train.svr: Train SVR

In ForAI: Forecasting with Artificial Intelligence

Description

Train SVR

Usage

train.svr(xtrain, ytrain, hypparameter, ErrorFunc, PercentValid = 20, 
          kfold = 1, SplitRandom = FALSE, kernel = "radial")

Arguments

xtrain	Data matrix (numeric) containing the input values (predictors) used to train the model.
ytrain	Response vector (numeric) used to train the model.
hypparameter	Hyper-parameters of the model.
ErrorFunc	Error function to be minimized.
PercentValid	Percentage of the data reserved for validation. Default is 20%.
kfold	if a integer value k>1 is specified, a k-fold cross validation on the training data is performed. Default is 1.
SplitRandom	Option whether to split the train set randomly. Default is FALSE.
kernel	Kernel function to be used. Default is 'radial'.

Value

hypparameter	Hyper-parameter of the best trained model.
forecast	A vector of predicted values generated by the best trained model.
svmf	An object of class "svm" containing the fitted model.
ffTrain	Error value of the training based on ErrorFunc. (It is NULL if kfold > 1)
ffValid	Error value of the validation based on ErrorFunc.
stepvars	Variables selected by the prescreening methods when they are used.

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--	or do  help(data=index)  for the standard data sets.

## The function is currently defined as
function (xtrain, ytrain, hypparameter, ErrorFunc, PercentValid = 20, 
    kfold = 1, SplitRandom = FALSE, kernel = "radial") 
{
    if (kfold > 1) {
        n.cases = nrow(xtrain)
        index.block <- xval.buffer(n.cases, kfold)
        pred.valid <- rep(Inf, n.cases)
        for (nb in 1:kfold) {
            svr.try <- try(svm(xtrain[index.block[[nb]]$train, 
                , drop = FALSE], ytrain[index.block[[nb]]$train, 
                , drop = FALSE], kernel = kernel, gamma = hypparameter[1], 
                epsilon = hypparameter[2], cost = hypparameter[3]), 
                silent = TRUE)
            if (class(svr.try) != "try-error") {
                pred.valid[index.block[[nb]]$valid] = predict(svr.try, 
                  xtrain[index.block[[nb]]$valid, , drop = FALSE])
            }
            else {
                return(list(error.svm = TRUE))
            }
        }
        return(list(error.svm = FALSE, fitted = pred.valid, ffValid = ErrorFunc(ytrain, 
            pred.valid)))
    }
    else {
        nTrain = nrow(xtrain)
        indValid <- nTrain - round((nTrain * (PercentValid/100)))
        if (SplitRandom) {
            cases <- sample(nTrain)
            x.fit <- x.train[cases, , drop = FALSE]
            y.fit <- y.train[cases]
        }
        x.fit.train <- xtrain[1:indValid, , drop = FALSE]
        x.fit.valid <- xtrain[(indValid + 1):nTrain, , drop = FALSE]
        y.fit.train <- ytrain[1:indValid]
        y.fit.valid <- ytrain[(indValid + 1):nTrain]
        svr.try <- try(svm(x.fit.train, y.fit.train, kernel = "radial", 
            gamma = hypparameter[1], epsilon = hypparameter[2], 
            cost = hypparameter[3]), silent = TRUE)
        if (class(svr.try) != "try-error") {
            sv <- svr.try
            return(list(error.svm = FALSE, fitted.train = sv$fitted, 
                fitted.valid = predict(sv, x.fit.valid), ffValid = ErrorFunc(y.fit.valid, 
                  predict(sv, x.fit.valid)), ffTrain = ErrorFunc(y.fit.train, 
                  sv$fitted)))
        }
        else {
            return(list(error.svm = TRUE))
        }
    }
  }

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