Movatterモバイル変換

sweep: Tidy Tools for Forecasting

Description

Thesweep package "tidies" up the modeling workflow of theforecast package.

Details

The model and forecast objects are not covered bythebroom package. It includes thesw_tidy(),sw_glance(),andsw_augment() functions that work in a similar capacity asbroom functions.In addition, it providessw_tidy_decomp() to tidy decomposition, andsw_sweep() to coerceforecast objects to "tibbles" for easy visualization withggplot2and manipulation withdplyr.

To learn more aboutsweep, start with the vignettes:browseVignettes(package = "sweep")

Author(s)

Maintainer: Matt Danchomdancho@business-science.io

Authors:

• Davis Vaughandvaughan@business-science.io

See Also

Useful links:

• https://business-science.github.io/sweep/

• https://github.com/business-science/sweep

• Report bugs athttps://github.com/business-science/sweep/issues

Adds a sequential index column to a data frame

Description

Adds a sequential index column to a data frame

Usage

add_index(ret, rename_index)

Arguments

Print the ARIMA model parameters

Description

Refer to forecast:::arima.string.forecast arima.R

Usage

arima_string(object, padding = FALSE)

Arguments

Print the BATS model parameters

Description

Refer to forecast:::makeText.forecast bats.R

Usage

bats_string(object)

Arguments

Fictional sales data for bike shops purchasing Cannondale bikes

Description

A dataset containing the fictional bicycle orders spanning 2011 through 2015.Hypothetically, thebike_sales data are similar to sales data mainatainedin a business' sales data base. The unit price and model names come fromdata provided by model for the bicycle manufacturer, Cannondale (2016).The customers (bicycle shops) including name, location, etc andthe orders including quantity purchased and order dates are fictional.The data is intended for implementing business analytics techniques(e.g. forecast, clustering, etc) to identify underlying trends.

Usage

bike_sales

Format

A data frame with 15644 rows and 17 variables:

order.date

Date the order was placed

order.id

A unique order identification number

order.line

The sequential identification number for products on and order

quantity

Number of units purchased

price

The unit price of the bicycle

price.ext

The extended price = price x quantity

customer.id

A unique customer identification number

bikeshop.name

The customer name

bikeshop.city

The city that the bike shop is located

bikeshop.state

The state that the bike shop is located

latitude

The geograhpic latitude of the customer location

longitude

The geograhpic longitude of the customer location

product.id

A unique product identification number

model

The model name of the bicycle

category.primary

The main bicycle category, either "Mountain" or "Road"

category.secondary

One of nine more specific bicycle categories

frame

The bicycle frame material, either "Carbon" or "Aluminum"

Source

The 2016 bicycle model names and prices originated fromhttps://www.cannondale.com/en-us

Objects exported from other packages

Description

These objects are imported from other packages. Follow the linksbelow to see their documentation.

dplyr

%>%

Augment data according to a tidied model

Description

Given an R statistical model or other non-tidy object, add columns to theoriginal dataset such as predictions, residuals and cluster assignments.

Usage

sw_augment(x, ...)

Arguments

Details

sw_augment() is a wrapper forbroom::augment(). The benefit ofsw_augmentis that it has methods for various time-series model classes such asHoltWinters,ets,Arima, etc.

For non-time series,sw_augment() defaults tobroom::augment().The only difference is that the return is a tibble.

Note that by convention the first argument is almost alwaysdata,which specifies the original data object. This is not part of the S3signature, partly because it prevents rowwise_df_tidiers fromtaking a column name as the first argument.

See Also

broom::augment()

Default augment method

Description

By default,sw_augment() usesbroom::augment() to convert its output.

Usage

## Default S3 method:sw_augment(x, ...)

Arguments

Value

A tibble generated bybroom::augment()

Augments data

Description

Augments data

Usage

sw_augment_columns(ret, data, rename_index, timetk_idx = FALSE)

Arguments

Construct a single row summary "glance" of a model, fit, or otherobject

Description

Construct a single row summary "glance" of a model, fit, or otherobject

Usage

sw_glance(x, ...)

Arguments

Details

sw_glance() is a wrapper forbroom::glance(). The benefit ofsw_glanceis that it has methods for various time-series model classes such asHoltWinters,ets,Arima, etc.sw_glance methods always return either a one-row tibble orNULL.The single row includes summary statistics relevent to the model accuracy,which can be used to assess model fit and quality.

For non-time series,sw_glance() defaults tobroom::glance().The only difference is that the return is a tibble.

Value

single-row tibble with model summary information.

See Also

broom::glance()

Default glance method

Description

By default,sw_glance() usesbroom::glance() to convert its output.

Usage

## Default S3 method:sw_glance(x, ...)

Arguments

Value

A tibble generated bybroom::glance()

Tidy forecast objects

Description

Tidy forecast objects

Usage

sw_sweep(x, fitted = FALSE, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Details

sw_sweep is designedto coerceforecast objects from theforecast packageintotibble objects in a "tidy" format (long).The returned object contains both the actual valuesand the forecasted values including the point forecast and upper and lowerconfidence intervals.

Thetimetk_idx argument is used to modify the return format of the index.

• Iftimetk_idx = FALSE, a regularized time index is always constructed.This may be in the format of numeric values (e.g. 2010.000) or thehigher orderyearmon andyearqtr classes from thezoo package.A higher order class is attempted to be returned.

• Iftimetk_idx = TRUE and a timetk index is present, an irregular time indexwill be returned that combines the original time series (i.e. date or datetime)along with a computed future time series created usingtimetk::tk_make_future_timeseries()from thetimetk package. The... can be used to pass additional argumentstotimetk::tk_make_future_timeseries() such asinspect_weekdays,skip_values, etcthat can be useful in tuning the future time series sequence.

The index column name can be changed using therename_index argument.

Value

Returns atibble object.

See Also

timetk::tk_make_future_timeseries()

Examples

library(forecast)library(dplyr)# ETS forecastsUSAccDeaths %>%    ets() %>%    forecast(level = c(80, 95, 99)) %>%    sw_sweep()

Tidy the result of a time-series model into a summary tibble

Description

Tidy the result of a time-series model into a summary tibble

Usage

sw_tidy(x, ...)

Arguments

Details

sw_tidy() is a wrapper forbroom::tidy(). The main benefit ofsw_tidy()is that it has methods for various time-series model classes such asHoltWinters,ets,Arima, etc.sw_tidy() methods always returns a "tidy" tibble with model coefficient / parameters.

For non-time series,sw_tidy() defaults tobroom::tidy().The only difference is that the return is a tibble.The output ofsw_tidy() is always a tibble with disposable row names. It istherefore suited for further manipulation by packages like dplyr andggplot2.

Value

a tibble

See Also

broom::tidy()

Examples

library(dplyr)library(forecast)WWWusage %>%    auto.arima() %>%    sw_tidy(conf.int = TRUE)

Default tidying method

Description

By default,sw_tidy() usesbroom::tidy() to convert its output.

Usage

## Default S3 method:sw_tidy(x, ...)

Arguments

Value

A tibble generated bybroom::tidy()

Coerces decomposed time-series objects to tibble format.

Description

Coerces decomposed time-series objects to tibble format.

Usage

sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Details

sw_tidy_decomp is designedto coerce time-series objects with decompositions totibble objects.

A regularized time index is always constructed. If no time index isdetected, a sequential index is returned as a default.The index column name can be changed using therename_index argument.

Value

Returns atibble object.

Examples

library(dplyr)library(forecast)library(sweep)# Decompose ETS modelUSAccDeaths %>%    ets() %>%    sw_tidy_decomp()# Decompose STL objectUSAccDeaths %>%    stl(s.window = 'periodic') %>%    sw_tidy_decomp()

Print the TBATS model parameters

Description

Refer to forecast:::makeTextTBATS.forecast bats.R

Usage

tbats_string(object)

Arguments

Tidying methods for HoltWinters modeling of time series

Description

These methods tidyHoltWinters models of univariate timeseries.

Usage

## S3 method for class 'HoltWinters'sw_tidy(x, ...)## S3 method for class 'HoltWinters'sw_glance(x, ...)## S3 method for class 'HoltWinters'sw_augment(x, data = NULL, rename_index = "index", timetk_idx = FALSE, ...)## S3 method for class 'HoltWinters'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy() returns one row for each model parameter,with two columns:

• term: The various parameters (alpha, beta, gamma, and coefficients)

• estimate: The estimated parameter value

sw_glance() returns one row with the following columns:

• model.desc: A description of the model

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion (NA for bats / tbats)

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

sw_tidy_decomp() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• observed: The original time series

• season: The seasonal component

• trend: The trend component

• remainder: observed - (season + trend)

• seasadj: observed - season (or trend + remainder)

See Also

stats::HoltWinters()

Examples

library(dplyr)library(forecast)fit_hw <- USAccDeaths %>%    stats::HoltWinters()sw_tidy(fit_hw)sw_glance(fit_hw)sw_augment(fit_hw)sw_tidy_decomp(fit_hw)

Tidying methods for StructTS (Error, Trend, Seasonal) / exponential smoothingmodeling of time series

Description

These methods tidy the coefficients of StructTS models of univariate timeseries.

Usage

## S3 method for class 'StructTS'sw_tidy(x, ...)## S3 method for class 'StructTS'sw_glance(x, ...)## S3 method for class 'StructTS'sw_augment(x, data = NULL, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy() returns one row for each model parameter,with two columns:

• term: The model parameters

• estimate: The estimated parameter value

sw_glance() returns one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

See Also

stats::StructTS()

Examples

library(dplyr)library(forecast)fit_StructTS <- WWWusage %>%    StructTS()sw_tidy(fit_StructTS)sw_glance(fit_StructTS)sw_augment(fit_StructTS)

Tidying methods for ARIMA modeling of time series

Description

These methods tidy the coefficients of ARIMA models of univariate timeseries.

Usage

## S3 method for class 'Arima'sw_tidy(x, ...)## S3 method for class 'Arima'sw_glance(x, ...)## S3 method for class 'Arima'sw_augment(x, data = NULL, rename_index = "index", timetk_idx = FALSE, ...)## S3 method for class 'stlm'sw_tidy(x, ...)

Arguments

Value

sw_tidy() returns one row for each coefficient in the model,with five columns:

• term: The term in the nonlinear model being estimated and tested

• estimate: The estimated coefficient

sw_glance() returns one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

sw_tidy() returns the underlying ETS or ARIMA model'ssw_tidy()one row for each coefficient in the model,with five columns:

• term: The term in the nonlinear model being estimated and tested

• estimate: The estimated coefficient

See Also

stats::arima()

Examples

library(dplyr)library(forecast)fit_arima <- WWWusage %>%    auto.arima()sw_tidy(fit_arima)sw_glance(fit_arima)sw_augment(fit_arima)

Tidying methods for BATS and TBATS modeling of time series

Description

Tidying methods for BATS and TBATS modeling of time series

Usage

## S3 method for class 'bats'sw_tidy(x, ...)## S3 method for class 'bats'sw_glance(x, ...)## S3 method for class 'bats'sw_augment(x, data = NULL, rename_index = "index", timetk_idx = FALSE, ...)## S3 method for class 'bats'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy() returns one row for each model parameter,with two columns:

• term: The various parameters (lambda, alpha, gamma, etc)

• estimate: The estimated parameter value

sw_glance() returns one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion (NA for bats / tbats)

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

sw_tidy_decomp() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• observed: The original time series

• level: The level component

• slope: The slope component (Not always present)

• season: The seasonal component (Not always present)

See Also

forecast::bats(),forecast::tbats()

Examples

library(dplyr)library(forecast)fit_bats <- WWWusage %>%    bats()sw_tidy(fit_bats)sw_glance(fit_bats)sw_augment(fit_bats)

Tidying methods for decomposed time series

Description

Tidying methods for decomposed time series

Usage

## S3 method for class 'decomposed.ts'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy_decomp() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• season: The seasonal component

• trend: The trend component

• random: The error component

• seasadj: observed - season

See Also

decompose()

Examples

library(dplyr)library(forecast)fit_decomposed <- USAccDeaths %>%    decompose()sw_tidy_decomp(fit_decomposed)

Tidying methods for ETS (Error, Trend, Seasonal) exponential smoothingmodeling of time series

Description

Tidying methods for ETS (Error, Trend, Seasonal) exponential smoothingmodeling of time series

Usage

## S3 method for class 'ets'sw_tidy(x, ...)## S3 method for class 'ets'sw_glance(x, ...)## S3 method for class 'ets'sw_augment(x, data = NULL, timetk_idx = FALSE, rename_index = "index", ...)## S3 method for class 'ets'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy() returns one row for each model parameter,with two columns:

• term: The smoothing parameters (alpha, gamma) and the initial states(l, s0 through s10)

• estimate: The estimated parameter value

sw_glance() returns one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

sw_tidy_decomp() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• observed: The original time series

• level: The level component

• slope: The slope component (Not always present)

• season: The seasonal component (Not always present)

See Also

forecast::ets()

Examples

library(dplyr)library(forecast)fit_ets <- WWWusage %>%    ets()sw_tidy(fit_ets)sw_glance(fit_ets)sw_augment(fit_ets)sw_tidy_decomp(fit_ets)

Tidying methods for Nural Network Time Series models

Description

These methods tidy the coefficients of NNETAR models of univariate timeseries.

Usage

## S3 method for class 'nnetar'sw_tidy(x, ...)## S3 method for class 'nnetar'sw_glance(x, ...)## S3 method for class 'nnetar'sw_augment(x, data = NULL, timetk_idx = FALSE, rename_index = "index", ...)

Arguments

Value

sw_tidy() returns one row for each model parameter,with two columns:

• term: The smoothing parameters (alpha, gamma) and the initial states(l, s0 through s10)

• estimate: The estimated parameter value

sw_glance() returns one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model (NA)

• AIC: The Akaike Information Criterion (NA)

• BIC: The Bayesian Information Criterion (NA)

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

See Also

forecast::nnetar()

Examples

library(dplyr)library(forecast)fit_nnetar <- lynx %>%    nnetar()sw_tidy(fit_nnetar)sw_glance(fit_nnetar)sw_augment(fit_nnetar)

Tidying methods for STL (Seasonal, Trend, Level) decomposition of time series

Description

Tidying methods for STL (Seasonal, Trend, Level) decomposition of time series

Usage

## S3 method for class 'stl'sw_tidy(x, ...)## S3 method for class 'stl'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)## S3 method for class 'stlm'sw_tidy_decomp(x, timetk_idx = FALSE, rename_index = "index", ...)## S3 method for class 'stlm'sw_glance(x, ...)## S3 method for class 'stlm'sw_augment(x, data = NULL, rename_index = "index", timetk_idx = FALSE, ...)

Arguments

Value

sw_tidy() wrapssw_tidy_decomp()

sw_tidy_decomp() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• season: The seasonal component

• trend: The trend component

• remainder: observed - (season + trend)

• seasadj: observed - season (or trend + remainder)

sw_glance() returns the underlying ETS or ARIMA model'ssw_glance() results one row with the columns

• model.desc: A description of the model including thethree integer components (p, d, q) are the AR order,the degree of differencing, and the MA order.

• sigma: The square root of the estimated residual variance

• logLik: The data's log-likelihood under the model

• AIC: The Akaike Information Criterion

• BIC: The Bayesian Information Criterion

• ME: Mean error

• RMSE: Root mean squared error

• MAE: Mean absolute error

• MPE: Mean percentage error

• MAPE: Mean absolute percentage error

• MASE: Mean absolute scaled error

• ACF1: Autocorrelation of errors at lag 1

sw_augment() returns a tibble with the following time series attributes:

• index: An index is either attempted to be extracted from the model ora sequential index is created for plotting purposes

• .actual: The original time series

• .fitted: The fitted values from the model

• .resid: The residual values from the model

See Also

stats::stl()

Examples

library(dplyr)library(forecast)library(sweep)fit_stl <- USAccDeaths %>%    stl(s.window = "periodic")sw_tidy_decomp(fit_stl)

Validates data frame has column named the same name as variable rename_index

Description

Validates data frame has column named the same name as variable rename_index

Usage

validate_index(ret, rename_index)

Arguments