[Master Index] [Index for export/lyngby]

lyngby_nn_etrain

(export/lyngby/lyngby_nn_etrain.m)


Function Synopsis

[E, Y, VOut, WOut] = lyngby_nn_etrain(X, T, VOld, WOld, Reg, ...

Help text

 lyngby_nn_etrain     - Entropic neural network, training 

       function [E, Y, VOut, WOut] = lyngby_nn_etrain(X, T, VOld, ...
           WOld, Reg, 'PropertyName', 'PropertyValue')

       Input:    X       Neural network input
                 T       Target output 
		  VOld    Old input weights
		  WOld    Old output weights
                 Reg     Regularization (weight decay)

       Property: MaxIteration  {200} Iteration stop criterion
                 Method        Optimization type
                 MinCost       {0} Iteration stop criterion
                 MinGradient   {10^(-7)} Iteration stop criterion
                 WeightAcc     [ {0} | ~0 ] Accumulate weights
                 Info          [ {0} | ~0 ] Reporting of
                               costfunction and gradient

       Output:   E     Entropic error (cost without regularization)
                 Y     Computed Outputs
                 WOut  New trained output weights or accumulated
		        weights (depending on 'WeightAcc')
                 VOut  New trained hidden weights or accumulated
                       weights (depending on 'WeightAcc') 

       This function trains a neural network, either pruned or fully
       connected. It will continue until one of the stop criterions
       are meet: maxIteration is the number of epochs (optimization
       steps), minCost is the highest acceptable value for the cost
       function, minGradient is the hightest acceptable value for the
       norm of the gradient.
   
       See also: LYNGBY, LYNGBY_NN_EMAIN, LYNGBY_NN_EFORWARD.

 $Id: lyngby_nn_etrain.m,v 1.12 2002/03/18 18:15:36 fnielsen Exp $

Cross-Reference Information

This function calls
This function is called by

Produced by mat2html on Wed Jul 29 15:43:40 2009
Cross-Directory links are: OFF