Keywords - Function groups - @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Library: nn
See also: nnrpredict ann

Quantlet: nnrnet
Description: trains a one hidden layer feed forward network. The optional parameter param consists of 8 values. Boolean values for linear output, entropy error function, log probability models and for skip connections. The fifth value is the maximum value for the starting weights, the sixth the weight decay, the seventh the number of maximal iterations and the last value generates some output if equal to one.

Reference(s):

Usage: net=nnrnet (x, y, weights, size{, param {, wts}})
Input:
x n x p matrix input variables
y n x q matrix output variables
weights n x 1 vector of weights
size scalar number of hidden units
param 7 x 1 vector of parameters
wts vector of predefined weights
Output:
net.n 3 x 1 vector number of input, hidden and output units
net.nunits scalar
net.nconn vector
net.conn vector
net.decay scalar weight decay parameter
net.entropy scalar
net.softmax scalar
net.value scalar value of error function
net.wts vector of weights
net.yh n x q estimated y's

Example:
library("nn")
x = read("bank2")
y =(1:rows(x)).>100
x =(x-min(x))./(max(x)-min(x))
par = 0|1|0|1|0.7|1.0e-3|1500|0
net = nnrnet(x, y, matrix(rows(x)), 10, par)

Result:
runs a neural network with 10 hidden units for the swiss
banknote data (1 forged banknote, 0 genuine banknote).



Author: S. Klinke, 19981207
(C) MD*TECH Method and Data Technologies, 05.02.2006