pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::ConcatDenoiser Class Referencefinal

A Module that reads [x_t, embedding(t), c] as one row, conditioned by concatenation as in tests/diffusion_integration_test.cpp: x_t, then a sinusoidal timestep embedding (SinusoidalTimestepEmbedding), then a fixed conditioning vector such as a one-hot class. Its input relevance is split back into those three parts. More...

#include <trajectory_lrp.hpp>

Inheritance diagram for pulsatrix::ConcatDenoiser:
Collaboration diagram for pulsatrix::ConcatDenoiser:

Public Member Functions

 ConcatDenoiser (Module &network, int64_t dim, int64_t time_dim, std::vector< float > condition, DeviceBackend *backend, LRPRuleConfig config=TrajectoryNetworkRules())
 
void set_condition (std::vector< float > condition)
 Replaces the conditioning, for instance with another class.
 
const std::vector< float > & condition () const
 
std::vector< float > input_row (const std::vector< float > &x_t, int64_t t) const
 The row the network reads at timestep t.
 
std::vector< float > predict (const std::vector< float > &x_t, int64_t t) override
 The predicted noise at timestep t, the same size as x_t.
 
DenoiserRelevance relevance (const std::vector< float > &relevance_eps) override
 Relevance on the last predict()'s inputs, from relevance_eps on its output.
 
- Public Member Functions inherited from pulsatrix::ContinuousDenoiser
virtual ~ContinuousDenoiser ()=default
 

Detailed Description

A Module that reads [x_t, embedding(t), c] as one row, conditioned by concatenation as in tests/diffusion_integration_test.cpp: x_t, then a sinusoidal timestep embedding (SinusoidalTimestepEmbedding), then a fixed conditioning vector such as a one-hot class. Its input relevance is split back into those three parts.

Constructor & Destructor Documentation

◆ ConcatDenoiser()

pulsatrix::ConcatDenoiser::ConcatDenoiser ( Module &  network,
int64_t  dim,
int64_t  time_dim,
std::vector< float >  condition,
DeviceBackend *  backend,
LRPRuleConfig  config = TrajectoryNetworkRules() 
)
Parameters
networkMaps (1, dim + time_dim + condition.size()) to (1, dim). It and the backend must outlive this.
configThe rules the network's relevance pass uses.
Exceptions
std::invalid_argumentif dim <= 0 or time_dim < 0.

Member Function Documentation

◆ condition()

const std::vector< float > & pulsatrix::ConcatDenoiser::condition ( ) const
inline

◆ input_row()

std::vector< float > pulsatrix::ConcatDenoiser::input_row ( const std::vector< float > &  x_t,
int64_t  t 
) const

The row the network reads at timestep t.

◆ predict()

std::vector< float > pulsatrix::ConcatDenoiser::predict ( const std::vector< float > &  x_t,
int64_t  t 
)
overridevirtual

The predicted noise at timestep t, the same size as x_t.

Implements pulsatrix::ContinuousDenoiser.

◆ relevance()

DenoiserRelevance pulsatrix::ConcatDenoiser::relevance ( const std::vector< float > &  relevance_eps)
overridevirtual

Relevance on the last predict()'s inputs, from relevance_eps on its output.

Implements pulsatrix::ContinuousDenoiser.

◆ set_condition()

void pulsatrix::ConcatDenoiser::set_condition ( std::vector< float >  condition)

Replaces the conditioning, for instance with another class.


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