components

Documentation

Overview

Package components provides reusable components for neural network layers. (Stability: stable)

Package components provides reusable components for neural network layers.

Index

• type HeInitializer
  • func NewHeInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...HeInitializerOption[T]) *HeInitializer[T]
  • func (h *HeInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)
• type HeInitializerOption
• type HeInitializerOptions
• type LinearGradientComputer
  • func NewLinearGradientComputer[T tensor.Numeric](engine compute.Engine[T], opts ...LinearGradientComputerOption[T]) *LinearGradientComputer[T]
  • func (g *LinearGradientComputer[T]) ComputeBothGradients(ctx context.Context, input, weights, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], *tensor.TensorNumeric[T], error)
  • func (g *LinearGradientComputer[T]) ComputeInputGradient(ctx context.Context, weights, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
  • func (g *LinearGradientComputer[T]) ComputeWeightGradient(ctx context.Context, input, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
• type LinearGradientComputerOption
• type LinearGradientComputerOptions
• type MatrixMultiplier
  • func NewMatrixMultiplier[T tensor.Numeric](engine compute.Engine[T], opts ...MatrixMultiplierOption[T]) *MatrixMultiplier[T]
  • func (m *MatrixMultiplier[T]) Multiply(ctx context.Context, a, b *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
  • func (m *MatrixMultiplier[T]) MultiplyWithDestination(ctx context.Context, a, b, dst *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
  • func (m *MatrixMultiplier[T]) Transpose(ctx context.Context, a *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
  • func (m *MatrixMultiplier[T]) TransposeWithDestination(ctx context.Context, a, dst *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)
• type MatrixMultiplierOption
• type MatrixMultiplierOptions
• type UniformInitializer
  • func NewUniformInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...UniformInitializerOption[T]) *UniformInitializer[T]
  • func (u *UniformInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)
• type UniformInitializerOption
  • func WithScale[T tensor.Numeric](scale float64) UniformInitializerOption[T]
• type UniformInitializerOptions
• type WeightInitializer
• type XavierInitializer
  • func NewXavierInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...XavierInitializerOption[T]) *XavierInitializer[T]
  • func (x *XavierInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)
• type XavierInitializerOption
• type XavierInitializerOptions

Types

type HeInitializer

type HeInitializer[T tensor.Numeric] struct {
	// contains filtered or unexported fields
}

HeInitializer implements He initialization. Weights are sampled from a normal distribution with variance 2/fan_in.

func NewHeInitializer

func NewHeInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...HeInitializerOption[T]) *HeInitializer[T]

NewHeInitializer creates a new He initializer.

func (*HeInitializer[T]) Initialize

func (h *HeInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)

Initialize generates weights using He initialization.

type HeInitializerOption

type HeInitializerOption[T tensor.Numeric] func(*HeInitializerOptions[T])

HeInitializerOption is a function that applies an option to HeInitializerOptions.

type HeInitializerOptions

type HeInitializerOptions[T tensor.Numeric] struct {
}

HeInitializerOptions holds configuration options for HeInitializer.

type LinearGradientComputer

type LinearGradientComputer[T tensor.Numeric] struct {
	// contains filtered or unexported fields
}

LinearGradientComputer handles gradient computation for linear layers.

func NewLinearGradientComputer

func NewLinearGradientComputer[T tensor.Numeric](engine compute.Engine[T], opts ...LinearGradientComputerOption[T]) *LinearGradientComputer[T]

NewLinearGradientComputer creates a new linear gradient computer.

func (*LinearGradientComputer[T]) ComputeBothGradients

func (g *LinearGradientComputer[T]) ComputeBothGradients(ctx context.Context, input, weights, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], *tensor.TensorNumeric[T], error)

ComputeBothGradients computes both weight and input gradients in one call. This can be more efficient when both gradients are needed.

func (*LinearGradientComputer[T]) ComputeInputGradient

func (g *LinearGradientComputer[T]) ComputeInputGradient(ctx context.Context, weights, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

ComputeInputGradient computes the gradient with respect to input. Formula: input_gradient = output_gradient * weights^T.

func (*LinearGradientComputer[T]) ComputeWeightGradient

func (g *LinearGradientComputer[T]) ComputeWeightGradient(ctx context.Context, input, outputGradient *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

ComputeWeightGradient computes the gradient with respect to weights. Formula: weight_gradient = input^T * output_gradient.

type LinearGradientComputerOption

type LinearGradientComputerOption[T tensor.Numeric] func(*LinearGradientComputerOptions[T])

LinearGradientComputerOption applies an option to LinearGradientComputerOptions.

type LinearGradientComputerOptions

type LinearGradientComputerOptions[T tensor.Numeric] struct {
}

LinearGradientComputerOptions holds configuration options for LinearGradientComputer.

type MatrixMultiplier

type MatrixMultiplier[T tensor.Numeric] struct {
	// contains filtered or unexported fields
}

MatrixMultiplier handles matrix multiplication operations for layers.

func NewMatrixMultiplier

func NewMatrixMultiplier[T tensor.Numeric](engine compute.Engine[T], opts ...MatrixMultiplierOption[T]) *MatrixMultiplier[T]

NewMatrixMultiplier creates a new matrix multiplier.

func (*MatrixMultiplier[T]) Multiply

func (m *MatrixMultiplier[T]) Multiply(ctx context.Context, a, b *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

Multiply performs matrix multiplication: result = a * b.

func (*MatrixMultiplier[T]) MultiplyWithDestination

func (m *MatrixMultiplier[T]) MultiplyWithDestination(ctx context.Context, a, b, dst *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

MultiplyWithDestination performs matrix multiplication with a pre-allocated destination tensor.

func (*MatrixMultiplier[T]) Transpose

func (m *MatrixMultiplier[T]) Transpose(ctx context.Context, a *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

Transpose transposes a matrix.

func (*MatrixMultiplier[T]) TransposeWithDestination

func (m *MatrixMultiplier[T]) TransposeWithDestination(ctx context.Context, a, dst *tensor.TensorNumeric[T]) (*tensor.TensorNumeric[T], error)

TransposeWithDestination transposes a matrix with a pre-allocated destination tensor.

type MatrixMultiplierOption

type MatrixMultiplierOption[T tensor.Numeric] func(*MatrixMultiplierOptions[T])

MatrixMultiplierOption applies an option to MatrixMultiplierOptions.

type MatrixMultiplierOptions

type MatrixMultiplierOptions[T tensor.Numeric] struct {
}

MatrixMultiplierOptions represents configuration options for MatrixMultiplier.

type UniformInitializer

type UniformInitializer[T tensor.Numeric] struct {
	// contains filtered or unexported fields
}

UniformInitializer implements simple uniform initialization.

func NewUniformInitializer

func NewUniformInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...UniformInitializerOption[T]) *UniformInitializer[T]

NewUniformInitializer creates a new uniform initializer with the given scale.

func (*UniformInitializer[T]) Initialize

func (u *UniformInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)

Initialize generates weights using uniform initialization.

type UniformInitializerOption

type UniformInitializerOption[T tensor.Numeric] func(*UniformInitializerOptions[T])

UniformInitializerOption is a function that applies an option to UniformInitializerOptions.

func WithScale

func WithScale[T tensor.Numeric](scale float64) UniformInitializerOption[T]

WithScale sets the scale parameter for UniformInitializer.

type UniformInitializerOptions

type UniformInitializerOptions[T tensor.Numeric] struct {
	Scale float64
}

UniformInitializerOptions holds configuration options for UniformInitializer.

type WeightInitializer

type WeightInitializer[T tensor.Numeric] interface {
	Initialize(inputSize, outputSize int) ([]T, error)
}

WeightInitializer defines the interface for weight initialization strategies.

type XavierInitializer

type XavierInitializer[T tensor.Numeric] struct {
	// contains filtered or unexported fields
}

XavierInitializer implements Xavier/Glorot initialization. Weights are sampled from a uniform distribution with variance 2/(fan_in + fan_out).

func NewXavierInitializer

func NewXavierInitializer[T tensor.Numeric](ops numeric.Arithmetic[T], opts ...XavierInitializerOption[T]) *XavierInitializer[T]

NewXavierInitializer creates a new Xavier initializer.

func (*XavierInitializer[T]) Initialize

func (x *XavierInitializer[T]) Initialize(inputSize, outputSize int) ([]T, error)

Initialize generates weights using Xavier initialization.

type XavierInitializerOption

type XavierInitializerOption[T tensor.Numeric] func(*XavierInitializerOptions[T])

XavierInitializerOption is a function that applies an option to XavierInitializerOptions.

type XavierInitializerOptions

type XavierInitializerOptions[T tensor.Numeric] struct {
}

XavierInitializerOptions holds configuration options for XavierInitializer.