pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::FeatureActivityTracker Class Reference

How often each feature fires, and how long since it last did. A feature that hasn't fired in a long window is dead: it no longer learns, and wastes capacity. One that fires on most inputs is dense, usually a sign it encodes a bias, not a feature. More...

#include <featurizer.hpp>

Public Member Functions

 FeatureActivityTracker (int64_t num_features)
 
void observe (const std::vector< float > &codes, float threshold=0.0f)
 Records a batch of codes, row-major (N, num_features); a code above threshold in magnitude counts as firing.
 
int64_t inputs_seen () const
 Inputs observed so far.
 
std::vector< double > firing_rates () const
 The share of observed inputs each feature fired on.
 
std::vector< int64_t > dead (int64_t window) const
 Features that haven't fired in the last window inputs (or ever, if fewer were seen), in index order.
 
double dead_fraction (int64_t window) const
 dead(window)'s share of all features.
 
std::vector< int64_t > dense (double rate=0.1) const
 Features firing on more than rate of the inputs observed.
 
void reset ()
 Forgets everything observed.
 

Detailed Description

How often each feature fires, and how long since it last did. A feature that hasn't fired in a long window is dead: it no longer learns, and wastes capacity. One that fires on most inputs is dense, usually a sign it encodes a bias, not a feature.

Constructor & Destructor Documentation

◆ FeatureActivityTracker()

pulsatrix::FeatureActivityTracker::FeatureActivityTracker ( int64_t  num_features)
explicit
Exceptions
std::invalid_argumentif num_features < 1.

Member Function Documentation

◆ dead()

std::vector< int64_t > pulsatrix::FeatureActivityTracker::dead ( int64_t  window) const

Features that haven't fired in the last window inputs (or ever, if fewer were seen), in index order.

◆ dead_fraction()

double pulsatrix::FeatureActivityTracker::dead_fraction ( int64_t  window) const

dead(window)'s share of all features.

◆ dense()

std::vector< int64_t > pulsatrix::FeatureActivityTracker::dense ( double  rate = 0.1) const

Features firing on more than rate of the inputs observed.

◆ firing_rates()

std::vector< double > pulsatrix::FeatureActivityTracker::firing_rates ( ) const

The share of observed inputs each feature fired on.

◆ inputs_seen()

int64_t pulsatrix::FeatureActivityTracker::inputs_seen ( ) const
inline

Inputs observed so far.

◆ observe()

void pulsatrix::FeatureActivityTracker::observe ( const std::vector< float > &  codes,
float  threshold = 0.0f 
)

Records a batch of codes, row-major (N, num_features); a code above threshold in magnitude counts as firing.

Exceptions
std::invalid_argumentfor a size that isn't a whole number of rows.

◆ reset()

void pulsatrix::FeatureActivityTracker::reset ( )

Forgets everything observed.


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