public void TestShapeOfSlice()
{
var v = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 13).As <float>(), "v");
AssertArray.WriteTheSame(new[] { 3 }, v[XSlicer.Range(5, 8)].Shape);
AssertArray.WriteTheSame(new[] { 3 }, v[XSlicer.Range(8, 5, -1)].Shape);
AssertArray.WriteTheSame(new[] { 4 }, v[XSlicer.Range(3, 11, 2)].Shape);
var M = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 8, 22).As <float>(), "M");
AssertArray.WriteTheSame(new[] { 5, 17 }, M[XSlicer.Range(6, 1, -1), XSlicer.From(5)].Shape);
}
public void TestShapeOfShared()
{
var v = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 13).As <float>(), "v");
AssertArray.WriteTheSame(new[] { 13 }, v.Shape);
var M = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 5, 4).As <float>(), "M");
AssertArray.WriteTheSame(new[] { 5, 4 }, M.Shape);
var V = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 5, 4, 8).As <float>(), "V");
AssertArray.WriteTheSame(new[] { 5, 4, 8 }, V.Shape);
}
public void TestShapeOfScan()
{
var W = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 5, 4).As <float>(), "W");
var X = T.Matrix <float>("X");
var acc0 = T.Shared(NN.Zeros <float>(5), "acc0");
var loop = T.Scan(fn: (x, acc) => acc + T.Dot(W, x), sequence: X, outputsInfo: acc0);
AssertArray.WriteTheSame(new[] { X.Shape[0], 5 }, loop.Shape);
var norm2 = T.Norm2(loop[-1]);
var grad = T.Grad(norm2, W);
grad.Shape.WillEqualTo(new Dim[] { 5, 4 });
}
public void TestShapeOfDot()
{
var v1 = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 13).As <float>(), "v1");
var v2 = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 13).As <float>(), "v2");
AssertArray.WriteTheSame(new[] { 13, 13 }, T.Dot(v1, v2, transposeY: true).Shape);
AssertArray.AreEqual(EmptyArray <Dim> .Value, T.Dot(v1, v2).Shape);
AssertArray.AreEqual(EmptyArray <Dim> .Value, T.Dot(v1, v2, transposeX: true).Shape);
var M1 = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 5, 13).As <float>(), "M1");
AssertArray.WriteTheSame(new[] { 5 }, T.Dot(M1, v1).Shape);
var M2 = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 13, 7).As <float>(), "M2");
AssertArray.WriteTheSame(new[] { 5, 7 }, T.Dot(M1, M2).Shape);
var M3 = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 8, 13).As <float>(), "M3");
AssertArray.WriteTheSame(new[] { 5, 8 }, T.Dot(M1, M3, transposeY: true).Shape);
}
public void TestShapeOfScanWithShared()
{
var W = T.Shared(0.2f * NN.Random.Uniform(-1.0f, 1.0f, 5, 4).As <float>(), "W");
Func <Tensor <float>, Tensor <float>, Tensor <float> > recurrence = (x, acc) =>
{
return(acc + T.Dot(W, x));
};
var X = T.Shared(NN.Zeros(13, 4), "X");
var acc0 = T.Shared(NN.Zeros <float>(5), "acc0");
var loop = T.Scan(fn: recurrence, sequences: new[] { X }, outputsInfo: acc0);
AssertArray.WriteTheSame(new [] { 13, 5 }, loop.Shape);
var norm2 = T.Norm2(loop[-1]);
var grad = T.Grad(norm2, W);
AssertArray.WriteTheSame(new[] { 5, 4 }, grad.Shape);
}