pulsatrix::SequentialModule Class Reference

Composes layers_[0..n-1] in forward() order; backward()/propagate_relevance() chain layers_[n-1..0] in reverse – correct reverse-mode composition order. More...

#include <sequential_module.hpp>

Inheritance diagram for pulsatrix::SequentialModule:

Collaboration diagram for pulsatrix::SequentialModule:

Public Member Functions
	SequentialModule (std::vector< Module * > layers)
	Constructs a container over an ordered sequence of layers.
Tensor	backward (const Tensor &grad_output) override
	Chains backward() across layers_ in reverse order.
OpType	op_type () const override
	Composite per charter's closed OpType set – a container wrapping several ops is genuinely not any single existing category.
Tensor	propagate_relevance (const Tensor &relevance_out, const LRPRuleConfig &config) override
	Chains propagate_relevance() across layers_ in reverse order.
bool	supports_lrp_rule (LRPRule rule) const override
	A rule is supported iff every contained layer supports it (the config is forwarded to all).
std::vector< NamedParamRef >	named_parameters () override
	Every contained layer's named_parameters(), prefixed with its index (0.weight).
void	set_training (bool training) override
	Sets this container's own training flag and cascades to every contained layer.
const std::vector< Module * > &	layers () const
	The contained layers, in forward-execution order.
Public Member Functions inherited from pulsatrix::Module
virtual	~Module ()=default
Tensor	forward (const Tensor &input)
	Runs this module's forward computation.
virtual std::optional< DeviceType >	compute_device () const
	The device this module computes on, so forward() can reject an input on another device before any kernel sees it (roadmap FND-8).
std::pair< Tensor, NodeId >	forward_traced (const Tensor &input, NodeId input_node, ComputationGraph &graph, Autograd &autograd)
	Runs forward() while also registering a ComputationGraph node (tagged with this module's op_type(), parented to input_node) and wiring an Autograd backward function that reuses this module's own backward() – the opt-in traced/explainable path, per Phase 2 Mission 0.
virtual std::vector< ParamRef >	parameters ()
	This module's trainable parameters and their gradients, for an optimizer to update uniformly across module types.
void	set_requires_grad (bool requires_grad, const std::string &prefix="")
	Freezes (false) or unfreezes (true) parameters by name (roadmap FND-2).
bool	is_training () const
	Whether this module is currently in training mode.

Protected Member Functions
Tensor	forward_impl (const Tensor &input) override
	Chains forward() across layers_ in order.

Detailed Description

Composes layers_[0..n-1] in forward() order; backward()/propagate_relevance() chain layers_[n-1..0] in reverse – correct reverse-mode composition order.

Note

layers_ is not owned – "not owned; must outlive this object", the same convention as every DeviceBackend* member in this codebase. A caller constructing a SequentialModule keeps its constituent modules alive independently.

propagate_relevance needs no new LRP theory – pure delegation to each contained layer's own already-cited rule. Conservation holds end-to-end because composing conserving functions conserves; this is verified numerically by an LRPConservationTest TEST_P case, not just asserted in prose.

set_training() overrides the (now virtual, since this mission) base method: sets its own flag and cascades to every contained layer – correct even when accessed through a Module* base pointer, not just SequentialModule's own concrete type.

Constructor & Destructor Documentation

SequentialModule()

pulsatrix::SequentialModule::SequentialModule ( std::vector< Module * >  layers )

explicit

Constructs a container over an ordered sequence of layers.

Parameters

layers

Layers in forward-execution order. Not owned; must outlive this object.

Exceptions

std::invalid_argument

if layers is empty, or any entry is nullptr – both external boundaries (a caller-constructed vector, e.g. from Phase 5's Python bindings, could be empty or contain a null entry with no upstream validation).

Member Function Documentation

backward()

Tensor pulsatrix::SequentialModule::backward ( const Tensor &  grad_output )

overridevirtual

Chains backward() across layers_ in reverse order.

Parameters

grad_output

Gradient w.r.t. this container's output. Must match the shape of the most recent forward() call's output (validated by the last layer's own backward(), not by SequentialModule itself).

Returns

Gradient w.r.t. this container's input.

Exceptions

std::logic_error

if forward() has never been called.

Implements pulsatrix::Module.

forward_impl()

Tensor pulsatrix::SequentialModule::forward_impl ( const Tensor &  input )

overrideprotectedvirtual

Chains forward() across layers_ in order.

Exceptions

Whatever

the first layer whose own forward() rejects input throws – SequentialModule adds no extra validation of its own beyond delegation.

Implements pulsatrix::Module.

layers()

const std::vector< Module * > & pulsatrix::SequentialModule::layers ( ) const

inline

The contained layers, in forward-execution order.

named_parameters()

std::vector< NamedParamRef > pulsatrix::SequentialModule::named_parameters ( )

overridevirtual

Every contained layer's named_parameters(), prefixed with its index (0.weight).

Reimplemented from pulsatrix::Module.

op_type()

OpType pulsatrix::SequentialModule::op_type ( ) const

inlineoverridevirtual

Composite per charter's closed OpType set – a container wrapping several ops is genuinely not any single existing category.

Implements pulsatrix::Module.

propagate_relevance()

Tensor pulsatrix::SequentialModule::propagate_relevance ( const Tensor &  relevance_out, const LRPRuleConfig &  config  )

overridevirtual

Chains propagate_relevance() across layers_ in reverse order.

Parameters

relevance_out	Relevance at this container's output. Must match the shape of the most recent forward() call's output (validated by the last layer's own propagate_relevance(), not by SequentialModule itself).
config	Forwarded unchanged to every contained layer's own rule.

Returns

Relevance at this container's input.

Exceptions

std::logic_error

if forward() has never been called.

Implements pulsatrix::Module.

set_training()

void pulsatrix::SequentialModule::set_training ( bool  training )

overridevirtual

Sets this container's own training flag and cascades to every contained layer.

Parameters

training

New training/eval mode.

Reimplemented from pulsatrix::Module.

supports_lrp_rule()

bool pulsatrix::SequentialModule::supports_lrp_rule ( LRPRule  rule ) const

inlineoverridevirtual

A rule is supported iff every contained layer supports it (the config is forwarded to all).

Reimplemented from pulsatrix::Module.

---

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

• include/pulsatrix/sequential_module.hpp