(gpu).FullyConnectedBackwardFilter(x.Entities, fc.InputInfo.Size, fc.OutputInfo.Size, x_gpu.Ptr, dy_gpu.Ptr, djdb_gpu.Ptr);
djdb_gpu.CopyToHost(1, fc.Weights.Length, out Tensor dJdw2);
Assert.IsTrue(dJdw1.ContentEquals(dJdw2));
Tensor.Free(x, dy, dJdw1, dJdw2);
}
}
