public void ExecuteTest()
{
Random random = new Random(1234);
const int batches = 1000, tests = 8;
float range = (float)Math.Sqrt(3);
foreach (int num in new int[] { 32, 64, 128, 256 })
{
List <float> variance_list = new List <float>();
for (int i = 0; i < tests; i++)
{
Tensor x = Tensor.UniformRandom(Shape.Map0D(num, batches), random) * (2 * range) - range;
Tensor w = new Tensor(Shape.Kernel0D(num, 1));
var initializer = new HeUniform(w, random);
initializer.Execute();
Tensor y = Tensor.Relu(Tensor.Dense(x, w));
float variance = Tensor.Average(y * y).State[0];
variance_list.Add(variance);
}
float mean_variance = variance_list.Average();
Assert.AreEqual(1, mean_variance, 0.1f, $"{num}:{mean_variance}");
Console.WriteLine($"{num}:{mean_variance}");
}
}
public void HeUniformTest()
{
var init = new HeUniform();
var m = init.Initialize(new Matrix(2, 2));
Assert.IsTrue(Math.Abs(m.FrobeniusNorm()) > 0.01, init.Type().ToString() + " Initializer.");
}
