cvforecast: Core function to compute multiple forecasts by...

In evandeilton/cvforecast: Multiple forecasts based on CrossValidation

Description

This is the core function of the package. It computes multiple forecasts by the technique of Cross-Validation. The decision about the best models are based on tests as linearity, trend and fit accuracy.

Usage

cvforecast(tsdata, tsControl = cvForecastControl(), fcMethod = NULL, ...)

Arguments

tsdata	data.frame type date-value, ts, mts or xts time series objects
tsControl	generic contol with several args for the modelling process. See cvForecastControl.
fcMethod	accept the forecast method fefined by the user. This argument can be a string or a list, eg. fcMethod = "fc_ets" or a list as fcMethod = list("fc_ets", "fc_hws"). If NULL, decision is made automatically.
...	other arguments

Value

A list of class 'cvforecast' containing several objetcts from the forecasting process. It includes: betso models (less tahn 6), crossValidation statistics for all models, accuracy of all models, the control, etc. As below.

Author(s)

LOPES, J. E.

Examples

#Define cross validation parameters
myControl <- cvForecastControl(
 minObs = 14,
 stepSize = 10,
 maxHorizon = 30,
 summaryFunc=tsSummary,
 cvMethod="MAPE",
 tsfrequency='day',
 OutlierClean=FALSE,
 dateformat='%d/%m/%Y %H:%M:%S')

#Paralell execution improves the processing time
#require(doParallel)
#cl <- makeCluster(4, type='SOCK')
#registerDoParallel(cl)

#Load data
require(plyr)
data(datasample, package="cvforecast")
dadosd <- ConvertData(datasample[,1:6], dateformat='%d/%m/%Y %H:%M:%S', tsfrequency = "day", OutType="ts")
table(sapply(dadosd, class))
dim(dadosd)

#Looping example

FF <- llply(dadosd[,1:2], function(X) {
 fit <- try(cvforecast(X, myControl))
 if(class(fit) != "try-error") {
   #plot(fit)
   return(fit)
 } else NA
}, .progress = "time")

table(sapply(FF, class))
plot(FF[[1]])
sapply(FF, names)
#stopCluster(cl)

evandeilton/cvforecast documentation built on May 16, 2019, 9:36 a.m.