public unsafe void SoftmaxBackwardOutput()
{
float[,]
x = WeightsProvider.NewFullyConnectedWeights(TensorInfo.Linear(400), 250, WeightsInitializationMode.GlorotNormal).AsSpan().AsMatrix(400, 250),
y = new float[400, 127];
for (int i = 0; i < 400; i++)
{
y[i, ThreadSafeRandom.NextInt(max: 127)] = 1;
}
OutputLayerBase
cpu = new SoftmaxLayer(TensorInfo.Linear(250), 127, WeightsInitializationMode.GlorotNormal, BiasInitializationMode.Gaussian),
gpu = new CuDnnSoftmaxLayer(cpu.InputInfo, cpu.OutputInfo.Size, cpu.Weights, cpu.Biases);
fixed(float *px = x, py = y)
{
Tensor.Reshape(px, x.GetLength(0), x.GetLength(1), out Tensor xt);
cpu.Forward(xt, out Tensor z, out Tensor a);
a.Duplicate(out Tensor a2);
Tensor.Reshape(py, y.GetLength(0), y.GetLength(1), out Tensor yt);
cpu.Backpropagate(a, yt, z);
gpu.Backpropagate(a2, yt, z);
Assert.IsTrue(a.ContentEquals(a2));
a.Free();
a2.Free();
z.Free();
}
}
public override float Loss(ComputeShader compute, Signal signal, Signal answer, bool train)
{
var predictSig = Predict(compute, signal, train);
var softmaxSig = softmax.Forward(compute, predictSig, train);
predictSig.Dispose();
return(CrossEntropyError.Loss(compute, softmaxSig, answer));
}
public void SoftmaxLayer_Forward()
{
var layer = new SoftmaxLayer();
layer.Setup(bottom, top);
layer.Forward(bottom, top);
Assert.Equal(bottom.Count, top.Count);
using (var topCpu = top.OnCpu())
using (var bottomCpu = bottom.OnCpu())
{
int count = bottom.Count;
int num = bottom.Num;
int channels = bottom.Channels;
for (int i = 0; i < num; i++)
{
double sum = 0;
for (int j = 0; j < channels; j++)
{
sum += topCpu.DataAt(i, j, 0, 0);
}
Assert.True(sum >= 0.999);
Assert.True(sum <= 1.001);
}
for (int i = 0; i < num; i++)
{
double scale = 0;
for (int j = 0; j < channels; j++)
{
scale += Math.Exp(bottomCpu.DataAt(i, j, 0, 0));
}
for (int j = 0; j < channels; j++)
{
Assert.True(topCpu.DataAt(i, j, 0, 0) + 1e-4f >= Math.Exp(bottomCpu.DataAt(i, j, 0, 0)) / scale);
Assert.True(topCpu.DataAt(i, j, 0, 0) - 1e-4f <= Math.Exp(bottomCpu.DataAt(i, j, 0, 0)) / scale);
}
}
}
}
public void SoftmaxLayer_Forward()
{
var layer = new SoftmaxLayer();
layer.Setup(bottom, top);
layer.Forward(bottom, top);
Assert.Equal(bottom.Count, top.Count);
using (var topCpu = top.OnCpu())
using (var bottomCpu = bottom.OnCpu())
{
int count = bottom.Count;
int num = bottom.Num;
int channels = bottom.Channels;
for (int i = 0; i < num; i++)
{
double sum = 0;
for (int j = 0; j < channels; j++)
sum += topCpu.DataAt(i, j, 0, 0);
Assert.True(sum >= 0.999);
Assert.True(sum <= 1.001);
}
for (int i = 0; i < num; i++)
{
double scale = 0;
for (int j = 0; j < channels; j++)
scale += Math.Exp(bottomCpu.DataAt(i, j, 0, 0));
for (int j = 0; j < channels; j++)
{
Assert.True(topCpu.DataAt(i, j, 0, 0) + 1e-4f >= Math.Exp(bottomCpu.DataAt(i, j, 0, 0)) / scale);
Assert.True(topCpu.DataAt(i, j, 0, 0) - 1e-4f <= Math.Exp(bottomCpu.DataAt(i, j, 0, 0)) / scale);
}
}
}
}
