pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::FeaturizerLoss Struct Reference

One batch's loss, split into its parts. More...

#include <featurizer.hpp>

Public Attributes

float total = 0.0f
 What was minimized: reconstruction + sparsity, except that a Matryoshka SAE adds every prefix's reconstruction error (TopKSparseAutoencoder).
 
float reconstruction = 0.0f
 Mean squared reconstruction error over every element.
 
float sparsity = 0.0f
 The sparsity penalty's contribution (0 for featurizers that set sparsity directly).
 

Detailed Description

One batch's loss, split into its parts.

Member Data Documentation

◆ reconstruction

float pulsatrix::FeaturizerLoss::reconstruction = 0.0f

Mean squared reconstruction error over every element.

◆ sparsity

float pulsatrix::FeaturizerLoss::sparsity = 0.0f

The sparsity penalty's contribution (0 for featurizers that set sparsity directly).

◆ total

float pulsatrix::FeaturizerLoss::total = 0.0f

What was minimized: reconstruction + sparsity, except that a Matryoshka SAE adds every prefix's reconstruction error (TopKSparseAutoencoder).


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