#include <mgnboost.h>
Inheritance diagram for MgnBoost:
Public Member Functions | |
MgnBoost (bool cvx=false, const cost::Cost &c=cost::_cost) | |
MgnBoost (std::istream &is) | |
virtual const id_t & | id () const |
virtual MgnBoost * | create () const |
Create a new object using the default constructor. | |
virtual MgnBoost * | clone () const |
Create a new object by replicating itself. | |
virtual void | train () |
Train with preset data set and sample weight. | |
Protected Member Functions | |
virtual void | train_gd () |
Training using gradient-descent. | |
Friends | |
struct | _mgn_gd |
MgnBoost is an implementation of the arc-gv boosting algorithm [1]. We add a proxy MgnCost to modify any cost function, though arc-gv only uses the exponential cost (which is the default). (Thus the use of other cost functions is just experimental.) The minimal margin is updated only before the gradient calculation, but not during the line-search step. This is exactly how arc-gv works. I've tried updating the minimal margin also in the line-search step, but it didn't work well (minimal margin remains very negative).
[1] L. Breiman. Prediction games and arcing algorithms. Neural Computation, 11(7):1493-1517, 1999.
Definition at line 51 of file mgnboost.h.
|
Definition at line 58 of file mgnboost.h. References Boosting::use_gradient_descent(). Referenced by MgnBoost::clone(), and MgnBoost::create(). |
|
Definition at line 61 of file mgnboost.h. |
|
Create a new object by replicating itself.
return new Derived(*this);
Reimplemented from Boosting. Definition at line 65 of file mgnboost.h. References MgnBoost::MgnBoost(). |
|
Create a new object using the default constructor. The code for a derived class Derived is always return new Derived(); Reimplemented from Boosting. Definition at line 64 of file mgnboost.h. References MgnBoost::MgnBoost(). |
|
Reimplemented from Boosting. |
|
Train with preset data set and sample weight.
Reimplemented from Boosting. Definition at line 13 of file mgnboost.cpp. References Boosting::convex, Boosting::grad_desc_view, and Boosting::train(). |
|
Training using gradient-descent.
Reimplemented from Boosting. Definition at line 18 of file mgnboost.cpp. References Boosting::convex, and lemga::iterative_optimize(). |
|
Definition at line 52 of file mgnboost.h. |