Struct TensorShape

Represents the shape of a tensor.

Inherited Members
Namespace: Unity.Sentis
Assembly: Unity.Sentis.dll
Syntax
[Serializable]
public struct TensorShape

Constructors

TensorShape(int)

Initializes and returns an instance of TensorShape with a rank of 1: (d0).

For example (9).

Declaration
public TensorShape(int d0)
Parameters
Type Name Description
int d0

Length of axis 0.

TensorShape(int, int)

Initializes and returns an instance of TensorShape with a rank of 2: (d1, d0).

For example (8, 9).

Declaration
public TensorShape(int d1, int d0)
Parameters
Type Name Description
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int)

Initializes and returns an instance of TensorShape with a rank of 3: (d2, d1, d0).

For example (7, 8, 9).

Declaration
public TensorShape(int d2, int d1, int d0)
Parameters
Type Name Description
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int, int)

Initializes and returns an instance of TensorShape with a rank of 4: (d3, d2, d1, d0).

For example (6, 7, 8, 9).

Declaration
public TensorShape(int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d3

Length of axis 3.
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int, int, int)

Initializes and returns an instance of TensorShape with a rank of 5: (d4, d3, d2, d1, d0).

For example (5, 6, 7, 8, 9).

Declaration
public TensorShape(int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d4

Length of axis 4.
int d3

Length of axis 3.
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int, int, int, int)

Initializes and returns an instance of TensorShape with a rank of 6: (d5, d4, d3, d2, d1, d0).

For example (4, 5, 6, 7, 8, 9).

Declaration
public TensorShape(int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d5

Length of axis 5.
int d4

Length of axis 4.
int d3

Length of axis 3.
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int, int, int, int, int)

Initializes and returns an instance of TensorShape with a rank of 7: (d6, d5, d4, d3, d2, d1, d0).

For example (3, 4, 5, 6, 7, 8, 9).

Declaration
public TensorShape(int d6, int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d6

Length of axis 6.
int d5

Length of axis 5.
int d4

Length of axis 4.
int d3

Length of axis 3.
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(int, int, int, int, int, int, int, int)

Initializes and returns an instance of TensorShape with a rank of 8: (d7, d6, d5, d4, d3, d2, d1, d0).

For example (2, 3, 4, 5, 6, 7, 8, 9).

Declaration
public TensorShape(int d7, int d6, int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d7

Length of axis 7.
int d6

Length of axis 6.
int d5

Length of axis 5.
int d4

Length of axis 4.
int d3

Length of axis 3.
int d2

Length of axis 2.
int d1

Length of axis 1.
int d0

Length of axis 0.

TensorShape(ReadOnlySpan<int>)

Initializes and returns an instance of TensorShape with a given shape. For example: TensorShape(new [] {3, 4, 5, 6}) returns a tensor with a shape of (3, 4, 5, 6).

Declaration
public TensorShape(ReadOnlySpan<int> shape)
Parameters
Type Name Description
ReadOnlySpan<int> shape

The shape as a span.

TensorShape(TensorShape)

Returns a copy of another TensorShape.

Declaration
public TensorShape(TensorShape shape)
Parameters
Type Name Description
TensorShape shape

The shape to copy.

Fields

maxRank

The maximum rank a TensorShape can have.

Declaration
public const int maxRank = 8
Field Value
Type Description
int

Properties

this[int]

Gets or sets the tensor shape at a given axis. A negative axis counts backwards from the inner dimension.

Declaration
public int this[int axis] { get; set; }
Parameters
Type Name Description
int axis

The axis to get or set.
Property Value
Type Description
int

length

The number of elements represented by the TensorShape. For example a shape of (1, 2, 3, 4) represents 24 elements: 1 * 2 * 3 * 4.

Declaration
public int length { get; }
Property Value
Type Description
int

rank

The rank of a TensorShape. For example, a tensor of shape (5) has a rank of 1. A tensor of shape (7, 3, 5) has a rank of 3.

Declaration
public int rank { get; }
Property Value
Type Description
int

Methods

Axis(int)

Returns the positive axis corresponding to a given axis. Negative axes counts backwards from the inner dimension.

Declaration
public int Axis(int axis)
Parameters
Type Name Description
int axis

The axis to wrap.
Returns
Type Description
int

The positive axis.

Broadcast(TensorShape)

Creates a TensorShape by applying numpy-style broadcasting between this and other.

Sentis broadcasts shapes from innermost to outermost dimensions. Two dimensions are compatible when they're equal, or one of the dimensions is 1.

Declaration
public TensorShape Broadcast(TensorShape other)
Parameters
Type Name Description
TensorShape other

The other tensor shape which which to broadcast.
Returns
Type Description
TensorShape

The broadcast tensor shape.

BroadcastToRank(int)

Creates a TensorShape of a given rank and with the same inner dimensions by adding outer dimensions of size 1 when needed.

For example, if the TensorShape is (256, 256, 3), and the value of rank is 5, the method returns (1, 1, 256, 256, 3).

Declaration
public TensorShape BroadcastToRank(int rank)
Parameters
Type Name Description
int rank

The rank to which to broadcast to.
Returns
Type Description
TensorShape

The broadcast tensor shape.

Concat(TensorShape, int)

Creates a TensorShape by concatenating this TensorShape with other along a given axis. The dimensions of the shapes must be equal, except at axis.

For example if this is (2, 3, 4, 5), other is (2, 2, 4, 5), and the value of axis is 1, the method returns (2, 5, 4, 5).

Declaration
public TensorShape Concat(TensorShape other, int axis)
Parameters
Type Name Description
TensorShape other

The other tensor shape which which to concatenate.
int axis

The axis along which to concatenate.
Returns
Type Description
TensorShape

The concatenated tensor shape.

Equals(object)

Determines whether the specified object is equal to the current TensorShape.

Declaration
public override bool Equals(object obj)
Parameters
Type Name Description
object obj

The object to compare to the shape.
Returns
Type Description
bool

Whether the object is equal to the shape.
Overrides

Flatten(int)

Creates a TensorShape by duplicating this and reshaping to a 2D matrix. The dimensions before axis are collapsed into the outer dimension and the remaining axes are collapsed into the inner dimension.

For example, if this is (2, 3, 4), and the value of axis is 2, the method returns (2 * 3, 4).

Declaration
public TensorShape Flatten(int axis = 1)
Parameters
Type Name Description
int axis

The axis at which to flatten.
Returns
Type Description
TensorShape

The flattened tensor shape.

GetHashCode()

Serves as the default hash function.

Declaration
public override int GetHashCode()
Returns
Type Description
int

The hash code of the tensor shape.
Overrides

HasZeroDims()

Calculates whether any axes are length 0. In this case the length is also 0.

Declaration
public bool HasZeroDims()
Returns
Type Description
bool

Whether the shape has any axes that are length 0.

Length(int)

Returns the number of elements represented by the TensorShape, starting from a given axis. A negative axis counts backwards from the inner dimension.

Declaration
public int Length(int start)
Parameters
Type Name Description
int start

The first axis to count length from.
Returns
Type Description
int

The number of elements in the shape.

Length(int, int)

Returns the number of elements represented by the TensorShape, between the start and end axes. Negative axes counts backwards from the inner dimension.

Declaration
public int Length(int start, int end)
Parameters
Type Name Description
int start

The first axis to count length from.
int end

The exclusive final axis to count length to.
Returns
Type Description
int

The number of elements in the shape.

MatMul(TensorShape)

Creates a TensorShape that results from performing a matrix multiplication between this and other with numpy-style broadcasting. For example, if this is (5, 2, 3), and other is (1, 3, 4), the method returns (5, 2, 4).

Declaration
public TensorShape MatMul(TensorShape other)
Parameters
Type Name Description
TensorShape other

The right hand tensor shape for the MatMul.
Returns
Type Description
TensorShape

The resultant tensor shape.

Ones(int)

Creates a TensorShape with a given rank where all of the dimensions are 1. For example if rank is 3, the method returns (1, 1, 1).

Declaration
public static TensorShape Ones(int rank)
Parameters
Type Name Description
int rank

The rank of the tensor shape.
Returns
Type Description
TensorShape

The created tensor shape.

Pad(ReadOnlySpan<int>)

Creates a TensorShape by padding axes. For example, if this is (1, 2, 3), and pads is {0, 0, 1, 0, 2, 2}, the method returns (1, 3, 7).

Declaration
public TensorShape Pad(ReadOnlySpan<int> pads)
Parameters
Type Name Description
ReadOnlySpan<int> pads

The lower and upper padding values for each dimension. For example [pad_left, pad_right] for 1D, or [pad_top, pad_bottom, pad_left, pad_right] for 2D.
Returns
Type Description
TensorShape

The padded tensor shape.

RavelIndex(int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (6,7) => 1 * (7) + 5 = 12

index: (1,5)

Declaration
public int RavelIndex(int d1, int d0)
Parameters
Type Name Description
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (5,6,7) => 2 * (6*7) + 1 * (7) + 5 = 96

index: (2,1,5)

Declaration
public int RavelIndex(int d2, int d1, int d0)
Parameters
Type Name Description
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (4,5,6,7) => 3 * (567) + 2 * (6*7) + 1 * (7) + 5 = 726

index: (3,2,1,5)

Declaration
public int RavelIndex(int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d3

The index along axis 3.
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (3,4,5,6,7) => 0 * (4567) + 3 * (567) + 2 * (67) + 1 * (7) + 5 = 726

index: (0,3,2,1,5)

Declaration
public int RavelIndex(int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d4

The index along axis 4.
int d3

The index along axis 3.
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int, int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (2,3,4,5,6,7) => 1 * (34567) + 0 * (4567) + 3 * (567) + 2 * (67) + 1 * (7) + 5 = 3246

index: (1,0,3,2,1,5)

Declaration
public int RavelIndex(int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d5

The index along axis 5.
int d4

The index along axis 4.
int d3

The index along axis 3.
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int, int, int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (1,2,3,4,5,6,7) => 0 * (234567) + 1 * (34567) + 0 * (4567) + 3 * (567) + 2 * (6*7) + 1 * (7) + 5 = 3246

index: (0,1,0,3,2,1,5)

Declaration
public int RavelIndex(int d6, int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d6

The index along axis 6.
int d5

The index along axis 5.
int d4

The index along axis 4.
int d3

The index along axis 3.
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

RavelIndex(int, int, int, int, int, int, int, int)

Converts the indexes of the TensorShape into a flat index.

shape: (5,1,2,3,4,5,6,7) => 2 * (1234567) + 0 * (234567) + 1 * (34567) + 0 * (4567) + 3 * (567) + 2 * (6*7) + 1 * (7) + 5 = 13326

index: (2,0,1,0,3,2,1,5)

Declaration
public int RavelIndex(int d7, int d6, int d5, int d4, int d3, int d2, int d1, int d0)
Parameters
Type Name Description
int d7

The index along axis 7.
int d6

The index along axis 6.
int d5

The index along axis 5.
int d4

The index along axis 4.
int d3

The index along axis 3.
int d2

The index along axis 2.
int d1

The index along axis 1.
int d0

The index along axis 0.
Returns
Type Description
int

The raveled index.

Reduce(int, bool)

Removes a dimension at axis. For example, if this is (2, 3, 4, 5), the value of axis is 1, and the value of keepDim is true, the method returns (2, 1, 4, 5).

Declaration
public TensorShape Reduce(int axis, bool keepDim = true)
Parameters
Type Name Description
int axis

The axis along which to reduce.
bool keepDim

When the value is true, Sentis replaces the dimension with 1.
Returns
Type Description
TensorShape

The reduced tensor shape.

Reduce(ReadOnlySpan<int>, bool)

Creates a TensorShape by removing the dimensions at axes. For example, if this is (2, 3, 4, 5), axis is {1, 2} and the value of keepDim is true, the method returns (2, 1, 1, 5).

Declaration
public TensorShape Reduce(ReadOnlySpan<int> axes, bool keepDim = true)
Parameters
Type Name Description
ReadOnlySpan<int> axes

The axes along which to reduce.
bool keepDim

When the value is true, Sentis replaces the reduced axes with 1. Otherwise Sentis removes the reduced axes.
Returns
Type Description
TensorShape

The reduced tensor shape.

Reshape(ReadOnlySpan<int>, bool)

Creates a TensorShape by duplicating this and reshaping the dimensions to those given.

If a dimension in the shape array is -1, Sentis infers the value from the size of the TensorShape and the remaining dimensions. Only one dimension can be -1.

Declaration
public TensorShape Reshape(ReadOnlySpan<int> shape, bool allowZero = false)
Parameters
Type Name Description
ReadOnlySpan<int> shape

The new shape as a span of integers.
bool allowZero

When the value is true, Sentis sets a dimension to zero if the new shape includes a zero. Otherwise Sentis retains the corresponding size at that axis from the original shape.
Returns
Type Description
TensorShape

Slice(ReadOnlySpan<int>, ReadOnlySpan<int>, ReadOnlySpan<int>, ReadOnlySpan<int>)

Creates a TensorShape that results from slicing this along given axes with given starts, ends, and steps.

Declaration
public TensorShape Slice(ReadOnlySpan<int> starts, ReadOnlySpan<int> ends, ReadOnlySpan<int> axes, ReadOnlySpan<int> steps)
Parameters
Type Name Description
ReadOnlySpan<int> starts

The start indices along each of the axes.
ReadOnlySpan<int> ends

The end indices along each of the axes.
ReadOnlySpan<int> axes

The optional axes along which to slice. The default value is [0, 1, 2...rank-1].
ReadOnlySpan<int> steps

The optional step sizes for each of the axes. The default value is [1, 1, 1...1]
Returns
Type Description
TensorShape

The sliced tensor shape.

Squeeze()

Creates a TensorShape by duplicating this and removing the dimensions of size 1. For example, if this is (5, 1, 3, 1), the method returns (5, 3).

Declaration
public TensorShape Squeeze()
Returns
Type Description
TensorShape

The squeezed tensor shape.

Squeeze(int)

Creates a TensorShape by duplicating this and removing the given axis of size 1. For example, if this is (5, 1, 3, 1), and axis is 1, the method returns (5, 3, 1).

Declaration
public TensorShape Squeeze(int axis)
Parameters
Type Name Description
int axis

The axis to squeeze.
Returns
Type Description
TensorShape

The squeezed tensor shape.

Squeeze(ReadOnlySpan<int>)

Creates a TensorShape by duplicating this and removing the given axes of size 1. For example, if this is (5, 1, 3, 1), and axes is {1, -1}, the method returns (5, 3).

Declaration
public TensorShape Squeeze(ReadOnlySpan<int> axes)
Parameters
Type Name Description
ReadOnlySpan<int> axes

The axes to squeeze.
Returns
Type Description
TensorShape

The squeezed tensor shape.

Strides(int)

Returns the product of the dimensions of the tensor shape after a given axis. Negative axes counts backwards from the inner dimension.

The strides of a tensor tell us how many elements we have to skip in flattened memory to move to the next position along a given index.

Declaration
public int Strides(int axis)
Parameters
Type Name Description
int axis

The axis to calculate the stride length at.
Returns
Type Description
int

The stride length at the axis.

Tile(ReadOnlySpan<int>)

Creates a TensorShape by repeating this TensorShape a number of times along each axis.

Declaration
public TensorShape Tile(ReadOnlySpan<int> repeats)
Parameters
Type Name Description
ReadOnlySpan<int> repeats

The repeat counts along each axis as a span of integers.
Returns
Type Description
TensorShape

The tiled tensor shape.

ToArray()

Returns the TensorShape as an array of integers. For example if the TensorShape is (5, 2, 3, 4), the method returns new[] {5, 2, 3, 4}.

Declaration
public int[] ToArray()
Returns
Type Description
int[]

An integer array representation of the shape.

ToString()

Returns a string that represents the TensorShape.

Declaration
public override string ToString()
Returns
Type Description
string

String representation of shape.
Overrides

Transpose()

Creates a TensorShape by reversing axes. For example, if this is (6, 7, 8, 9), the method returns (9, 8, 7, 6).

Declaration
public TensorShape Transpose()
Returns
Type Description
TensorShape

The transposed tensor shape.

Transpose(ReadOnlySpan<int>)

Creates a TensorShape by permuting axes. For example, if this is (6, 7, 8, 9), and permutations is {3, 0, 1, 2}, the method returns (9, 6, 7, 8).

Declaration
public TensorShape Transpose(ReadOnlySpan<int> permutations)
Parameters
Type Name Description
ReadOnlySpan<int> permutations

An array indexing the new tensor axis from the old ones.
Returns
Type Description
TensorShape

The transposed tensor shape.

Unsqueeze(int)

Creates a TensorShape by duplicating this and inserting a dimension of size one at a given axis. For example if this is (2), and the value of axis is 0, the method returns (1, 2).

Declaration
public TensorShape Unsqueeze(int axis)
Parameters
Type Name Description
int axis

The axis at which to unsqueeze.
Returns
Type Description
TensorShape

The unsqueezed tensor shape.

Unsqueeze(ReadOnlySpan<int>)

Creates a TensorShape by duplicating this and inserting a dimension of size one at a given axis. For example if this is (2), and the value of axis is 0, the method returns (1, 2).

Declaration
public TensorShape Unsqueeze(ReadOnlySpan<int> axes)
Parameters
Type Name Description
ReadOnlySpan<int> axes

The axes at which to unsqueeze.
Returns
Type Description
TensorShape

The unsqueezed tensor shape.

Operators

operator ==(TensorShape, TensorShape)

Compares two TensorShape objects. Returns true if the two objects have the same rank, and all their dimensions are equal.

Declaration
public static bool operator ==(TensorShape a, TensorShape b)
Parameters
Type Name Description
TensorShape a

The first shape to compare.
TensorShape b

The second shape to compare.
Returns
Type Description
bool

Whether the two shapes are equal.

operator !=(TensorShape, TensorShape)

Compares two TensorShape objects.

Declaration
public static bool operator !=(TensorShape a, TensorShape b)
Parameters
Type Name Description
TensorShape a

The first shape to compare.
TensorShape b

The second shape to compare.
Returns
Type Description
bool

Whether the two shapes are not equal.