Tensor

public class Tensor: Hashable, Equatable, TensorSymbol

A Tensor object is a n-dimension page-aligned data container with fixed-size memory space. The attribute shape specifying the dimension information of a Tensor object.

All elements stored as `Float` values

No matter what type of array user passing in: Double or Float or Int, inside a Tensor object values will be converted and stored as Float (32-bit signle precision). Serrano chose this because 32-bit Float is the maximum precise floating data type Metal could support (v1.2).

Inside Memory Layout

Inside a Tensor object, it maintains a virtual 1-d array as a contiguous memory space manually allocated. Elements are stored following row-major order. Details can be found in TensorShape docs.

Tensor-Tensor arithmetic operators

Besides Operators which contains many math and NN operations, Tensor object itself implements/overloads common arithmetic operators. Tensor object support element-wise arithmetic operation with or without broadcasting. Also it supports in-place operation choice.

Element-wise operation without broadcasting:

+ Addition without broadcasting
- Substraction without broadcasting
* Multiplication without broadcasting
/ Division without broadcasting

Element-wise in-place operation:

&+ Addition without broadcasting
&- Substraction without broadcasting
&* Multiplication without broadcasting
&/ Division without broadcasting

Element-wise operation with broadcasting:

.+ Addition without broadcasting
.- Substraction without broadcasting
.* Multiplication without broadcasting
./ Division without broadcasting

Element-wise in-place operation with broadcasting:

.&+ Addition without broadcasting
.&- Substraction without broadcasting
.&* Multiplication without broadcasting
.&/ Division without broadcasting

Example usage:

/// Element wise addition without broadcasting
let tensorA = Tensor(repeatingValue: 1.0,
                     tensorShape: TensorShape(dataType:.float, shape: [2, 5]))
let tensorB = Tensor(repeatingValue: 2.0,
                     tensorShape: TensorShape(dataType:.float, shape: [2, 5]))
let result1 = tensorA + tensorB

/// Element wise in-place addition without broadcasting
let result2 = tensorA &+ tensorB
print(result2 == tensorA) // true

/// Element wise addition with broadcasting
let tensorD = Tensor(repeatingValue: 2.0,
                     tensorShape: TensorShape(dataType:.float, shape: [1, 5]))
let result3 = tensorA .+ tensorD

/// Element wise in-place addition with broadcasting
let resunt4 = tensorA .&+ tensorD
print(result4 == tensorA) // true

Directly used as TensorSymbol

Tensor conforms to TensorSymbol protocol. So a tensor object could be used as a symbol participating in graph computation.

Properties

_shape
shape
Declaration
Swift

internal var _shape: TensorShape
Show on GitHub
_allocatedSize
Allocated aligned memory size in bytes
Declaration
Swift

internal var _allocatedSize: Int
Show on GitHub
_dataMemoryBaseAdrress
The base address of allocated memory.
Declaration
Swift

internal var _dataMemoryBaseAdrress: UnsafeMutablePointer<Float>
Show on GitHub
_elementReader
Convenience Float reader
Declaration
Swift

internal var _elementReader: UnsafeMutableBufferPointer<Float>
Show on GitHub
_capacity
How many elements can store
Declaration
Swift

internal var _capacity: Int
Show on GitHub
_sliceMarker
Slice marker. Indicate if this tensor object is a slice from another tensor. If true, this object is just a reference slice and should not own the memeory.
Declaration
Swift

internal var _sliceMarker: Bool = false
Show on GitHub
_sliceRootTensor
The root tensor of a sliced tensor
Declaration
Swift

internal var _sliceRootTensor: Tensor? = nil
Show on GitHub
_sliceIndex
The index array of this slice object in raw tensor. nil if not a sliced tensor
Declaration
Swift

internal var _sliceIndex: [Int]?
Show on GitHub
_mtlbuffer
MTLBuffer binded to this tensor.

Note
If the tensor is sliced, this is nil.
Declaration
Swift

internal var _mtlbuffer: MTLBuffer?
Show on GitHub
count
Count of data elements tored
Declaration
Swift

public var count: Int
Show on GitHub
dimension
Dimension of the array
Declaration
Swift

public var dimension: Int
Show on GitHub
capacity
How many elements can store
Declaration
Swift

public var capacity: Int
Show on GitHub
shape
Shape
Declaration
Swift

public var shape: TensorShape
Show on GitHub
allocatedBytes
Allocated memeory bytes
Declaration
Swift

public var allocatedBytes: Int
Show on GitHub
description
Readable description of a object
Declaration
Swift

public var description: String
Show on GitHub
label
A readable label for distinguishing different tensors. Serrano does not check label unique.
Declaration
Swift

public var label: String = "Tensor"
Show on GitHub
contentsAddress
Base address of allocated memeory space

Warning
User could reuse a Tensor object with this pointer by assigning new values. However, it must be very careful with the boundary.
Declaration
Swift

public var contentsAddress: UnsafeMutablePointer<Float>
Show on GitHub
floatValueReader
Float value reader pointer
Declaration
Swift

public var floatValueReader: UnsafeMutableBufferPointer<Float>
Show on GitHub
rank
Rank
Declaration
Swift

public var rank: Int
Show on GitHub
isSliceTensor
Slice marker. Indicate if this tensor object is a slice from another tensor. If true, this object is just a reference slice and should not own the memeory.
Declaration
Swift

public var isSliceTensor: Bool
Show on GitHub
sliceRootTensor
The root tensor of a sliced tensor

Note
nil if _sliceMarker is false
Declaration
Swift

public var sliceRootTensor: Tensor?
Show on GitHub
sliceIndex
The index of this slice object in raw tensor. nil if not a sliced tensor
Declaration
Swift

public var sliceIndex: [Int]?
Show on GitHub
symbolType
Conforms to protocol TensorSymbol.
Declaration
Swift

public var symbolType: SymbolType = SymbolType.Tensor
Show on GitHub
UID
Conforms to protocol TensorSymbol.
Declaration
Swift

public var UID: String = ""
Show on GitHub
symbolLabel
Conforms to protocol TensorSymbol.
Declaration
Swift

public var symbolLabel: String = ""
Show on GitHub
inBounds
Inbound symbols list. Conforms to GraphSymbol. To prevent from cycle reference, we dynamic constructing this attribute from inBoundsWeak.
Declaration
Swift

public var inBounds: [GraphSymbol]
Show on GitHub
outBounds
Outbound symbols list. Conforms to GraphSymbol. To prevent from cycle reference, we dynamic constructing this attribute from inBoundsWeak.
Declaration
Swift

public var outBounds: [GraphSymbol]
Show on GitHub


                    
                    
                    inBoundsWeak

Weak reference array of inbounds objects

Declaration

Swift

internal var inBoundsWeak: [WeakSerranoGraphSymbol] = [WeakSerranoGraphSymbol]()

Show on GitHub


                    
                    
                    outBoundsWeak

Weak reference array of outbounds objects

Declaration

Swift

internal var outBoundsWeak: [WeakSerranoGraphSymbol] = [WeakSerranoGraphSymbol]()

Show on GitHub

bindedData
Read-only. Conforms to protocol TensorSymbol.

Note
If a tensor object used as a TensorSymbol, atrribute bindedData could not be modifed.
Declaration
Swift

public var bindedData: DataSymbolSupportedDataType?
Show on GitHub
dataSource
Conforms to protocol TensorSymbol.
Declaration
Swift

public var dataSource: SymbolDataSource = SymbolDataSource.Other
Show on GitHub
updatable
If differentiable
Declaration
Swift

public var updatable = false
Show on GitHub


                    
                    
                    currentGrad

Current grad

Declaration

Swift

public var currentGrad: DataSymbolSupportedDataType?

Show on GitHub

historyGradsEnabled
If enabled history grads recording. Default is false.
Declaration
Swift

public var historyGradsEnabled = false
Show on GitHub


                    
                    
                    historyGrads

history grads

Declaration

Swift

public var historyGrads: [DataSymbolSupportedDataType] = [Tensor]()

Show on GitHub

subscript


                    
                    
                    indexIsValid(_:)

Check if index is valid for fetching a single element

Declaration

Swift

public func indexIsValid(_ index: [Int]) -> Bool

Parameters


                                index

index description

Return Value

return value description

Show on GitHub


                    
                    
                    offSetFromIndex(_:)

Get offset in terms of element counting from valid index

Declaration

Swift

internal func offSetFromIndex(_ index: [Int]) -> Int

Parameters


                                index

valid index list

Return Value

offset

Show on GitHub


                    
                    
                    subscript(_:)

Custom subscript to fetch single Float element

Declaration

Swift

public subscript(_ index: Int...) -> Float

Parameters


                                index

Indices list

Show on GitHub


                    
                    
                    subscript(_:)

Custom subscript to fetch single Float element

Declaration

Swift

public subscript(_ index: [Int]) -> Float

Parameters


                                index

Indices list

Show on GitHub


                    
                    
                    subscript(withoutChecking:)

Internal use, already checked boundary before fetch or set

Warning

May cause unexpected result or fatal error if index is not valid.

Declaration

Swift

public subscript(withoutChecking index:[Int]) -> Float

Parameters


                                index

index array

Show on GitHub


                    
                    
                    fetchValueOrDefault(_:missingValue:)

Get element value from index. If input index is invalid, return missingValue.

Declaration

Swift

public func fetchValueOrDefault(_ index: [Int], missingValue: Float = 0.0) -> Float

Parameters

`index`	index array
`missingValue`	default value for missing elements. Default is `0.0`

Return Value

value

Show on GitHub

Initializers


                    
                    
                    init(shape:allocateSize:capacity:dataMemoryBaseAddres:elementReader:)

Designated init.

Declaration

Swift

internal init(shape: TensorShape, allocateSize: Int, capacity: Int,
            dataMemoryBaseAddres: UnsafeMutablePointer<Float>, elementReader:UnsafeMutableBufferPointer<Float>)

Parameters

`shape`	shape
`allocateSize`	allocateSize
`capacity`	capacity
`dataMemoryBaseAddres`	dataMemoryBaseAddres
`elementReader`	elementReader

Show on GitHub


                    
                    
                    init(repeatingValue:tensorShape:)

Initial a tensor with repeating value

Declaration

Swift

public convenience init(repeatingValue value: Float, tensorShape shape: TensorShape)

Parameters

`repeatingValue`	repeat value
`shape`	shape

Show on GitHub


                    
                    
                    init(fromFlatArray:tensorShape:)

Construct a tensor from a flatten (1-D) swift array.

Declaration

Swift

public convenience init(fromFlatArray array:[SupportedScalarDataType], tensorShape shape: TensorShape)

Parameters

`array`	1-D array.
`shape`	The shape which the constructed tensor have. The count of this shape should be the same of count of passed in array.

Show on GitHub


                    
                    
                    init(dataArray:tensorShape:)

Construct a tensor from a nested (N-D) swift array.

Declaration

Swift

public convenience init(dataArray array:[Any], tensorShape shape: TensorShape)

Parameters

`array`	The nested array object. Note that this array should only contains objects of type `SupportedScalarDataType`. Also this dimension of this array should be matching with parameter `tensorShape`.
`shape`	The shape describing the passed in array.

Show on GitHub


                    
                    
                    init(sliceTensorFrom:sliceContentAddress:count:shape:index:)

Used internal only to generate slice tensor

Declaration

Swift

internal init(sliceTensorFrom rootTensor: Tensor,
              sliceContentAddress: UnsafeMutablePointer<Float>,
              count: Int, shape: TensorShape, index: [Int])

Parameters

`sliceTensorFrom`	contentAddress sliceContentAddress: slice content address
`count`	count
`shape`	shape

Show on GitHub

Deinitializers

deinit
Need to free allcoated memory space manually.
Declaration
Swift

deinit
Show on GitHub

Make tensors randomly


                    
                    
                    randomTensor(_:min:max:)

Generate a random tensor.

Declaration

Swift

public static func randomTensor(_ shape: TensorShape, min: Float = 0.0, max: Float = 1.0) -> Tensor

Parameters

`shape`
`min`
`max`

Return Value

Show on GitHub

Util methods


                    
                    
                    reloadData(fromFlatArray:tensorShape:)

Reload data from a flat array.

Note

If array size < tensor’s count

Declaration

Swift

public func reloadData(fromFlatArray array:[SupportedScalarDataType], tensorShape shape: TensorShape)

Parameters


                                array

Show on GitHub


                    
                    
                    resetValues(_:)

Set tensor’s all selements to a same value.

Warning

This operation will erase current values of tensor object.

Declaration

Swift

public func resetValues(_ value:SupportedScalarDataType)

Parameters


                                value

initial value

Show on GitHub


                    
                    
                    clear()

Set tensor’s all elements to 0.

Warning

This operation will erase current values of tensor object.

Declaration

Swift

public func clear()

Parameters


                                value

initial value

Show on GitHub


                    
                    
                    copyValues(_:)

Copy values from another tensor.

Dimension check

This function just check cp from tensor’s total element count. If two tensors has same number of elements, do the copy. If not, raise fatalError

Declaration

Swift

public func copyValues(_ tensor: Tensor)

Parameters


                                tensor

cp from tensor

Show on GitHub

Array representation

flatArrayFloat()
Return flat array containing all elements stored in tensor with Float type.
Declaration
Swift

public func flatArrayFloat() -> [Float]
Return Value

Type of declared.

Show on GitHub
nestedArrayFloat()
Return nested array following object’s shape.
Declaration
Swift

public func nestedArrayFloat() -> [Any]
Return Value

A nested array with same shape in shape attribute.

Show on GitHub


                    
                    
                    constructNestedArrayFrom(location:)

Recursively construct nested array

Declaration

Swift

internal func constructNestedArrayFrom(location: [Int]) -> [Any]

Parameters


                                shape

shape needs to construct

Return Value

result

Show on GitHub

Hashable protocol

hashValue
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
Declaration
Swift

public var hashValue: Int
Show on GitHub

Equatable protocol

==(_:_:)
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
Declaration
Swift

public static func ==(lhs: Tensor, rhs: Tensor) -> Bool
Show on GitHub

Metal Buffer Binding APIs

gpuBufferResource()
Get a MTLBuffer associated with this tensor object.
Declaration
Swift

public func gpuBufferResource() -> MTLBufferResource
Return Value

Show on GitHub

Tensor manipulation

abs()
Return a new tensor contains the absolute values of this tensor. Same effect as using AbsOperator.
Declaration
Swift

public func abs() -> Tensor
Return Value

new tensor object with absed values.

Show on GitHub

Tensor-Tensor arithmetic operations (not support broadcasting)


                    
                    
                    +(_:_:)

Element-wise addition. Not support braodcasting. Same effect as using AddOperator.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func +(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    -(_:_:)

Element-wise substraction. Not support braodcasting. Same effect as using SubOperator.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func -(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    *(_:_:)

Element-wise multiplication. Not support braodcasting. Same effect as using AddOperator.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func *(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    /(_:_:)

Element-wise division. Not support braodcasting. Same effect as using DivOperator.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func /(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub

Tensor-Tensor in-place arithmetic operations (not support broadcasting)


                    
                    
                    &+(_:_:)

Element-wise in-place addition. Not support braodcasting. Same effect as using AddOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func &+(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calculated.

Show on GitHub


                    
                    
                    &-(_:_:)

Element-wise in-place substraction. Not support braodcasting. Same effect as using SubOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func &-(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calculated.

Show on GitHub


                    
                    
                    &*(_:_:)

Element-wise in-place multiplication. Not support braodcasting. Same effect as using MultOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func &*(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calculated.

Show on GitHub


                    
                    
                    &/(_:_:)

Element-wise in-place division. Not support braodcasting. Same effect as using DivOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func &/(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calculated.

Show on GitHub

Tensor-Tensor arithmetic operations (support broadcasting)


                    
                    
                    .+(_:_:)

Element-wise addition. support braodcasting. Same effect as using BrodcastAddOperator.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func .+(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    .-(_:_:)

Element-wise substraction. support braodcasting. Same effect as using BroadcastSubOperator.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func .-(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    .*(_:_:)

Element-wise multiplication. support braodcasting. Same effect as using BroadcastMultOperator.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func .*(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor

Show on GitHub


                    
                    
                    ./(_:_:)

Element-wise division. support braodcasting. Same effect as using BroadcastDivOperator.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

left: left tensor
right: right tensor

Declaration

Swift

public static func ./(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

result tensor, a new created tensor.

Show on GitHub

Tensor-Tensor in-place arithmetic operations (support broadcasting)


                    
                    
                    .&+(_:_:)

Element-wise in-place addition. support braodcasting. Same effect as using BrodcastAddOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func .&+(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calcualted

Show on GitHub


                    
                    
                    .&-(_:_:)

Element-wise in-place substraction. support braodcasting. Same effect as using BroadcastSubOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func .&-(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calcualted

Show on GitHub


                    
                    
                    .&*(_:_:)

Element-wise in-place multiplication. support braodcasting. Same effect as using BroadcastMultOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func .&*(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calcualted

Show on GitHub


                    
                    
                    .&/(_:_:)

Element-wise in-place division. support braodcasting. Same effect as using BroadcastDivOperator.

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Warning

If two tensors don’t have same shape or one of them cannot be broadcasted to anoter one, program would be aborted (fatalError() called).

Declaration

Swift

@discardableResult public static func .&/(left: Tensor, right:Tensor) -> Tensor

Parameters

`left`	left tensor
`right`	right tensor

Return Value

left tensor after calcualted

Show on GitHub

Tensor-Scalar arithmetic operations.


                    
                    
                    +(_:_:)

A tensor plus a scalar variable. Ex. [2, 3] + 0.5 --> [2.5, 3.5].

Declaration

Swift

public static func + (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    -(_:_:)

A tensor substract a scalar variable. Ex. [2, 3] - 0.5 --> [1.5, 2.5].

Declaration

Swift

public static func - (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    *(_:_:)

A tensor multiply a scalar variable. Ex. [2, 3] * 0.5 --> [1.0, 1.5].

Declaration

Swift

public static func * (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    /(_:_:)

A tensor divide a scalar variable. Ex. [2, 3] / 0.5 --> [4.0, 6.0].

Declaration

Swift

public static func / (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Result tensor, new created.

Show on GitHub

Tensor-Scalar in-place arithmetic operations.


                    
                    
                    &+(_:_:)

A tensor plus a scalar variable. Ex. [2, 3] + 0.5 --> [2.5, 3.5].

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &+ (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Show on GitHub


                    
                    
                    &-(_:_:)

A tensor substract a scalar variable. Ex. [2, 3] - 0.5 --> [1.5, 2.5].

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &- (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Show on GitHub


                    
                    
                    &*(_:_:)

A tensor multiply a scalar variable. Ex. [2, 3] * 0.5 --> [1.0, 1.5].

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &* (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

left tensor after calculated.

Show on GitHub


                    
                    
                    &/(_:_:)

A tensor divide a scalar variable. Ex. [2, 3] / 0.5 --> [4.0, 6.0].

Note

This is an in place operation. Result stored in left tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &/ (left: Tensor,  rightScalar: SupportedScalarDataType) -> Tensor

Parameters

`left`	A tensor object
`rightScalar`	A scalar variable

Return Value

Show on GitHub

Scalar-Tensor arithmetic operations.


                    
                    
                    +(_:_:)

A scalar plus a tensor variable. Ex. 0.5 + [2, 3] --> [2.5, 3.5].

Declaration

Swift

public static func + (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    -(_:_:)

A scalar substracts a tensor variable. Ex. 0.5 - [2, 3] --> [-1.5, -2.5].

Declaration

Swift

public static func - (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    *(_:_:)

A scalar multiplies a tensor variable. Ex. 0.5 * [2, 3] --> [1.0, 1.5].

Declaration

Swift

public static func * (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    /(_:_:)

A scalar divides by a tensor variable. Ex. 0.5 / [2, 3] --> [0.25, 0.16].

Declaration

Swift

public static func / (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub

Scalar-Tensor in-place arithmetic operations.


                    
                    
                    &+(_:_:)

A scalar plus a tensor variable. Ex. 0.5 + [2, 3] --> [2.5, 3.5].

Note

This is an in place operation. Result stored in right tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &+ (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result right tensor.

Show on GitHub


                    
                    
                    &-(_:_:)

A scalar substracts a tensor variable. Ex. 0.5 - [2, 3] --> [-1.5, -2.5].

Note

This is an in place operation. Result stored in right tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &- (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result right tensor.

Show on GitHub


                    
                    
                    &*(_:_:)

A scalar multiplies a tensor variable. Ex. 0.5 * [2, 3] --> [1.0, 1.5].

Note

This is an in place operation. Result stored in right tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &* (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub


                    
                    
                    &/(_:_:)

A scalar divides by a tensor variable. Ex. 0.5 / [2, 3] --> [0.25, 0.16].

Note

This is an in place operation. Result stored in right tensor and return the calcualted left tensor.

Declaration

Swift

@discardableResult public static func &/ (leftScalar: SupportedScalarDataType, right: Tensor) -> Tensor

Parameters

`leftScalar`	A scalar variable
`right`	A tensor object

Return Value

Result tensor, new created.

Show on GitHub

Slice tensor management


                    
                    
                    slice(sliceIndex:)

Make a slice tensor from this object.

Declaration

Swift

public func slice(sliceIndex:[Int]) -> Tensor

Parameters


                                sliceIndex

Slice Inde

Return Value

A sliced tensor

Show on GitHub


                    
                    
                    containsSlice(_:)

Check if a slice tensor object belong to this tensor.

Note

If the passed in tensor is not a slice tensor, always returns false.

Declaration

Swift

public func containsSlice(_ slice: Tensor) -> Bool

Parameters


                                slice

slice tensor

Return Value

Bool. Result

Show on GitHub


                    
                    
                    slicedTensorOffset(_:)

Get the bytes offset for a slice tensor spawned from this tensor.

Note

The function will first check if contains this slice, if not it will return nil

Declaration

Swift

public func slicedTensorOffset(_ slice: Tensor) -> Int?

Parameters


                                slice

slice object

Return Value

offset in bytes

Show on GitHub

Protocol TensorSymbol

evaluate()
Return self
Declaration
Swift

public func evaluate() -> [DataSymbolSupportedDataType]?
Return Value

tensor

Show on GitHub
addToInBound(_:)
Add new symbol to inBounds. Should check duplicate.
Declaration
Swift

public func addToInBound(_ symbol: GraphSymbol)
Show on GitHub
addToOutBound(_:)
Add new symbol to outBounds Should check duplicate.
Declaration
Swift

public func addToOutBound(_ symbol: GraphSymbol)
Show on GitHub


                    
                    
                    bindData(_:)

A tensor object could not be binded to another tensor.

Warning

This function will do nonthing and always return false. A error logging will be gaven.

Declaration

Swift

public func bindData(_ data:DataSymbolSupportedDataType) -> Bool

Parameters


                                data

data. Should be a Tensor object.

Return Value

always False

Show on GitHub

MPS utils


                    
                    
                    getMPSImage(dataFormat:)

Undocumented

Declaration

Swift

public func getMPSImage(dataFormat: TensorChannelOrder) -> MPSImage

Show on GitHub

scarlarValue
Undocumented
Declaration
Swift

public var scarlarValue: Float
Show on GitHub
tensorValue
Undocumented
Declaration
Swift

public var tensorValue: Tensor
Show on GitHub

Tensor

All elements stored as Float values

Inside Memory Layout

Tensor-Tensor arithmetic operators

Element-wise operation without broadcasting:

Element-wise in-place operation:

Element-wise operation with broadcasting:

Element-wise in-place operation with broadcasting:

Directly used as TensorSymbol

Properties

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

Declaration

subscript

Declaration

Parameters

Return Value

Declaration

Parameters

Return Value

Declaration

Parameters

Declaration

Parameters

Declaration

Parameters

Declaration

Parameters

Return Value

Initializers

Declaration

Parameters

Declaration

Parameters

Declaration

Parameters

Declaration

Parameters

Declaration

Parameters

Deinitializers

Declaration

Make tensors randomly

Declaration

Parameters

Return Value

Util methods

Declaration

All elements stored as `Float` values