tensor

package
v3.0.3 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jul 5, 2024 License: MIT Imports: 6 Imported by: 0

Documentation

Overview

Package tensor defines the tensor interface.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

This section is empty.

Types

type CPUOperator 

type CPUOperator struct {
}

CPUOperator can process CPU tensors.

func (CPUOperator) Adam 

func (to CPUOperator) Adam(
	params, gradients Tensor,
	vHistory, sHistory Tensor,
	smoothFactor1, smoothFactor2, learningRate float64,
)

Adam runs the adam gradient descent algorithm.

func (CPUOperator) AvgPoolingBackward 

func (to CPUOperator) AvgPoolingBackward(forwardIn, backwardIn Tensor,
	kernelSize, padding, stride []int) Tensor

AvgPoolingBackward calculates the backward propagation results for AvgPooling layers.

func (CPUOperator) AvgPoolingForward 

func (to CPUOperator) AvgPoolingForward(
	t Tensor,
	kernelSize, padding, stride []int,
) Tensor

AvgPoolingForward calculates the forward propagation results for AvgPooling layers.

func (CPUOperator) Clear 

func (to CPUOperator) Clear(t Tensor)

Clear sets all the element to 0.

func (CPUOperator) Clone 

func (to CPUOperator) Clone(t Tensor) Tensor

Clone duplicates the tensor.

func (CPUOperator) Copy 

func (to CPUOperator) Copy(dst, src Tensor)

Copy moves the data, size, and descriptor from the src tensor to the destination tensor.

func (CPUOperator) Create 

func (to CPUOperator) Create(size []int) Tensor

Create creates a new CPU tensor.

func (CPUOperator) CreateWithData 

func (to CPUOperator) CreateWithData(
	data []float64,
	size []int,
	descriptor string,
) Tensor

CreateWithData creates a new CPU tensor and fill it with the given data.

func (CPUOperator) CrossEntropy 

func (to CPUOperator) CrossEntropy(t Tensor, label []int) float64

CrossEntropy calculates the cross entropy.

func (CPUOperator) CrossEntropyDerivative 

func (to CPUOperator) CrossEntropyDerivative(t Tensor, label []int) Tensor

CrossEntropyDerivative generators the final derivative of the forward pass.

func (CPUOperator) Dilate 

func (to CPUOperator) Dilate(t Tensor, dilate []int) Tensor

Dilate add 0s in the rows and columns.

func (CPUOperator) Dump 

func (to CPUOperator) Dump(t Tensor) string

Dump converts the content of the tensor to a string.

func (CPUOperator) ElementWiseMul 

func (to CPUOperator) ElementWiseMul(t1, t2 Tensor) Tensor

ElementWiseMul performs element-wise multiplication operation.

func (CPUOperator) Free 

func (to CPUOperator) Free(t Tensor)

Free does nothing as Go's gabage collector will take care of it.

func (CPUOperator) Gemm 

func (to CPUOperator) Gemm(
	transA, transB bool,
	alpha, beta float64,
	a, b, c Tensor,
) Tensor

Gemm performs alpha x A x B + beta x C operation.

func (CPUOperator) Im2Col 

func (to CPUOperator) Im2Col(
	t Tensor,
	kernelSize, padding, stride, dilation []int,
) Tensor

Im2Col performs the im2col operation.

func (CPUOperator) Init 

func (to CPUOperator) Init(t Tensor, data []float64)

Init initialize tensor with data and size.

func (CPUOperator) MaxPoolingBackward 

func (to CPUOperator) MaxPoolingBackward(
	forwardIn, backwardIn, mask Tensor,
	kernelSize, padding, stride []int,
) Tensor

MaxPoolingBackward calculates the back propagation results for MaxPooling layers.

func (CPUOperator) MaxPoolingForward 

func (to CPUOperator) MaxPoolingForward(
	t Tensor,
	kernelSize, padding, stride []int,
) (out, mask Tensor)

MaxPoolingForward calculates the forward propagation results for MaxPooling layers.

func (CPUOperator) RMSProp 

func (to CPUOperator) RMSProp(
	params, gradients Tensor,
	sHistory Tensor,
	smoothFactor, learningRate float64,
)

RMSProp runs the rmsProp gradient descent algorithm.

func (CPUOperator) ReluBackward 

func (to CPUOperator) ReluBackward(forwardIn, backwardIn Tensor) Tensor

ReluBackward runs the relu backward propagation algorithm.

func (CPUOperator) ReluForward 

func (to CPUOperator) ReluForward(in Tensor) Tensor

ReluForward runs the ReLU forward propagation algorithm.

func (CPUOperator) Repeat 

func (to CPUOperator) Repeat(t Tensor, times int) Tensor

Repeat will create a new tensor that duplicates the input tensor by n times.

func (CPUOperator) Reshape 

func (to CPUOperator) Reshape(t Tensor, newSize []int) Tensor

Reshape creates a new tensor wit the same element but a different shape.

func (CPUOperator) Rotate180 

func (to CPUOperator) Rotate180(t Tensor) Tensor

Rotate180 rotates all the lowest level matrics by 180 degree.

func (CPUOperator) ScaleAdd 

func (to CPUOperator) ScaleAdd(alpha, beta float64, a, b Tensor) Tensor

ScaleAdd performs the alpht*A + beta*B operation

func (CPUOperator) Slice 

func (to CPUOperator) Slice(
	t Tensor,
	start, end int,
) Tensor

Slice creates another tensor that shares part of the underlying buffer bounded by [start, end).

func (CPUOperator) Softmax 

func (to CPUOperator) Softmax(t Tensor) Tensor

Softmax caculates the softmax vector.

func (CPUOperator) SoftmaxCrossEntropyDerivative 

func (to CPUOperator) SoftmaxCrossEntropyDerivative(
	t Tensor,
	label []int,
) Tensor

SoftmaxCrossEntropyDerivative generators the final derivative of the forward pass.

func (CPUOperator) Sum 

func (to CPUOperator) Sum(t Tensor, axis []int) Tensor

Sum calculate sums over given axis.

func (CPUOperator) Transpose 

func (to CPUOperator) Transpose(t Tensor, order []int) Tensor

Transpose can reorder the axes.

func (CPUOperator) Zeros 

func (to CPUOperator) Zeros(size []int) Tensor

Zeros creates a tensor with given size and fill the tensor with 0s.

type Operator 

type Operator interface {
	Create(size []int) Tensor
	CreateWithData(data []float64, size []int, descriptor string) Tensor
	Free(t Tensor)
	Copy(dst, src Tensor)
	Clone(t Tensor) Tensor
	Dump(t Tensor) string

	Init(t Tensor, data []float64)
	Slice(t Tensor, start, end int) Tensor
	Repeat(t Tensor, times int) Tensor
	Clear(t Tensor)
	Zeros(size []int) Tensor

	Reshape(t Tensor, newSize []int) Tensor
	Transpose(t Tensor, order []int) Tensor
	Rotate180(t Tensor) Tensor
	Dilate(t Tensor, dilate []int) Tensor

	Sum(t Tensor, axis []int) Tensor
	Gemm(transA, transB bool, alpha, beta float64, a, b, c Tensor) Tensor
	Im2Col(t Tensor, kernelSize, padding, stride, dilation []int) Tensor
	MaxPoolingForward(t Tensor,
		kernelSize, padding, stride []int,
	) (out, mask Tensor)
	MaxPoolingBackward(
		forwardIn, backwardIn, mask Tensor,
		kernelSize, padding, stride []int) Tensor
	AvgPoolingForward(t Tensor,
		kernelSize, padding, stride []int) Tensor
	AvgPoolingBackward(forwardIn, backwardIn Tensor,
		kernelSize, padding, stride []int) Tensor
	Softmax(t Tensor) Tensor
	CrossEntropy(t Tensor, label []int) float64
	CrossEntropyDerivative(t Tensor, label []int) Tensor
	SoftmaxCrossEntropyDerivative(t Tensor, label []int) Tensor

	ElementWiseMul(t1, t2 Tensor) Tensor
	ScaleAdd(alpha, beta float64, a, b Tensor) Tensor

	RMSProp(params, gradient, sHistory Tensor,
		smoothFactor, learningRate float64)
	Adam(params, gradients, vHistory, sHistory Tensor,
		smoothFactor1, smoothFactor2, learningRate float64)

	ReluForward(in Tensor) Tensor
	ReluBackward(forwardIn, backwardIn Tensor) Tensor
}

Operator can process tensors

type SimpleTensor 

type SimpleTensor struct {
	// contains filtered or unexported fields
}

A SimpleTensor is a multi-dimensional matrix.

func (SimpleTensor) Descriptor 

func (t SimpleTensor) Descriptor() string

Descriptor returns the descriptor of the tensor.

func (SimpleTensor) Dim 

func (t SimpleTensor) Dim() int

Dim returns the number of dimensions that the tensor has.

func (SimpleTensor) NumElement 

func (t SimpleTensor) NumElement() int

NumElement returns the total number of elements in the tensor.

func (*SimpleTensor) SetDescriptor 

func (t *SimpleTensor) SetDescriptor(d string)

SetDescriptor sets the descriptor of the tensor

func (*SimpleTensor) SetSize 

func (t *SimpleTensor) SetSize(newSize []int)

SetSize sets the size of the tensor

func (SimpleTensor) Size 

func (t SimpleTensor) Size() []int

Size returns the size of the tensor.

func (SimpleTensor) Vector 

func (t SimpleTensor) Vector() []float64

Vector returns the raw data of the tensor.

type Tensor 

type Tensor interface {
	// Dim returns the number of dimension of the tensor
	Dim() int

	// NumElement returns the total number of elements in the tensor
	NumElement() int

	// Size returns the length of the tensor in each dimension, from the
	// outermost dimension to the innermost dimension.
	Size() []int

	// SetSize sets the size of the tensor.
	SetSize([]int)

	// Vector returns the data of the tensor represented in a pure vector.
	// Here, we use float64. However, the concrete tensor implementation can
	// use lower-precision numbers.
	Vector() []float64

	// Descriptor represents what does each dimension of the tensor represents.
	//
	// Tokens that represent the meaning of the dimensions include N, C, H, W.
	Descriptor() string

	// SetDescriptor sets the descriptor of the tensor.
	SetDescriptor(d string)
}

A Tensor is a multi-dimension matrix.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.