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

CEM's settings. The defaults are the paper's MNIST settings, scaled to its 1,000 steps. More...

#include <contrastive.hpp>

Public Attributes

float kappa = 0.0f
 The margin κ the class change (or keep) must reach in the outputs.
 
float beta = 0.1f
 The L1 weight β: larger gives sparser explanations.
 
float c_init = 10.0f
 The first weight c of the prediction loss; the binary search moves it.
 
int search_steps = 9
 Rounds of the binary search over c.
 
int steps = 1000
 FISTA iterations in each round.
 
float learning_rate = 0.01f
 
std::vector< float > baseline
 Each feature's absence value; empty for 0.
 
std::vector< float > upper
 The most a pertinent negative may raise each feature to; empty for no limit.
 

Detailed Description

CEM's settings. The defaults are the paper's MNIST settings, scaled to its 1,000 steps.

Member Data Documentation

◆ baseline

std::vector<float> pulsatrix::ContrastiveOptions::baseline

Each feature's absence value; empty for 0.

◆ beta

float pulsatrix::ContrastiveOptions::beta = 0.1f

The L1 weight β: larger gives sparser explanations.

◆ c_init

float pulsatrix::ContrastiveOptions::c_init = 10.0f

The first weight c of the prediction loss; the binary search moves it.

◆ kappa

float pulsatrix::ContrastiveOptions::kappa = 0.0f

The margin κ the class change (or keep) must reach in the outputs.

◆ learning_rate

float pulsatrix::ContrastiveOptions::learning_rate = 0.01f

◆ search_steps

int pulsatrix::ContrastiveOptions::search_steps = 9

Rounds of the binary search over c.

◆ steps

int pulsatrix::ContrastiveOptions::steps = 1000

FISTA iterations in each round.

◆ upper

std::vector<float> pulsatrix::ContrastiveOptions::upper

The most a pertinent negative may raise each feature to; empty for no limit.


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