---

gnn_gradient_descent : Gradient Descent Algorithm.
[gnn_trainer : Trainers for Models.]

---

Detailed Description

The present trainer provides an implementation of the gradient descent algorithm for parameter optimization. Gradient descent is the simplest parameter optimization procedure based on the gradient. At each step, the parameters are updated using the following rule:

where is called the "learning rate", and determines the step-size that are taken. The value of the learning rate depends on the problem, but tipical values lie between . Smaller values tend to get trapped at local minima, but larger values often overshoot optimum values.

Functions
int	gnn_gradient_descent_reset (gnn_trainer *trainer)
	The trainer's "reset" implementation.
int	gnn_gradient_descent_train (gnn_trainer *trainer)
	The trainer's "train" implementation.
void	gnn_gradient_descent_destroy (gnn_trainer *trainer)
	The trainers "destroy" implementation.
gnn_trainer *	gnn_gradient_descent_new (gnn_node *node, gnn_criterion *crit, gnn_dataset *data, double mu)
	Creates a new gradient descent trainer.
double	gnn_gradient_descent_get_mu (gnn_trainer *trainer)
	Gets the learning rate.
int	gnn_gradient_descent_set_mu (gnn_trainer *trainer, double mu)
	Sets the learning rate.

---

Function Documentation

void gnn_gradient_descent_destroy (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_gradient_descent. Definition at line 146 of file gnn_gradient_descent.c.

double gnn_gradient_descent_get_mu (  gnn_trainer *    trainer )

This function returns the learning rate used by the gradient descent trainer. Parameters: trainer  A pointer to a gnn_gradient_descent. Returns: Returns the learning rate . Definition at line 244 of file gnn_gradient_descent.c.

gnn_trainer* gnn_gradient_descent_new (  gnn_node *    node, gnn_criterion *    crit, gnn_dataset *    data, double    mu )

This function creates a new gradient descent trainer (gnn_gradient_descent). It uses the learning rate given by "mu". Parameters: node  A pointer to a gnn_node. crit  A pointer to a gnn_criterion : Basic Criterion Function.. data  A pointer to a gnn_dataset : Datasets for Training.. mu  The learning rate . Returns: Returns a pointer to a new gnn_gradient_descent trainer. Definition at line 179 of file gnn_gradient_descent.c.

int gnn_gradient_descent_reset (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_gradient_descent. Returns: Returns 0 if suceeded. Definition at line 88 of file gnn_gradient_descent.c.

int gnn_gradient_descent_set_mu (  gnn_trainer *    trainer, double    mu )

This function sets a new value for the learning rate used by the gradient descent trainer. The learning rate should be strictly positive. Parameters: trainer  A pointer to a gnn_gradient_descent. mu  The learning rate . Returns: Returns 0 if suceeded. Definition at line 266 of file gnn_gradient_descent.c.

int gnn_gradient_descent_train (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_gradient_descent. Returns: Returns the mean cost. Definition at line 110 of file gnn_gradient_descent.c.

---