oldFunctions/fast_time_series_catch_all.R
In ryanbieber/Time-Series-Catch-All: Looks at time series data and produces the best forecast from them
#' ONLY USE IF YOU ALREADY RAN THE FIRST MODEL(time_series_catch_all) AND HAVE A VEcTOR OF MODEL NAMES
#'
#' @param data original data with Date in first column
#' @param f frequency of data i.e. month/day/year 12/365/1, interger
#' @param h forecasting steps up to 3, integer
#' @param trainStart numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param trainEnd numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param test numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param n integer, number of cores you want to use
#' @param is.XREG if the data has external regressors in an external matrix with is.XREG = TRUE
#' @param seasonaility if quarterly choose seasonaility = "Quarter" otherwise dont change default is monthly
#' @param bestModels vector of best modelNames, comes from best_model_name function
#' @param xreg, a numerical vector or matrix of external regressors, which must have the same number of rows as y. (It should not be a data frame.)
#' @param OutOfSample, boolean idicating if your forecast is out of sample or not
#'
#' @return a list of data frames that only contain the best model found by using the best_model_name function
#' @export
#'
#' @examples the fast version of time-series catch all, can oly be run after you run the first model due to the dependency on the bestModels variable
#' as it will only run those types of models in the future. To use this you need to take the output from names_of_best_model and input it into the
#' bestModels variable. therefore it will only run the functions that were chosen to be the best model.
#'
fast_time_series_catch_all <- function(data, f, h, trainStart, trainEnd, test, n, bestModels, is.XREG = FALSE, seasonaility = "Monthly", xreg = NULL, OutOfSample = FALSE){
# initializing list
cols <- colnames(data)
DF_Fast=list()
# xts timeseries
ts_data <- xts(data, order.by = data$Date)
ts_data<- t(ts_data)
ts_data<- tslist(ts_data)
for (i in 2:length(data)){ #goes through all columns assume col 1 is the date field and skipping that one
## making the train starting date dynamic as some columns have missing data at the beginning
frame <- ts_data[[1]]
na <- sum(is.na(ts_data[[i]]))
frame <- frame[na+1]
year <- as.numeric(substr(frame, 1, 4))
month <- as.numeric(substr(frame, 6, 7))
## formatting time series for other packages
#setting up the main time series list that will be used in the rest of the models
time <-ts(ts_data[[i]], frequency = f, start = trainStart)
n_train<-window(time , start = c(year, month), end = trainEnd) #training columns
if (bestModels[i-1]=="Prophet"){
#running facebooks prophet model and setting up a certain data frame as it is picky in how it likes the inputs
Z <- ts_prophet(data, h, i, OutOfSample)
} else if (bestModels[i-1]=="DLM"){
#dlm model
Z <- ts_dlm_model(time, n_train, h, seasonaility)
Z <- as.numeric(n_train)
Z <- c(a1, a)
} else if (bestModels[i-1]=="ETS"){
#exponntial smoothing
ETSe <- ets(as.double(n_train))
Zf <- fitted.values(ETSe)
Z <- forecast(ETSe, h=h)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else if (bestModels[i-1]=="ARIMAX"){
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
xregnntrain <- xreg[c(1:(NROW(n_train))),]
xregnntest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),]
## if you go to 1 forecasted value it messes up the format of multiple external regressors
if (h==1){
xregnntest <- t(as.matrix(xregnntest))
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, xreg = xregtrain, biasadj = TRUE, ic ="aicc")
Zf <- fitted.values(ARIMAa)
Z <- forecast(ARIMAa,
h=h, xreg = xregtest)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, biasadj = TRUE, ic ="aicc")
Zf <- fitted.values(ARIMAa)
Z <- forecast(ARIMAa,
h=h)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="TBATS"){
#uses tbats model
#uses tbats model
TBATSt <- tbats(as.double(n_train),
lambda=0, use.parallel = TRUE, num.cores = n)
Zf <- fitted.values(TBATSt)
TBATS <- forecast(TBATSt,
h=h)
Z<- ts(TBATS$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else if (bestModels[i-1]=="HybridE"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridIn") {
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridENN"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridInNN"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="Ensamble InvW"){
#dlm model
a <- ts_dlm_model(time, n_train, h, seasonaility)
a1 <- as.numeric(n_train)
a <- c(a1, a)
#exponntial smoothing
ETSe <- ets(as.double(n_train))
ETSf <- fitted.values(ETSe)
ETS <- forecast(ETSe, h=h)
if (is.XREG==TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
xregnntrain <- xreg[c(1:(NROW(n_train))),]
xregnntest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),]
## if you go to 1 forecasted value it messes up the format of multiple external regressors
if (h==1){
xregnntest <- t(as.matrix(xregnntest))
}
}else{}
if(is.XREG==TRUE){
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, xreg = xregtrain, biasadj = TRUE, ic ="aicc")
ARIMAf <- fitted.values(ARIMAa)
ARIMA <- forecast(ARIMA,
h=h, xreg = xregtest)
}else{
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, biasadj = TRUE, ic ="aicc")
ARIMAf <- fitted.values(ARIMAa)
ARIMA <- forecast(ARIMAa,
h=h)
}
#uses tbats model
TBATSt <- tbats(as.double(n_train),
lambda=0, use.parallel = TRUE, num.cores = n)
TBATSf <- fitted.values(TBATSt)
TBATS <- forecast(TBATSt,
h=h)
#gets the forecast values
ETS<- ts(ETS$mean,
frequency = f, start = test)
ARIMA<- ts(ARIMA$mean,
frequency = f, start = test)
TBATS<- ts(TBATS$mean,
frequency = f, start = test)
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG==TRUE){
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Hybrid1f <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Hybrid2h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Hybrid2f <- fitted.values(Hybrid2h)
Hybrid2 <- forecast(Hybrid2h,
h=h, xreg = xregtest) # ?forecasthybrid
Hybrid3h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregnntrain))
Hybrid3f <- fitted.values(Hybrid3h)
Hybrid3 <- forecast(Hybrid3h,
h=h, xreg = xregnntest)
Hybrid4h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregnntrain))
Hybrid4f <- fitted.values(Hybrid4h)
Hybrid4 <- forecast(Hybrid4h,
h=h, xreg = xregnntest)
}else{
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Hybrid1f <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Hybrid2h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Hybrid2f <- fitted.values(Hybrid2h)
Hybrid2 <- forecast(Hybrid2h,
h=h) # ?forecasthybrid
Hybrid3h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Hybrid3f <- fitted.values(Hybrid3h)
Hybrid3 <- forecast(Hybrid3h,
h=h) # ?forecasthybrid
Hybrid4h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Hybrid4f <- fitted.values(Hybrid4h)
Hybrid4 <- forecast(Hybrid4h,
h=h) # ?forecasthybrid
}
Hybrid1 <-ts(Hybrid1$mean, frequency = f, start = test)
Hybrid2 <-ts(Hybrid2$mean, frequency = f, start = test)
Hybrid3 <-ts(Hybrid3$mean, frequency = f, start = test)
Hybrid4 <-ts(Hybrid4$mean, frequency = f, start = test)
# comibining all columns to a giant data frame of forecasted values
X <- cbind( ETS=ETS,
ARIMA=ARIMA,
TBATS=TBATS,
Hybrid1,
Hybrid2,
Hybrid3,
Hybrid4,
Prophet=tail(fcastprophet, h),
DLM=tail(a, h))
Xf <- cbind.data.frame(ETSf, ARIMAf, TBATSf, Hybrid1f, Hybrid2f, Hybrid3f, Hybrid4f)
X1 <- data.frame(X)
X1 <- X1[c(1:7)]
X1[] <- lapply(X1, as.numeric)
Xf[] <- lapply(Xf, as.numeric)
colnames(Xf) <- c("ETS", "ARIMA", "TBATS", "Hybrid1", "Hybrid2", "Hybrid3", "Hybrid4")
X2 <- rbind(Xf, X1)
Q <- cbind(X2, fcastprophet, a)
k <- which(is.na(Q), arr.ind=TRUE)
Q[k] <- rowMeans(Q, na.rm=TRUE)[k[,1]]
Q <- head(Q, -h)
Z2 <- foreccomb(n_train, Q, newpreds = X)
Z1<- comb_InvW(Z2)
Z <- c(Z1$Fitted, Z1$Forecasts_Test)
}
final <- cbind(c(as.numeric(time), rep(NA, h)), Z)
colnames(final) <- c(cols[i], bestModels[i-1])
DF_Fast[[i]] <- final
print(i)
}
return(DF_Fast)
}
ryanbieber/Time-Series-Catch-All documentation built on Oct. 13, 2020, 9:59 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
#' ONLY USE IF YOU ALREADY RAN THE FIRST MODEL(time_series_catch_all) AND HAVE A VEcTOR OF MODEL NAMES
#'
#' @param data original data with Date in first column
#' @param f frequency of data i.e. month/day/year 12/365/1, interger
#' @param h forecasting steps up to 3, integer
#' @param trainStart numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param trainEnd numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param test numeric vector, i.e. c(2016, 1) for Jan 2016
#' @param n integer, number of cores you want to use
#' @param is.XREG if the data has external regressors in an external matrix with is.XREG = TRUE
#' @param seasonaility if quarterly choose seasonaility = "Quarter" otherwise dont change default is monthly
#' @param bestModels vector of best modelNames, comes from best_model_name function
#' @param xreg, a numerical vector or matrix of external regressors, which must have the same number of rows as y. (It should not be a data frame.)
#' @param OutOfSample, boolean idicating if your forecast is out of sample or not
#'
#' @return a list of data frames that only contain the best model found by using the best_model_name function
#' @export
#'
#' @examples the fast version of time-series catch all, can oly be run after you run the first model due to the dependency on the bestModels variable
#' as it will only run those types of models in the future. To use this you need to take the output from names_of_best_model and input it into the
#' bestModels variable. therefore it will only run the functions that were chosen to be the best model.
#'
fast_time_series_catch_all <- function(data, f, h, trainStart, trainEnd, test, n, bestModels, is.XREG = FALSE, seasonaility = "Monthly", xreg = NULL, OutOfSample = FALSE){
# initializing list
cols <- colnames(data)
DF_Fast=list()
# xts timeseries
ts_data <- xts(data, order.by = data$Date)
ts_data<- t(ts_data)
ts_data<- tslist(ts_data)
for (i in 2:length(data)){ #goes through all columns assume col 1 is the date field and skipping that one
## making the train starting date dynamic as some columns have missing data at the beginning
frame <- ts_data[[1]]
na <- sum(is.na(ts_data[[i]]))
frame <- frame[na+1]
year <- as.numeric(substr(frame, 1, 4))
month <- as.numeric(substr(frame, 6, 7))
## formatting time series for other packages
#setting up the main time series list that will be used in the rest of the models
time <-ts(ts_data[[i]], frequency = f, start = trainStart)
n_train<-window(time , start = c(year, month), end = trainEnd) #training columns
if (bestModels[i-1]=="Prophet"){
#running facebooks prophet model and setting up a certain data frame as it is picky in how it likes the inputs
Z <- ts_prophet(data, h, i, OutOfSample)
} else if (bestModels[i-1]=="DLM"){
#dlm model
Z <- ts_dlm_model(time, n_train, h, seasonaility)
Z <- as.numeric(n_train)
Z <- c(a1, a)
} else if (bestModels[i-1]=="ETS"){
#exponntial smoothing
ETSe <- ets(as.double(n_train))
Zf <- fitted.values(ETSe)
Z <- forecast(ETSe, h=h)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else if (bestModels[i-1]=="ARIMAX"){
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
xregnntrain <- xreg[c(1:(NROW(n_train))),]
xregnntest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),]
## if you go to 1 forecasted value it messes up the format of multiple external regressors
if (h==1){
xregnntest <- t(as.matrix(xregnntest))
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, xreg = xregtrain, biasadj = TRUE, ic ="aicc")
Zf <- fitted.values(ARIMAa)
Z <- forecast(ARIMAa,
h=h, xreg = xregtest)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, biasadj = TRUE, ic ="aicc")
Zf <- fitted.values(ARIMAa)
Z <- forecast(ARIMAa,
h=h)
Z<- ts(Z$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="TBATS"){
#uses tbats model
#uses tbats model
TBATSt <- tbats(as.double(n_train),
lambda=0, use.parallel = TRUE, num.cores = n)
Zf <- fitted.values(TBATSt)
TBATS <- forecast(TBATSt,
h=h)
Z<- ts(TBATS$mean,
frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else if (bestModels[i-1]=="HybridE"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridIn") {
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridENN"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="HybridInNN"){
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG == TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain))
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
} else {
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Zf <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Z <-ts(Hybrid1$mean, frequency = f, start = test)
Zf <- as.numeric(Zf)
Z <- as.numeric(Z)
Z <- c(Zf, Z)
}
} else if (bestModels[i-1]=="Ensamble InvW"){
#dlm model
a <- ts_dlm_model(time, n_train, h, seasonaility)
a1 <- as.numeric(n_train)
a <- c(a1, a)
#exponntial smoothing
ETSe <- ets(as.double(n_train))
ETSf <- fitted.values(ETSe)
ETS <- forecast(ETSe, h=h)
if (is.XREG==TRUE){
#setting up xregs for ARIMA and Hybridmodel
xregtrain <- xreg[c(1:(NROW(n_train))),1]
xregtest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),1]
xregnntrain <- xreg[c(1:(NROW(n_train))),]
xregnntest <- xreg[c((NROW(n_train)+1):(NROW(n_train)+h)),]
## if you go to 1 forecasted value it messes up the format of multiple external regressors
if (h==1){
xregnntest <- t(as.matrix(xregnntest))
}
}else{}
if(is.XREG==TRUE){
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, xreg = xregtrain, biasadj = TRUE, ic ="aicc")
ARIMAf <- fitted.values(ARIMAa)
ARIMA <- forecast(ARIMA,
h=h, xreg = xregtest)
}else{
#arima modelling
ARIMAa <- auto.arima(as.double(n_train),
stepwise = FALSE, parallel = TRUE, biasadj = TRUE, ic ="aicc")
ARIMAf <- fitted.values(ARIMAa)
ARIMA <- forecast(ARIMAa,
h=h)
}
#uses tbats model
TBATSt <- tbats(as.double(n_train),
lambda=0, use.parallel = TRUE, num.cores = n)
TBATSf <- fitted.values(TBATSt)
TBATS <- forecast(TBATSt,
h=h)
#gets the forecast values
ETS<- ts(ETS$mean,
frequency = f, start = test)
ARIMA<- ts(ARIMA$mean,
frequency = f, start = test)
TBATS<- ts(TBATS$mean,
frequency = f, start = test)
#setting N-train to numeric as hybridModel needs it like that
n_train<-as.numeric(n_train)
if (is.XREG==TRUE){
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Hybrid1f <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h, xreg = xregtest) # ?forecasthybrid
Hybrid2h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregtrain),
a.args = list(xreg = xregtrain))
Hybrid2f <- fitted.values(Hybrid2h)
Hybrid2 <- forecast(Hybrid2h,
h=h, xreg = xregtest) # ?forecasthybrid
Hybrid3h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregnntrain))
Hybrid3f <- fitted.values(Hybrid3h)
Hybrid3 <- forecast(Hybrid3h,
h=h, xreg = xregnntest)
Hybrid4h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,
n.args = list(xreg = xregnntrain))
Hybrid4f <- fitted.values(Hybrid4h)
Hybrid4 <- forecast(Hybrid4h,
h=h, xreg = xregnntest)
}else{
Hybrid1h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Hybrid1f <- fitted.values(Hybrid1h)
Hybrid1 <- forecast(Hybrid1h,
h=h) # ?forecasthybrid
Hybrid2h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Hybrid2f <- fitted.values(Hybrid2h)
Hybrid2 <- forecast(Hybrid2h,
h=h) # ?forecasthybrid
Hybrid3h <- hybridModel(n_train, models = "aefnt", weights="equal", parallel = TRUE, num.cores = n,)
Hybrid3f <- fitted.values(Hybrid3h)
Hybrid3 <- forecast(Hybrid3h,
h=h) # ?forecasthybrid
Hybrid4h <- hybridModel(n_train, models = "aefnt", weights="insample", parallel = TRUE, num.cores = n,)
Hybrid4f <- fitted.values(Hybrid4h)
Hybrid4 <- forecast(Hybrid4h,
h=h) # ?forecasthybrid
}
Hybrid1 <-ts(Hybrid1$mean, frequency = f, start = test)
Hybrid2 <-ts(Hybrid2$mean, frequency = f, start = test)
Hybrid3 <-ts(Hybrid3$mean, frequency = f, start = test)
Hybrid4 <-ts(Hybrid4$mean, frequency = f, start = test)
# comibining all columns to a giant data frame of forecasted values
X <- cbind( ETS=ETS,
ARIMA=ARIMA,
TBATS=TBATS,
Hybrid1,
Hybrid2,
Hybrid3,
Hybrid4,
Prophet=tail(fcastprophet, h),
DLM=tail(a, h))
Xf <- cbind.data.frame(ETSf, ARIMAf, TBATSf, Hybrid1f, Hybrid2f, Hybrid3f, Hybrid4f)
X1 <- data.frame(X)
X1 <- X1[c(1:7)]
X1[] <- lapply(X1, as.numeric)
Xf[] <- lapply(Xf, as.numeric)
colnames(Xf) <- c("ETS", "ARIMA", "TBATS", "Hybrid1", "Hybrid2", "Hybrid3", "Hybrid4")
X2 <- rbind(Xf, X1)
Q <- cbind(X2, fcastprophet, a)
k <- which(is.na(Q), arr.ind=TRUE)
Q[k] <- rowMeans(Q, na.rm=TRUE)[k[,1]]
Q <- head(Q, -h)
Z2 <- foreccomb(n_train, Q, newpreds = X)
Z1<- comb_InvW(Z2)
Z <- c(Z1$Fitted, Z1$Forecasts_Test)
}
final <- cbind(c(as.numeric(time), rep(NA, h)), Z)
colnames(final) <- c(cols[i], bestModels[i-1])
DF_Fast[[i]] <- final
print(i)
}
return(DF_Fast)
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.