newff {AMORE}R Documentation

Feedforward Neural Network

Description

Creates a feedforward artificial neural network according to the structure established by the AMORE package standard.

Usage

newff(n.inputs, n.hidden, n.outputs, learning.rate.global, momentum.global, error.criterium, Stao, hidden.layer, output.layer) 

Arguments

n.inputs Number of input neurons or predictors.
n.hidden Number of hidden layer neurons.
n.outputs Number of output layer neurons.
learning.rate.global Learning rate.
momentum.global Momentum (Set to 0 if you do not want to use it).
error.criterium Criterium used to measure to proximity of the neural network prediction to its target. Currently we can choose amongst:
  • "MSE": Mean Squared Error
  • "LMLS": Least Mean Logarithm Squared (Liano 1996).
  • "TAO": TAO Error (Pernia, 2004).
Stao Stao parameter for the TAO error criterium. Unused by the rest of criteria.
hidden.layer Activation function of the hidden layer neurons. Available functions are:
  • "purelin".
  • "tansig".
  • "sigmoid".
  • "hardlim".
output.layer Activation function of the hidden layer neurons according to the former list shown above.

Value

newff returns a feedforward neural network object.

Author(s)

Manuel Castejón Limas. manuel.castejon@unileon.es
Joaquin Ordieres Meré. joaquin.ordieres@dim.unirioja.es
Ana González Marcos. ana.gonzalez@unileon.es
Alpha V. Pernía Espinoza. alpha.pernia@alum.unirioja.es
Eliseo P. Vergara Gonzalez. eliseo.vergara@dim.unirioja.es
Francisco Javier Martinez de Pisón. francisco.martinez@dim.unirioja.es
Fernando Alba Elías. fernando.alba@unavarra.es

References

Pernia Espinoza, A.V. TAO-robust backpropagation learning algorithm. Neural Networks. In press.

Simon Haykin. Neural Networks. A comprehensive foundation. 2nd Edition.

See Also

init.neuron, random.init.NeuralNet, random.init.neuron, select.activation.function , init.neuron

Examples

#Example 1

library(AMORE)
# P is the input vector
P <- matrix(sample(seq(-1,1,length=1000), 1000, replace=FALSE), ncol=1) 
# The network will try to approximate the target P^2
target <- P^2                                   
#We create a feedforward network, with 2 neurons in the hidden layer. Tansig and Purelin activation functions.
net <- newff(n.inputs=1,n.hidden=2,n.outputs=1,learning.rate.global=1e-1, momentum.global=0.5 , error.criterium="MSE", hidden.layer="tansig", output.layer="purelin")
net <- train(net,P,target,n.epochs=100, g=adapt.NeuralNet,error.criterium="MSE", Stao=NA, report=TRUE, show.step=10 )

[Package AMORE version 0.1.1 Index]