FlexNN::Layers::Conv1D Class Reference

#include <Conv1D.hpp>

Public Member Functions
	Conv1D (Conv1DParams p, Activations::Activation act=Activations::Activation::ReLU, const Activations::ActivationParameters &params=Activations::ActivationParameters{})
	Construct from params and activation.
	Conv1D (int inChannels, int outChannels, int kernelSize, int stride=1, int padding=0, int dilation=1, Activations::Activation act=Activations::Activation::ReLU, const Activations::ActivationParameters &params=Activations::ActivationParameters{})
	Convenience ctor with explicit ints.
LayerType	type () const noexcept
Activations::Activation	activation () const noexcept
const Activations::ActivationParameters &	activationParams () const noexcept
const Conv1DParams &	params () const noexcept
const Eigen::MatrixXd &	weights () const noexcept
const Eigen::VectorXd &	biases () const noexcept
void	setWeights (const Eigen::MatrixXd &W)
void	setBiases (const Eigen::VectorXd &b)
void	setActivationParameters (const Activations::ActivationParameters &p)
std::pair< Eigen::MatrixXd, Eigen::MatrixXd >	forward (const Eigen::MatrixXd &input) const
	Forward: Z[oc][ol] = b[oc] + Σic Σk W[oc][ic*K+k] * Xpad[ic][ol*stride - pad + k*dilation]
Eigen::MatrixXd	backward (const Eigen::MatrixXd &nextW, const Eigen::MatrixXd &nextdZ, const Eigen::MatrixXd &currZ) const
	Backward for hidden layers (old signature, for dense_test compatibility).
Eigen::MatrixXd	backward (const Eigen::MatrixXd &upstream, const Eigen::MatrixXd &currZ) const
	Backward with already-propagated upstream gradient.
Eigen::MatrixXd	propagate (const Eigen::MatrixXd &dZ) const
	Propagate gradient to previous layer: dX = W^T * dZ via col2im.
std::pair< Eigen::MatrixXd, Eigen::VectorXd >	grad (const Eigen::MatrixXd &dZ, const Eigen::MatrixXd &input) const
	Compute gradients dW and db for SGD update.
void	update (const Eigen::MatrixXd &dW, const Eigen::VectorXd &db, double lr) noexcept
void	updateWeights (const Eigen::MatrixXd &dW, const Eigen::VectorXd &db, double lr) noexcept

Constructor & Destructor Documentation

Conv1D() [1/2]

FlexNN::Layers::Conv1D::Conv1D ( Conv1DParams  p, Activations::Activation  act = Activations::Activation::ReLU, const Activations::ActivationParameters &  params = Activations::ActivationParameters{}  )

explicit

Construct from params and activation.

Initializes W and b with Random * 0.5 (same as Dense for v0.1).

Parameters

p	Conv1DParams with inChannels/outChannels/kernelSize/stride/pad/dilation
act	Activation (default ReLU)

Conv1D() [2/2]

FlexNN::Layers::Conv1D::Conv1D	(	int	inChannels,
		int	outChannels,
		int	kernelSize,
		int	stride = 1,
		int	padding = 0,
		int	dilation = 1,
		Activations::Activation	act = Activations::Activation::ReLU,
		const Activations::ActivationParameters &	params = Activations::ActivationParameters{}
	)

Convenience ctor with explicit ints.

Member Function Documentation

activation()

Activations::Activation FlexNN::Layers::Conv1D::activation ( ) const

inlinenoexcept

activationParams()

const Activations::ActivationParameters & FlexNN::Layers::Conv1D::activationParams ( ) const

inlinenoexcept

backward() [1/2]

Eigen::MatrixXd FlexNN::Layers::Conv1D::backward	(	const Eigen::MatrixXd &	nextW,
		const Eigen::MatrixXd &	nextdZ,
		const Eigen::MatrixXd &	currZ
	)		const

Backward for hidden layers (old signature, for dense_test compatibility).

Computes dA = W_next^T * dZ_next then ‘dZ = f’(Z) ⊙ dA. Kept forDense-style testing; new code should usebackward(upstream, currZ)`.

backward() [2/2]

Eigen::MatrixXd FlexNN::Layers::Conv1D::backward	(	const Eigen::MatrixXd &	upstream,
		const Eigen::MatrixXd &	currZ
	)		const

Backward with already-propagated upstream gradient.

Parameters

upstream	Gradient w.r.t. A of this layer (already W_next^T*dZ_next or propagate result)
currZ	Pre-activation Z of this layer from forward

Returns

Gradient w.r.t. Z of this layer

biases()

const Eigen::VectorXd & FlexNN::Layers::Conv1D::biases ( ) const

inlinenoexcept

forward()

std::pair< Eigen::MatrixXd, Eigen::MatrixXd > FlexNN::Layers::Conv1D::forward

(

const Eigen::MatrixXd &

input

)

const

Forward: Z[oc][ol] = b[oc] + Σic Σk W[oc][ic*K+k] * Xpad[ic][ol*stride - pad + k*dilation]

Parameters

input

Flattened input [C_in*L_in × batch] (each column is a sample)

Returns

Pair (Z, A) each [C_out*L_out × batch]

grad()

std::pair< Eigen::MatrixXd, Eigen::VectorXd > FlexNN::Layers::Conv1D::grad	(	const Eigen::MatrixXd &	dZ,
		const Eigen::MatrixXd &	input
	)		const

Compute gradients dW and db for SGD update.

Uses im2col: col [C_in*K × L_out] per sample, dW = Σ_n dZ_n * col_n^T / batch, db = rowMean over (batch*L_out) per channel (mean over batch and spatial).

Parameters

dZ	Gradient w.r.t. Z [C_out*L_out × batch]
input	Original input to forward [C_in*L_in × batch]

Returns

Pair (dW, db) where dW is [outChannels × inChannels*K], db is [outChannels]

params()

const Conv1DParams & FlexNN::Layers::Conv1D::params ( ) const

inlinenoexcept

propagate()

Eigen::MatrixXd FlexNN::Layers::Conv1D::propagate

(

const Eigen::MatrixXd &

dZ

)

const

Propagate gradient to previous layer: dX = W^T * dZ via col2im.

For Conv1D, this is the transpose convolution (col2im). Input dZ is [C_out*L_out × batch], output dX is [C_in*L_in × batch] where L_in is derived from dZ's implied L_out and params. The caller must know L_in — we derive it from dZ rows and params assuming the forward L_out formula. For the common case where Conv1D is followed by Dense, L_in is not needed because Dense's dZ is already flattened; this method is used when Conv1D is the next layer for a previous Dense — the Dense's upstream is propagate(dZ_next) from Conv1D.

Parameters

dZ	Gradient w.r.t. Z of this Conv1D layer [C_out*L_out × batch]

Returns

Gradient w.r.t. input of this layer [C_in*L_in × batch] (derived)

setActivationParameters()

void FlexNN::Layers::Conv1D::setActivationParameters ( const Activations::ActivationParameters &  p )

inline

setBiases()

void FlexNN::Layers::Conv1D::setBiases ( const Eigen::VectorXd &  b )

inline

setWeights()

void FlexNN::Layers::Conv1D::setWeights ( const Eigen::MatrixXd &  W )

inline

type()

LayerType FlexNN::Layers::Conv1D::type ( ) const

inlinenoexcept

update()

void FlexNN::Layers::Conv1D::update ( const Eigen::MatrixXd &  dW, const Eigen::VectorXd &  db, double  lr  )

noexcept

updateWeights()

void FlexNN::Layers::Conv1D::updateWeights ( const Eigen::MatrixXd &  dW, const Eigen::VectorXd &  db, double  lr  )

inlinenoexcept

weights()

const Eigen::MatrixXd & FlexNN::Layers::Conv1D::weights ( ) const

inlinenoexcept

---

The documentation for this class was generated from the following files:

• include/layers/Conv1D.hpp
• lib/layers/Conv1D.cpp