RSNNS (version 0.4-14)

mlp: Create and train a multi-layer perceptron (MLP)

Description

This function creates a multilayer perceptron (MLP) and trains it. MLPs are fully connected feedforward networks, and probably the most common network architecture in use. Training is usually performed by error backpropagation or a related procedure.

There are a lot of different learning functions present in SNNS that can be used together with this function, e.g., Std_Backpropagation, BackpropBatch, BackpropChunk, BackpropMomentum, BackpropWeightDecay, Rprop, Quickprop, SCG (scaled conjugate gradient), ...

Usage

mlp(x, ...)
# S3 method for default
mlp(
  x,
  y,
  size = c(5),
  maxit = 100,
  initFunc = "Randomize_Weights",
  initFuncParams = c(-0.3, 0.3),
  learnFunc = "Std_Backpropagation",
  learnFuncParams = c(0.2, 0),
  updateFunc = "Topological_Order",
  updateFuncParams = c(0),
  hiddenActFunc = "Act_Logistic",
  shufflePatterns = TRUE,
  linOut = FALSE,
  outputActFunc = if (linOut) "Act_Identity" else "Act_Logistic",
  inputsTest = NULL,
  targetsTest = NULL,
  pruneFunc = NULL,
  pruneFuncParams = NULL,
  ...
)

Value

an rsnns object.

Arguments

x

a matrix with training inputs for the network

...

additional function parameters (currently not used)

y

the corresponding targets values

size

number of units in the hidden layer(s)

maxit

maximum of iterations to learn

initFunc

the initialization function to use

initFuncParams

the parameters for the initialization function

learnFunc

the learning function to use

learnFuncParams

the parameters for the learning function

updateFunc

the update function to use

updateFuncParams

the parameters for the update function

hiddenActFunc

the activation function of all hidden units

shufflePatterns

should the patterns be shuffled?

linOut

sets the activation function of the output units to linear or logistic (ignored if outputActFunc is given)

outputActFunc

the activation function of all output units

inputsTest

a matrix with inputs to test the network

targetsTest

the corresponding targets for the test input

pruneFunc

the pruning function to use

pruneFuncParams

the parameters for the pruning function. Unlike the other functions, these have to be given in a named list. See the pruning demos for further explanation.

Details

Std_Backpropagation, BackpropBatch, e.g., have two parameters, the learning rate and the maximum output difference. The learning rate is usually a value between 0.1 and 1. It specifies the gradient descent step width. The maximum difference defines, how much difference between output and target value is treated as zero error, and not backpropagated. This parameter is used to prevent overtraining. For a complete list of the parameters of all the learning functions, see the SNNS User Manual, pp. 67.

The defaults that are set for initialization and update functions usually don't have to be changed.

References

Rosenblatt, F. (1958), 'The perceptron: A probabilistic model for information storage and organization in the brain', Psychological Review 65(6), 386--408.

Rumelhart, D. E.; Clelland, J. L. M. & Group, P. R. (1986), Parallel distributed processing :explorations in the microstructure of cognition, Mit, Cambridge, MA etc.

Zell, A. et al. (1998), 'SNNS Stuttgart Neural Network Simulator User Manual, Version 4.2', IPVR, University of Stuttgart and WSI, University of Tübingen. http://www.ra.cs.uni-tuebingen.de/SNNS/welcome.html

Zell, A. (1994), Simulation Neuronaler Netze, Addison-Wesley. (in German)

Examples

if (FALSE) demo(iris)
if (FALSE) demo(laser)
if (FALSE) demo(encoderSnnsCLib)


data(iris)

#shuffle the vector
iris <- iris[sample(1:nrow(iris),length(1:nrow(iris))),1:ncol(iris)]

irisValues <- iris[,1:4]
irisTargets <- decodeClassLabels(iris[,5])
#irisTargets <- decodeClassLabels(iris[,5], valTrue=0.9, valFalse=0.1)

iris <- splitForTrainingAndTest(irisValues, irisTargets, ratio=0.15)
iris <- normTrainingAndTestSet(iris)

model <- mlp(iris$inputsTrain, iris$targetsTrain, size=5, learnFuncParams=c(0.1), 
              maxit=50, inputsTest=iris$inputsTest, targetsTest=iris$targetsTest)

summary(model)
model
weightMatrix(model)
extractNetInfo(model)

par(mfrow=c(2,2))
plotIterativeError(model)

predictions <- predict(model,iris$inputsTest)

plotRegressionError(predictions[,2], iris$targetsTest[,2])

confusionMatrix(iris$targetsTrain,fitted.values(model))
confusionMatrix(iris$targetsTest,predictions)

plotROC(fitted.values(model)[,2], iris$targetsTrain[,2])
plotROC(predictions[,2], iris$targetsTest[,2])

#confusion matrix with 402040-method
confusionMatrix(iris$targetsTrain, encodeClassLabels(fitted.values(model),
                                                       method="402040", l=0.4, h=0.6))