Elm.train: Fit Extreme Learning Machine (ELM)

In ForAI: Forecasting with Artificial Intelligence

Description

Fit Extreme Learning Machine.

Usage

Elm.train(X.fit, Y.fit, Number.hn = 10, autorangeweight = FALSE, 
          rangeweight = 1, activation = "TANH", outputBias = FALSE,
          rangebias = 1)

Arguments

X.fit	Data matrix (numeric) containing the input values (predictors) used to train the model.
Y.fit	Response vector (numeric) used to train the model.
Number.hn	Number of units in the hidden layer. Default is 10.
autorangeweight	Option whether to use the automated range used for the weights. Default is FALSE.
rangeweight	Initial random weights on [-rangeweight,rangeweight]. The default is 1

activation	Activation function of the hidden layer neurons. Available functions are: 'TANH' (default) and 'SIG'.
outputBias	Option whether to use the bias parameter in the output layer
rangebias	Initial random bias on [-rangebias,rangebias]. The default is NULL

Value

inputWeight	Set of weights used.
biasofHN	Set of bias used.
matrixBeta	Set of weights adjusted.
matrixP	matrixP.
predictionTrain	The fitted values for the training data.
rangeweight	Used range of the random weight initialization ([-rangeweight,rangeweight]

.

activation	Activation function of the hidden layer neurons.
outputBias	Option whether to use the bias parameter in the output layer.
rangebias	Used range of the random bias initialization ([-rangebias,rangebias].

Note

To achieve better results, the use of a pre-processing technique (e.g. standardization of variables) is important.

Author(s)

Aranildo Lima

References

G.-B. Huang, Q.-Y. Zhu, C.-K. Siew (2006) Extreme learning machine: Theory and applications Neurocomputing 70 (2006) 489-501

Lima, A.R.; A.J. Cannon and W.W. Hsieh. Nonlinear Regression In Environmental Sciences Using Extreme Learning Machines. Submited to: Environmental Modelling and Software - ELSEVIER (submitted 2014/2/3).

Examples

  set.seed(123)
  attach(wtloss)
  library("scales")

  #scaling the inputs/outputs
  x.train <- rescale(as.matrix(wtloss$Days), to=c(-1,1))
  y.train <- rescale(as.matrix(wtloss$Weight), to=c(-1,1))

  #training the ELM
  trained.elm <- Elm.train(x.train,y.train,Number.hn =5)
  
  #rescaling back the elm outputs
  elm.fit.values <- rescale(trained.elm$predictionTrain,to= range(as.matrix(wtloss$Weight)),from=c(-1,1))
  
  RMSE(wtloss$Weight,elm.fit.values)
  
  oldpar <- par(mar = c(5.1, 4.1, 4.1, 4.1))
  plot(wtloss$Days, wtloss$Weight, type = "p", ylab = "Weight (kg)",main="Weight Reduction")
  lines(wtloss$Days, elm.fit.values,col=2,type='b')

ForAI documentation built on May 2, 2019, 6:14 p.m.