a set of C++ library classes
for neural networks development



class GradientDescent

Gradient descent learning algorithm - HTML documentaiton under construction.

Inheritance:


Public Methods

[more] GradientDescent (double x = DEFAULT_ETA)
[more]virtual ~GradientDescent ()
[more]void Update_delta_weights (unsigned n_layers, LayerNetTrain layer[] )
[more]void Update_weigths (unsigned n_layers, LayerNetTrain layer[])
[more]void Update_param (double err, unsigned iter)

Protected Methods

[more]void Zero_Delta_W (unsigned n_layers, LayerNetTrain layer[])


Inherited from Learning:

Public Methods

ovoid Init_rate(double val=DEFAULT_ETA)
odouble Read_rate(void)

Protected Fields

ostatic const double DEFAULT_ETA
ostatic const double DEFAULT_DECR_FACTOR
ostatic const double MIN_ETA
ostatic const double DEFAULT_DECR_LIN_FACTOR
ostatic const double DEFAULT_ALPHA
ostatic const double DEFAULT_ETA_INCR
ostatic const double DEFAULT_ETA_DECR
odouble eta


Documentation

Gradient descent learning class. It updates weights through simple gradient descent with fixed learning rate.
ovoid Zero_Delta_W(unsigned n_layers, LayerNetTrain layer[])

o GradientDescent(double x = DEFAULT_ETA)

ovirtual ~GradientDescent()

ovoid Update_delta_weights(unsigned n_layers, LayerNetTrain layer[] )

ovoid Update_weigths(unsigned n_layers, LayerNetTrain layer[])

ovoid Update_param(double err, unsigned iter)


Direct child classes:
MomentumDescent
GradDescDecrEtaLin
GradDescDecrEtaExp
BoldDriver

Alphabetic index HTML hierarchy of classes or Java


Last Updated February 2001
For comments and suggestions mail to Giorgio Valentini