Type: Package
Title: Time Series Prediction Tools
Version: 0.1.1
Date: 2018-04-26
Author: Alberto Vico Moreno [aut, cre], Antonio Jesus Rivera Rivas [aut, ths], Maria Dolores Perez Godoy [aut, ths]
Maintainer: Alberto Vico Moreno <avm00016@red.ujaen.es>
Description: Makes the time series prediction easier by automatizing this process using four main functions: prep(), modl(), pred() and postp(). Features different preprocessing methods to homogenize variance and to remove trend and seasonality. Also has the potential to bring together different predictive models to make comparatives. Features ARIMA and Data Mining Regression models (using caret).
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
Encoding: UTF-8
LazyData: true
URL: https://github.com/avm00016/predtoolsTS
RoxygenNote: 6.0.1
Imports: caret, forecast, graphics, methods, Metrics, stats, TSPred, tseries, utils
NeedsCompilation: no
Packaged: 2018-04-29 13:45:20 UTC; Alberto Vico Moreno
Repository: CRAN
Date/Publication: 2018-04-29 13:54:50 UTC

Building predictive models

Description

This function give us the tools to build predictive models for time series.

Usage

modl(tserie, method = "arima", algorithm = NULL, formula = NULL,
  initialWindow = NULL, horizon = NULL, fixedWindow = NULL)

Arguments

tserie

A ts or prep object.

method

A string. Current methods available are "arima" and "dataMining". Method "arima" is set as default.

algorithm

A string. In case method is "dataMining", pick the algorithm you want to use. There is a complete list of available algorithms here (only regression type allowed): http://topepo.github.io/caret/train-models-by-tag.html.

formula

An integer vector. Contains the indexes from the time series wich will indicate how to extract the features. The last value will be the class index. Default value: c(1:16)

initialWindow

An integer. The initial number of consecutive values in each training set sample. Default value: 30.

horizon

An integer. The number of consecutive values in test set sample. Default value: 15.

fixedWindow

A logical: if FALSE, the training set always start at the first sample and the training set size will vary over data splits. Default value: TRUE.

Details

Returns an object modl which stores all the information related to the final chosen model (errors, parameters, model).

Currently this function covers two different methods: the widely know ARIMA and the "not so used for prediction" data mining. For the data mining we make use of the caret package.

The caret package offers plenty of data mining algorithms. For the data splitting here we use a rolling forecasting origin technique, wich works better on time series.

Value

A list is returned of class modl containing:

tserie

Original time serie.

tserieDF

Time serie converted to data frame.

method

Method used to build the model.

algorithm

If method is data mining, indicates wich algorithm was used.

horizon

Horizon for the splitting.

model

Model result from caret. It is a list, result of the caret::train function.

errors

Contains three different metrics to evaluate the model.

Author(s)

Alberto Vico Moreno

References

http://topepo.github.io/caret/index.html

See Also

prep modl.arima, modl.tsToDataFrame, modl.trControl, modl.dataMining

Examples

p <- prep(AirPassengers)
modl(p,method='arima')
modl(p,method='dataMining',algorithm='rpart')

Automatic ARIMA model

Description

Assuming "tserie" is stationary, returns the best arima model

Usage

modl.arima(tserie)

Arguments

tserie

A ts object.

Value

ARIMA model.

Author(s)

Alberto Vico Moreno

Examples

modl.arima(AirPassengers)

Train the data

Description

Train the time serie(as data frame) to build the model.

Usage

modl.dataMining(form, tserieDF, algorithm, timeControl, metric = "RMSE",
  maximize = FALSE)

Arguments

form

A formula of the form y ~ x1 + x2 + ...

tserieDF

Data frame.

algorithm

A string. Algorithm to perform the training. Full list at http://topepo.github.io/caret/train-models-by-tag.html. Only regression types allowed.

timeControl

trainControl object.

metric

A string. Specifies what summary metric will be used to select the optimal model. Possible values in caret are "RMSE" and "Rsquared". "RMSE" set as default. If you used a custom summaryFunction(see ?trainControl) your metrics will prevail over default.

maximize

A logical. Should the metric be maximized or minimized? Default is FALSE, since that is what makes sense for time series.

Value

train object

Author(s)

Alberto Vico Moreno

Examples


modl.dataMining(form=Class ~ .,
 tserieDF=modl.tsToDataFrame(AirPassengers,formula=c(1:20)),
 algorithm='rpart',
 timeControl=modl.trControl(initialWindow=30,horizon=15,fixedWindow=TRUE))

Control the splitting to train the data

Description

Creates the needed caret::trainControl object to control the training splitting.

Usage

modl.trControl(initialWindow, horizon, fixedWindow, givenSummary = FALSE)

Arguments

initialWindow

An integer. The initial number of consecutive values in each training set sample. Default value: 30.

horizon

An integer. The number of consecutive values in test set sample. Default value: 15.

fixedWindow

A logical: if FALSE, the training set always start at the first sample and the training set size will vary over data splits. Default value: TRUE.

givenSummary

A logical. Indicates if it should be used the customized summaryFunction(?trainControl for more info) modl.sumFunction or not. Default is FALSE; this will use default caret metrics.

Details

We always split using method "timeslice", wich is the better for time series. More information on how this works on http://topepo.github.io/caret/data-splitting.html#data-splitting-for-time-series.

Value

trainControl object

Author(s)

Alberto Vico Moreno

Examples

modl.trControl(initialWindow=30,horizon=15,fixedWindow=TRUE,givenSummary=TRUE)

Ts to data frame transformation

Description

Transform a ts object into a data frame using the given formula.

Usage

modl.tsToDataFrame(tserie, formula = NULL)

Arguments

tserie

A ts object.

formula

An integer vector. Contains the indexes from the tserie wich will indicate how to extract the features. The last value will be the class index. Default value: c(1:16). Has to be length 6 minimum.

Value

the time serie as data frame

Author(s)

Alberto Vico Moreno

Examples

modl.tsToDataFrame(AirPassengers,formula=c(1,3,4,5,6,7))
modl.tsToDataFrame(AirPassengers,formula=c(1:20))

Generic function

Description

Plots object prep

Usage

## S3 method for class 'pred'
plot(x, ylab = "Values", main = "Predictions", ...)

Arguments

x

pred object

ylab

ylab

main

main

...

ignored

Examples

plot(pred(modl(prep(AirPassengers))))

Generic function

Description

Plots object prep

Usage

## S3 method for class 'prep'
plot(x, ylab = "Preprocessed time serie", xlab = "", ...)

Arguments

x

prep object

ylab

ylab

xlab

xlab

...

ignored

Examples

plot(prep(AirPassengers),ylab="Stationary AisPassengers")

Post-processing of pre-processed data

Description

Using the prep data we undo the changes on a pred object.

Usage

postp(prd, pre)

Arguments

prd

A pred object.

pre

A prep object.

Value

A pred object with reverted transformations.

Author(s)

Alberto Vico Moreno

See Also

pred prep, postp.homogenize.log, postp.homogenize.boxcox, postp.detrend.differencing, postp.detrend.sfsm, postp.deseason.differencing

Examples

preprocess <- prep(AirPassengers)
prediction <- pred(modl(preprocess),n.ahead=30)
postp.prediction <- postp(prediction,preprocess)

Undo deseason(differencing)

Description

Uses inverse seasonal differences to reverse the changes

Usage

postp.deseason.differencing(tserie, nsd, firstseasons, frequency)

Arguments

tserie

A ts object.

nsd

Number of seasonal differences.

firstseasons

Values lost on the original differences

frequency

Frequency of the original time serie

Value

A ts object.

Author(s)

Alberto Vico Moreno

Examples

p <- prep.deseason.differencing(AirPassengers)
postp.deseason.differencing(p$tserie,p$nsd,p$firstseasons,frequency(AirPassengers))

Undo detrend(differencing)

Description

Uses inverse differences to revert the changes

Usage

postp.detrend.differencing(tserie, nd, firstvalues)

Arguments

tserie

A ts object.

nd

Number of differences.

firstvalues

Values lost on the original differences

Value

A ts object.

Author(s)

Alberto Vico Moreno

Examples

p <- prep.detrend.differencing(AirPassengers)
postp.detrend.differencing(p$tserie,p$nd,p$firstvalues)

Undo detrend(substracting full-means method)

Description

Undo detrend(substracting full-means method)

Usage

postp.detrend.sfsm(tserie, means, start, frequency)

Arguments

tserie

A ts object.

means

A numeric vector.

start

Start of original time serie

frequency

Frequency of the original time serie

Value

A ts object.

Author(s)

Alberto Vico Moreno

Examples

p <- prep.detrend.sfsm(AirPassengers)
postp.detrend.sfsm(p$tserie,p$means,start(AirPassengers),frequency(AirPassengers))

Undo Box-Cox transformation

Description

Undo Box-Cox transformation

Usage

postp.homogenize.boxcox(tserie, lambda)

Arguments

tserie

A ts object.

lambda

A numeric.

Value

A ts object.

Author(s)

Alberto Vico Moreno

Examples

p <- prep.homogenize.boxcox(AirPassengers)
postp.homogenize.boxcox(p$tserie,p$lambda)

Undo logarithmic transformation

Description

Uses exponent to reverse the logarithm

Usage

postp.homogenize.log(tserie)

Arguments

tserie

A ts object.

Value

A ts object.

Author(s)

Alberto Vico Moreno

Examples

postp.homogenize.log(prep.homogenize.log(AirPassengers))

Predictions

Description

Performs predictions over a trained model.

Usage

pred(model = NULL, n.ahead = 20, tserie = NULL, predictions = NULL)

Arguments

model

A modl object. Contains the trained model we want to predict with.

n.ahead

Number of values to predict ahead of the end of the original time serie. Default value is 20. Must ve lower than 100.

tserie

A ts object.

predictions

A ts object.

Details

Predicts future values over a "modl" object which can be ARIMA or data mining, and returns the predictions. Data mining predictions start right after the last value contained in the training data, so they overlap with the end of the original.

The object contains only two time series: the original one and the predictions. You can just set these series aswell.

Value

A list is returned of class pred containing:

tserie

Original time serie.

predictions

Time serie with the predictions.

Author(s)

Alberto Vico Moreno

See Also

modl pred.arima, pred.dataMining, pred.compareModels

Examples

prediction <- pred(model=modl(prep(AirPassengers)),n.ahead=25)
pred(tserie=prediction$tserie, predictions=prediction$predictions)

Predicts for ARIMA

Description

Performs predictions over an ARIMA model using the stats::predict function.

Usage

pred.arima(model, n.ahead)

Arguments

model

An ARIMA model.

n.ahead

Number of values to predict.

Value

A ts object containing the predictions.

Author(s)

Alberto Vico Moreno

Examples

pred.arima(forecast::auto.arima(prep(AirPassengers)$tserie),n.ahead=30)

Compare different predictions

Description

Plots the original time serie along with 2-5 predictive models.

Usage

pred.compareModels(originalTS, p_1, p_2, p_3 = NULL, p_4 = NULL,
  p_5 = NULL, legendNames = NULL, colors = NULL, legend = TRUE,
  legendPosition = NULL, yAxis = "Values", title = "Predictions")

Arguments

originalTS

A ts object

p_1

A ts object

p_2

A ts object

p_3

A ts object. Default is NULL.

p_4

A ts object. Default is NULL.

p_5

A ts object. Default is NULL.

legendNames

String vector with the names for the legend. Has to be same length as number of time series we are plotting(including the original one). Default is NULL.

colors

Vector with the colors. Has to be same length as number of time series we are plotting(including the original one). Default is NULL.

legend

A logical. Do we want a legend? Default is TRUE.

legendPosition

A string with the position of the legend (bottomright, topright, ...). Default is NULL.

yAxis

A string. Name for the y axis. "Values" as default.

title

A string. Title for the plot. Default is "Predictions".

Details

This function aims to ease the comparation between different predictive models by plotting them into the same graphic.

Author(s)

Alberto Vico Moreno

Examples


data(AirPassengers)
#pre-processing
p <- prep(AirPassengers)
#modelling
arima.modl <- modl(p)
cart.modl <- modl(p,method='dataMining',algorithm='rpart')
#predicting
arima.pred <- pred(arima.modl,n.ahead=30)
cart.pred <- pred(cart.modl,n.ahead=45)
#post-processing
arima.pred <- postp(arima.pred,p)
cart.pred <- postp(cart.pred,p)
#visual comparison
pred.compareModels(AirPassengers,arima.pred$predictions,cart.pred$predictions
,legendNames=c('AirPassengers','ARIMA','CART'),yAxis='Passengers',legendPosition = 'topleft')

Predicts for data mining methods

Description

Performs predictions over a data mining model using the caret::predict.train function.

Usage

pred.dataMining(model, n.ahead)

Arguments

model

A modl object.

n.ahead

Number of values to predict.

Value

A ts object containing the predictions.

Author(s)

Alberto Vico Moreno

Examples

m <- modl(prep(AirPassengers),method='dataMining',algorithm='rpart')
pred.dataMining(m,n.ahead=15)

Automatic pre-preprocessing

Description

This function performs pre-processing on a time series object(ts) to treat heterocedasticity, trend and seasonality in order to make the serie stationary.

Usage

prep(tserie, homogenize.method = "log", detrend.method = "differencing",
  nd = NULL, deseason.method = "differencing", nsd = NULL,
  detrend.first = TRUE)

Arguments

tserie

A ts object.

homogenize.method

A string. Current methods available are "log" and "boxcox". Method "log" is set as default. If you don't want to perform this transformation, set method as "none".

detrend.method

A string. Current methods available are "differencing" and "sfsm". Method "differencing" is set as default. If you don't want to perform this transformation, set method as "none".

nd

A number. Number of differences you want to apply to the "differencing" detrending method. As default its value is NULL, which means nd will be calculated internally.

deseason.method

A string. Current methods available are "differencing". Method "differencing" is set as default. If you don't want to perform this transformation, set method as "none".

nsd

A number. Number of seasonal differences you want to apply to the "differencing" deseasoning method. As default its value is NULL, which means nsd will be calculated internally.

detrend.first

A boolean. TRUE if detrending method is applied first, then deseasoning. FALSE if deseasoning method is applied first. Default is TRUE.

Details

Returns an object prep which stores all data needed to undo the changes later on.

This function provides an automatic way of pre-processing based on unit root tests, but this is not the perfect way to do it. You should always check manually if the given time serie is actually stationary, and modify the parameters according to your thoughts.

Value

A list is returned of class prep containing:

tserie

Processed ts object.

homogenize.method

Method used for homogenizing.

detrend.method

Method used for detrending.

nd

Number of differences used on detrending through differencing.

firstvalues

First nd values of the original series.

deseason.method

Method used for deseasoning.

nsd

Number of seasonal differences used on deseasoning through differencing.

firstseasons

First nsd seasons of the original series.

detrend.first

Processed ts object

means

Vector of means used in "sfsm" detrending method.

lambda

Coefficient used in "boxcox" transformation.

start

Start of the original time serie.

length

Length of the original time serie.

Author(s)

Alberto Vico Moreno

References

https://www.otexts.org/fpp/8/1

See Also

prep.homogenize.log, prep.homogenize.boxcox, prep.detrend.differencing, prep.detrend.sfsm, prep.deseason.differencing, prep.check.acf, prep.check.adf

Examples

prep(AirPassengers)
prep(AirPassengers,homogenize.method='boxcox',detrend.method='none')

Autocorrelation function

Description

Plots the autocorrelation function to check stationarity

Usage

prep.check.acf(tserie)

Arguments

tserie

a ts or a prep object

Details

For a stationary time series, the ACF will drop to zero relatively quickly, while the ACF of non-stationary data decreases slowly. Also, for non-stationary data, the value is often large and positive.

Examples

prep.check.acf(AirPassengers)
prep.check.acf(prep(AirPassengers))

Augmented Dickey-Fuller test

Description

Performs ADF test just as another tool to check stationarity.

Usage

prep.check.adf(tserie)

Arguments

tserie

a ts or a prep object

Details

Shows the results of an ADF test. A p-value<0.05 suggests the data is stationary.

Examples

prep.check.adf(AirPassengers)
prep.check.adf(prep(AirPassengers))

Deseason with differencing method

Description

Performs differencing with lag=frequency.

Usage

prep.deseason.differencing(tserie, nsd = NULL)

Arguments

tserie

a ts object

nsd

number of seasonal differences to apply. As default its value is NULL; in this case, the function will perform an automatic estimation of nsd.

Details

If no number of differences is specified, the function will make an estimation of the number of differences needed based on unit root test provided by forecast::nsdiffs

Value

A list is returned containing:

tserie

Transformed ts object.

nsd

Number of seasonal differencies applied.

firstseasons

Lost values after differencing.

Examples

prep.deseason.differencing(AirPassengers)
prep.deseason.differencing(AirPassengers,nsd=2)

Detrend with differencing method

Description

Performs differencing with lag=1.

Usage

prep.detrend.differencing(tserie, nd = NULL)

Arguments

tserie

a ts object

nd

number of differences to apply. As default its value is NULL; in this case, the function will perform an automatic estimation of nd.

Details

If no number of differences is specified, the function will make an estimation of the number of differences needed based on unit root test provided by forecast::ndiffs

Value

A list is returned containing:

tserie

Transformed ts object.

nd

Number of differencies applied.

firstvalues

Lost values after differencing.

Examples

prep.detrend.differencing(AirPassengers)
prep.detrend.differencing(AirPassengers,nd=2)

Detrend with "substracting full-season means" method

Description

Performs "substracting full-season means" method to go for a totally automatic approach.

Usage

prep.detrend.sfsm(tserie)

Arguments

tserie

a ts object

Details

Under this detrending scheme, a series is first split into segments. The length of the segments is equal to the length of seasonality(12 for monthly). The mean of the historical observations within each of these segments is substacted from every historical observation in the segment. To get the detrended serie we do: ds = xi - m Being xi the actual values on the time series and m the mean of the segment of xi

Value

A list is returned containing:

tserie

Transformed ts object.

means

Vector containing the historical means.

Examples

prep.detrend.sfsm(AirPassengers)

Box-Cox transformation

Description

Performs a Box-Cox transformation to a time serie.

Usage

prep.homogenize.boxcox(tserie)

Arguments

tserie

a ts object

Value

A list is returned containing:

boxcox

Transformed ts object.

lambda

Lambda value.

References

Box-Cox transformation: https://en.wikipedia.org/wiki/Power_transform#Box.E2.80.93Cox_transformation

Examples

prep.homogenize.log(AirPassengers)

Logarithmic transformation

Description

Performs a logarithmic transformation to a time serie.

Usage

prep.homogenize.log(tserie)

Arguments

tserie

a ts object

Value

ts object with transformed time serie

Examples

prep.homogenize.log(AirPassengers)

Generic function

Description

Prints object modl

Usage

## S3 method for class 'modl'
print(x, ...)

Arguments

x

prep object

...

ignored

Examples

print(modl(prep(AirPassengers)))

Generic function

Description

Prints object pred

Usage

## S3 method for class 'pred'
print(x, ...)

Arguments

x

prep object

...

ignored

Examples

print(pred(modl(prep(AirPassengers))))

Generic function

Description

Prints object prep

Usage

## S3 method for class 'prep'
print(x, ...)

Arguments

x

prep object

...

ignored

Examples

print(prep(AirPassengers))

Generic function

Description

Summary of object modl

Usage

## S3 method for class 'modl'
summary(object, ...)

Arguments

object

prep object

...

ignored

Examples

summary(modl(prep(AirPassengers)))

Generic function

Description

Summary of object pred

Usage

## S3 method for class 'pred'
summary(object, ...)

Arguments

object

prep object

...

ignored

Examples

summary(pred(modl(prep(AirPassengers))))

Generic function

Description

Summary of object prep

Usage

## S3 method for class 'prep'
summary(object, ...)

Arguments

object

prep object

...

ignored

Examples

summary(prep(AirPassengers))