opennn::DropoutOp Struct Reference

Inverted dropout: at training time zeros activations with probability rate and rescales survivors. More...

#include <operators.h>

Inheritance diagram for opennn::DropoutOp:

Public Member Functions
bool	active () const
	Returns true when the dropout rate is non-zero.
void	set_rate (float new_rate)
	Sets the drop probability (0 disables dropout).
void	forward_propagate (ForwardPropagation &fp, size_t layer, bool is_training) noexcept override
	Runs the operator's forward computation.
void	back_propagate (ForwardPropagation &fp, BackPropagation &bp, size_t layer) const noexcept override
	Runs the operator's backward computation, accumulating into gradient/delta buffers.
void	apply_cpu (TensorView &output)
	CPU forward implementation; samples the mask and rescales survivors in place.
void	apply_gpu (TensorView &output)
	GPU forward implementation; samples the mask and rescales survivors in place.
void	apply_delta (TensorView &delta) const
	Applies the cached mask to a gradient tensor during the backward pass.
void	to_JSON (JsonWriter &w) const override
	Serializes the operator configuration to a JSON writer.
void	from_JSON (const Json *parent) override
	Restores the operator configuration from a JSON node.
void	destroy_cuda () override
	Releases CUDA resources owned by the operator; called from destructors.
	~DropoutOp () override
	DropoutOp ()=default
	DropoutOp (DropoutOp &&) noexcept=default
DropoutOp &	operator= (DropoutOp &&) noexcept=default
Public Member Functions inherited from opennn::Operator
virtual	~Operator ()=default
virtual vector< TensorSpec >	parameter_specs () const
	Returns the tensor specs of trainable parameters owned by this operator.
virtual vector< TensorSpec >	state_specs () const
	Returns the tensor specs of persistent state owned by this operator.
virtual void	link_parameters (span< const TensorView >)
	Binds parameter views provided by the hosting layer.
virtual void	link_gradients (span< const TensorView >)
	Binds gradient views provided by the hosting layer.
virtual void	link_states (span< const TensorView >)
	Binds state views provided by the hosting layer.
virtual void	set_parameters_random ()
	Initializes parameters with random values.
virtual void	set_parameters_glorot ()
	Initializes parameters using Glorot (Xavier) initialization.
virtual void	load_state_from_JSON (const Json *)
	Restores persistent state (e.g. running statistics) from a JSON node.
TensorView &	get_input (ForwardPropagation &fp, size_t layer, size_t i=0) const noexcept
vector< TensorView > &	get_inputs (ForwardPropagation &fp, size_t layer, size_t i=0) const noexcept
TensorView &	get_output (ForwardPropagation &fp, size_t layer, size_t i=0) const noexcept
TensorView &	get_output_delta (BackPropagation &bp, size_t layer, size_t i=0) const noexcept
TensorView &	get_input_delta (BackPropagation &bp, size_t layer, size_t i=0) const noexcept

Public Attributes
float	rate = 0.0f
Buffer	mask
vector< size_t >	save_slots
Public Attributes inherited from opennn::Operator
vector< size_t >	input_slots = {0}
vector< size_t >	output_slots = {1}
vector< size_t >	input_delta_slots = {1}
vector< size_t >	output_delta_slots = {0}

Detailed Description

Inverted dropout: at training time zeros activations with probability rate and rescales survivors.

Constructor & Destructor Documentation

~DropoutOp()

opennn::DropoutOp::~DropoutOp ( )

inlineoverride

DropoutOp() [1/2]

opennn::DropoutOp::DropoutOp ( )

default

DropoutOp() [2/2]

opennn::DropoutOp::DropoutOp ( DropoutOp && )

defaultnoexcept

Member Function Documentation

active()

bool opennn::DropoutOp::active ( ) const

inline

Returns true when the dropout rate is non-zero.

apply_cpu()

void opennn::DropoutOp::apply_cpu

(

TensorView &

output

)

CPU forward implementation; samples the mask and rescales survivors in place.

apply_delta()

void opennn::DropoutOp::apply_delta

(

TensorView &

delta

)

const

Applies the cached mask to a gradient tensor during the backward pass.

apply_gpu()

void opennn::DropoutOp::apply_gpu

(

TensorView &

output

)

GPU forward implementation; samples the mask and rescales survivors in place.

back_propagate()

void opennn::DropoutOp::back_propagate ( ForwardPropagation & fp, BackPropagation & bp, size_t layer ) const

overridevirtualnoexcept

Runs the operator's backward computation, accumulating into gradient/delta buffers.

Parameters

fp	Forward propagation workspace (read-only).
bp	Back propagation workspace receiving gradients and deltas.
layer	Index of the hosting layer in the workspace.

Reimplemented from opennn::Operator.

destroy_cuda()

void opennn::DropoutOp::destroy_cuda ( )

overridevirtual

Releases CUDA resources owned by the operator; called from destructors.

Reimplemented from opennn::Operator.

forward_propagate()

void opennn::DropoutOp::forward_propagate ( ForwardPropagation & fp, size_t layer, bool is_training )

overridevirtualnoexcept

Runs the operator's forward computation.

Parameters

fp	Forward propagation workspace.
layer	Index of the hosting layer in the workspace.
is_training	If true, enables training-only behavior (e.g. dropout sampling).

Reimplemented from opennn::Operator.

from_JSON()

void opennn::DropoutOp::from_JSON ( const Json * )

overridevirtual

Restores the operator configuration from a JSON node.

Reimplemented from opennn::Operator.

operator=()

DropoutOp & opennn::DropoutOp::operator= ( DropoutOp && )

defaultnoexcept

set_rate()

void opennn::DropoutOp::set_rate

(

float

new_rate

)

Sets the drop probability (0 disables dropout).

to_JSON()

void opennn::DropoutOp::to_JSON ( JsonWriter & ) const

overridevirtual

Serializes the operator configuration to a JSON writer.

Reimplemented from opennn::Operator.

Member Data Documentation

mask

Buffer opennn::DropoutOp::mask

rate

float opennn::DropoutOp::rate = 0.0f

save_slots

vector<size_t> opennn::DropoutOp::save_slots