Elm.search.hc: Search Number of Hidden Neurons Using Hill Climbing

In ForAI: Forecasting with Artificial Intelligence

Description

Finds the number of hidden neurons using the hill climbing procedure.

Usage

Elm.search.hc(X.fit, Y.fit, n.ensem= 10, n.blocks=5, ErrorFunc=RMSE, PercentValid=20,
              maxHiddenNodes = NULL, Trace=TRUE, 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.
n.ensem	Number of ensemble members. Default is 10.
n.blocks	an integer specifying the desired number of cross-validation folds. Default is 5.
ErrorFunc	Error function to be minimized. The default is the function MSE from package qualV but the function could be customized.
PercentValid	Percentage of the data reserved for validation (if n.blocks < 2. Default is 20%.
maxHiddenNodes	Maximum number of hidden nodes. Default is NULL which means number of cases -1.
Trace	If TRUE, information is printed during the running of svres. Default is TRUE.
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 1

Value

The best number of hidden neurons found by the automatic procedure.

References

Lima, A.R., A.J. Cannon and W.W. Hsieh. Nonlinear regression in environmental sciences using extreme learning machines. Environmental Modelling and Software (submitted 2014/2/3)

Yuret, D., 1994. From genetic algorithms to efficient optimization. Technical Report 1569. MIT AI Laboratory.

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))

#Finding the best number of hidden neurons
number.hn <- Elm.search.hc(x.train,y.train)

#training the ELM
trained.elm <- Elm.train(x.train,y.train,Number.hn = number.hn)

#rescaling back the elm outputs
elm.fit.values <- rescale(trained.elm$predictionTrain,to= range(as.matrix(wtloss$Weight)),from=c(-1,1))

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.