|
pulsatrix 1.1.0
|
SoftDecisionTree's settings. More...
#include <soft_decision_tree.hpp>
Public Attributes | |
| int | depth = 4 |
| Levels of inner nodes: 2^depth leaves. Between 1 and 16. | |
| float | beta = 1.0f |
| The gates' inverse temperature β: larger is closer to a hard tree. | |
| float | penalty = 0.0f |
| The balance penalty's λ; 0 turns it off. Frosst and Hinton use 0.1 or so. | |
| uint64_t | seed = 0 |
| The gates' weights and biases start uniform in ±1/√inputs, as nn.Linear's do; the leaves start at zero, the uniform distribution. | |
SoftDecisionTree's settings.
| float pulsatrix::SoftDecisionTreeOptions::beta = 1.0f |
The gates' inverse temperature β: larger is closer to a hard tree.
| int pulsatrix::SoftDecisionTreeOptions::depth = 4 |
Levels of inner nodes: 2^depth leaves. Between 1 and 16.
| float pulsatrix::SoftDecisionTreeOptions::penalty = 0.0f |
The balance penalty's λ; 0 turns it off. Frosst and Hinton use 0.1 or so.
| uint64_t pulsatrix::SoftDecisionTreeOptions::seed = 0 |
The gates' weights and biases start uniform in ±1/√inputs, as nn.Linear's do; the leaves start at zero, the uniform distribution.