public static Tensor <T> Create <T>(Tensor <T> x, Tensor <int> indexes, int axis, Dim[] shape)
{
x.AssertOfDim(shape.Length);
for (int i = 0; i < x.NDim; ++i)
{
if (i != axis)
{
ShapeExtension.Bind(ref x.Shape[i], ref shape[i]);
}
}
return(new UnArgmax <T>(x, indexes, axis, shape));
}
static Dim[] EinsteinShape(Dim[] shapeX, Dim[] shapeY, Einstein[] einstein)
{
var ndim = einstein.Sum(e => e.axisZ == null ? 0 : 1);
var shapeZ = new Dim[ndim];
foreach (var e in einstein)
{
switch (e.mode)
{
case EinsteinMode.INNER:
ShapeExtension.Bind(ref shapeX[(int)e.axisX], ref shapeY[(int)e.axisY]);
if (e.axisZ != null)
{
shapeZ[(int)e.axisZ] = 1;
}
break;
case EinsteinMode.ELEMENTWISE:
ShapeExtension.Bind(ref shapeX[(int)e.axisX], ref shapeY[(int)e.axisY]);
shapeZ[(int)e.axisZ] = shapeX[(int)e.axisX];
break;
case EinsteinMode.OUTERX:
shapeZ[(int)e.axisZ] = shapeX[(int)e.axisX];
break;
case EinsteinMode.OUTERY:
shapeZ[(int)e.axisZ] = shapeY[(int)e.axisY];
break;
case EinsteinMode.SUMX:
case EinsteinMode.SUMY:
if (e.axisZ != null)
{
shapeZ[(int)e.axisZ] = 1;
}
break;
}
}
return(shapeZ);
}
protected Concat(int axis, params Tensor <T>[] inputs) : base("Concat", (Scalar <int>)axis, inputs.ToStructArray())
{
int ndim = inputs.Max(i => i.NDim);
if (axis < 0)
{
axis += ndim;
}
foreach (var x in inputs)
{
x.AssertOfDim(ndim);
}
this.Shape = new Dim[ndim];
for (int a = 0; a < ndim; ++a)
{
Shape[a] = inputs[0].Shape[a];
}
_slices = new XSlice[inputs.Length];
_slices[0] = Range(0, this.Shape[axis]);
for (int i = 1; i < inputs.Length; ++i)
{
inputs[i].AssertOfDim(ndim);
for (int a = 0; a < ndim; ++a)
{
if (a != axis)
{
ShapeExtension.Bind(ref inputs[i].Shape[a], ref this.Shape[a]);
}
}
var start = Shape[axis];
Shape[axis] += inputs[i].Shape[axis];
_slices[i] = Range(start, Shape[axis]);
}
_inputs = inputs;
_axis = axis;
}
public static Tensor <float> Create(Tensor <float> a, IEnumerable <int> axesA, Tensor <float> b, IEnumerable <int> axesB)
{
var removeA = axesA.Select(i => i < 0 ? i + a.NDim : i).ToArray();
var removeB = axesB.Select(i => i < 0 ? i + b.NDim : i).ToArray();
int n = removeA.Length;
if (removeB.Length != n)
{
throw new RankException(string.Format(
"The axes parameters of TensorDot should have the same size. Found [{0}] and [{1}].",
string.Join(", ", removeA.AsEnumerable()), string.Join(", ", removeB.AsEnumerable())));
}
for (int d = 0; d < n; ++d)
{
ShapeExtension.Bind(ref a.Shape[removeA[d]], ref b.Shape[removeB[d]]);
}
var keptX = Enumerable.Range(0, a.NDim).Where(d => !removeA.Contains(d)).ToArray();
var keptY = Enumerable.Range(0, b.NDim).Where(d => !removeB.Contains(d)).ToArray();
// Move the axes to sum over to the end of "a"
var keptA = Enumerable.Range(0, a.NDim).Where(d => !removeA.Contains(d));
var at = a.DimShuffle(keptA.Concat(removeA).ToArray());
// Move the axes to sum over to the front of "b"
var keptB = Enumerable.Range(0, b.NDim).Where(d => !removeB.Contains(d));
var bt = b.DimShuffle(removeB.Concat(keptB).ToArray());
var resultShape = keptA.Select(axis => a.Shape[axis]).Concat(keptB.Select(axis => b.Shape[axis])).ToArray();
var a2d = Reshape2D(at, a.NDim - n);
var b2d = Reshape2D(bt, n);
var res = Op.Dot(a2d, b2d).Reshape(resultShape);
res.Comment = "TensorDot";
return(res);
//return new TensorDot(a, ToAxes(_axesX), b, ToAxes(_axesY));
}
private Dot(Tensor <float> x, Tensor <float> y, bool transposeX = false, bool transposeY = false)
: base("Dot", x, y, transposeX.Named("transA"), transposeY.Named("transB"))
{
this.TransposeX = transposeX;
this.TransposeY = transposeY;
if (x.NDim == 1 && y.NDim == 1)
{
// rowV dot rowV (forbidden)
if (TransposeX && TransposeY)
{
throw Rank(this, "Can't dot two row vectors: {0} and {1}");
}
// colV dot colV (used as inner product)
if (!TransposeX && !TransposeY)
{
TransposeX = true;
}
// colV dot rowV (outer product)
if (!TransposeX && TransposeY)
{
_shape = new Dim[] { x.Shape[0], y.Shape[0] }
}
;
// rowV dot colV (inner product)
if (TransposeX && !TransposeY)
{
if (!x.Shape[0].CanEqualTo(y.Shape[0]))
{
throw Rank(this, "Can't dot {0} with {1}");
}
ShapeExtension.Bind(ref x.Shape[0], ref y.Shape[0]);
_shape = new Dim[] { };
}
return;
}
// mat dot colV
if (y.NDim == 1)
{
TransposeY = false;
if (!x.Shape[TransposeX ? 0 : x.NDim - 1].CanEqualTo(y.Shape[0]))
{
throw Rank(this, "Can't dot {0} with {1}");
}
ShapeExtension.Bind(ref x.Shape[TransposeX ? 0 : x.NDim - 1], ref y.Shape[0]);
_shape = (TransposeX ? x.Shape.Reverse().ToArray() : x.Shape).DropRight(1);
return;
}
else if (y.NDim == 0)
{
TransposeY = false;
_shape = x.Shape.ToArray();
}
// mat dot mat
else
{
if (x.NDim == 1)
{
TransposeX = true;
}
_shape = new Dim[x.NDim + y.NDim - 2];
var axisX = x.Shape[TransposeX ? 0 : x.NDim - 1];
var axisY = y.Shape[TransposeY ? 1 : y.NDim - 2];
if (!axisX.CanEqualTo(axisY))
{
throw Rank(this, "Can't dot {0} with {1}");
}
ShapeExtension.Bind(ref axisX, ref axisY);
// copy x dims but (n - 1)
if (!TransposeX)
{
for (int i = 0; i < x.NDim - 1; ++i)
{
_shape[i] = x.Shape[i];
}
}
else
{
for (int i = 0; i < x.NDim - 1; ++i)
{
_shape[i] = x.Shape[x.NDim - 1 - i];
}
}
// copy y dims but (n - 2)
if (!TransposeY)
{
for (int i = 0; i < y.NDim - 2; ++i)
{
_shape[x.NDim + i] = y.Shape[i];
}
_shape[x.NDim + y.NDim - 3] = y.Shape[y.NDim - 1];
}
else
{
for (int i = 0; i < y.NDim - 2; ++i)
{
_shape[x.NDim + i] = y.Shape[y.NDim - 1 - i];
}
_shape[x.NDim + y.NDim - 3] = y.Shape[0];
}
}
}