---

gnn_momentum : Gradient Descent with Momentum Term.
[gnn_trainer : Trainers for Models.]

---

Detailed Description

The present trainer provides an implementation of the gradient descent with momentum term algorithm for parameter optimization. At each step, the parameters are updated using the following rule:

where is the "learning rate", is the "momentum rate", and is the update.

The learning rate is tipically one order of magnitude smaller than the one used in the gradient descent algorithm. On the other hand, a tipical choice of is .

Functions
int	gnn_momentum_reset (gnn_trainer *trainer)
	Reset function.
int	gnn_momentum_train (gnn_trainer *trainer)
	Train function.
void	gnn_momentum_destroy (gnn_trainer *trainer)
	Destructor.
gnn_trainer *	gnn_momentum_new (gnn_node *node, gnn_criterion *crit, gnn_dataset *data, double mu, double eta)
	Creates a new gradient descent with momentum trainer.
double	gnn_momentum_get_mu (gnn_trainer *trainer)
	Gets the learning rate.
int	gnn_momentum_set_mu (gnn_trainer *trainer, double mu)
	Sets the learning rate.
double	gnn_momentum_get_eta (gnn_trainer *trainer)
	Gets the momentum rate.
int	gnn_momentum_set_eta (gnn_trainer *trainer, double eta)
	Sets the momentum rate.

---

Function Documentation

void gnn_momentum_destroy (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_momentum. Definition at line 154 of file gnn_momentum.c.

double gnn_momentum_get_eta (  gnn_trainer *    trainer )

This function returns the momentum rate used by the gradient descent with momentum trainer. Parameters: trainer  A pointer to a gnn_momentum. Returns: Returns the learning rate . Definition at line 321 of file gnn_momentum.c.

double gnn_momentum_get_mu (  gnn_trainer *    trainer )

This function returns the learning rate used by the gradient descent with momentum trainer. Parameters: trainer  A pointer to a gnn_momentum. Returns: Returns the learning rate . Definition at line 268 of file gnn_momentum.c.

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

This function creates a new gradient descent with momentum trainer (gnn_momentum). It uses the learning rate given by "mu" and the momentum rate given by "eta", where and . 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 . eta  The momentum rate . Returns: Returns a pointer to a new gnn_momentum trainer. Definition at line 193 of file gnn_momentum.c.

int gnn_momentum_reset (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_momentum. Returns: Returns 0 if succeeded. Definition at line 89 of file gnn_momentum.c.

int gnn_momentum_set_eta (  gnn_trainer *    trainer, double    eta )

This function sets a new value for the momentum rate used by the gradient descent with momentum trainer. The momentum rate should be positive. Parameters: trainer  A pointer to a gnn_momentum. eta  The momentum rate . Returns: Returns 0 if suceeded. Definition at line 344 of file gnn_momentum.c.

int gnn_momentum_set_mu (  gnn_trainer *    trainer, double    mu )

This function sets a new value for the learning rate used by the gradient descent with momentum trainer. The learning rate should be strictly positive. Parameters: trainer  A pointer to a gnn_momentum. mu  The learning rate . Returns: Returns 0 if suceeded. Definition at line 291 of file gnn_momentum.c.

int gnn_momentum_train (  gnn_trainer *    trainer )  [static]

Parameters: trainer  A pointer to a gnn_momentum. Returns: Returns 0 if succeeded. Definition at line 114 of file gnn_momentum.c.

---