Skip to content

Convenient functions for ensemble forecasts in R combining approaches from the {forecast} package

License

Notifications You must be signed in to change notification settings

ellisp/forecastHybrid

Repository files navigation

Travis-CI Build Status Coverage Status CRAN version CRAN RStudio mirror downloads

forecastHybrid

Convenient functions for ensemble forecasts in R combining approaches from the forecast package.

For a more detailed description of the package and usage, consult the vignette.

The package is still under development, but many basic features have been implemented. Some features (such as optimized parallelization between rather than within models and automatically selecting the optimal combination of base models) have not yet been developed.

Installation

The stable release of the package is hosted on CRAN and can be installed as usual:

install.packages("forecastHybrid")

The latest development version can be installed using the devtools package.

devtools::install_github("ellisp/forecastHybrid/pkg")

Version updates to CRAN will be published frequently after new features are implemented, so the development version is not recommended unless you plan to modify the code or a particular bugfix is needed.

Usage

library(forecastHybrid)

# Build a hybrid forecast on the wineind dataset using auto.arima, ets, and tbats models.
# Each model is given equal weight
hm1 <- hybridModel(wineind, models = "aet", weights = "equal")
## Fitting the auto.arima model
## Fitting the ets model
## Fitting the tbats model
# Now plot the forecast for the next 48 periods
plot(forecast(hm1, h = 48))

plot of chunk unnamed-chunk-2

# Build the ensemble model on the same data but this time use auto.arima, nnetar, stlm, and tbats models.
hm2 <- hybridModel(wineind, models = "anst", weights = "equal")
## Fitting the auto.arima model
## Fitting the nnetar model
## Fitting the stlm model
## Fitting the tbats model
# Now plot the forecast for the next 48 periods
plot(forecast(hm2, h = 48))

plot of chunk unnamed-chunk-2

# Extract the point forecasts from this model
fc <- forecast(hm2, h = 48)
fc$mean
##           Jan      Feb      Mar      Apr      May      Jun      Jul
## 1994                                                               
## 1995 16201.35 20881.82 24108.16 24289.46 24099.33 24461.89 29209.40
## 1996 15867.52 20527.05 24004.26 24567.30 23822.42 24483.89 29136.83
## 1997 15739.09 20266.73 23692.47 24480.22 23639.51 24186.83 29006.33
## 1998 15655.18 20278.66 23605.34 24629.38 23644.49 24286.92 29177.41
##           Aug      Sep      Oct      Nov      Dec
## 1994          24963.59 26536.81 31868.13 36154.68
## 1995 26880.68 24310.59 26440.99 31474.53 35888.28
## 1996 26157.92 24352.95 25981.21 31503.84 35587.18
## 1997 26435.72 24452.37 25985.84 31418.15 36013.63
## 1998 25880.67
# Extract the (default) upper 80% and 95% prediction intervals
fc$upper
##            80%      95%
##  [1,] 28787.15 30519.50
##  [2,] 30475.49 32052.98
##  [3,] 36277.30 38475.03
##  [4,] 40369.80 41955.43
##  [5,] 20128.66 21356.34
##  [6,] 24420.61 26024.45
##  [7,] 27275.07 28949.58
##  [8,] 28878.89 30503.16
##  [9,] 27202.87 28837.47
## [10,] 28331.57 29976.48
## [11,] 32902.53 34557.70
## [12,] 29897.88 31763.53
## [13,] 28248.31 30062.69
## [14,] 30773.89 32595.66
## [15,] 35453.14 37287.12
## [16,] 40866.07 42714.60
## [17,] 19605.19 21469.25
## [18,] 24971.92 26851.88
## [19,] 27486.70 29382.63
## [20,] 29457.39 31369.24
## [21,] 27792.47 29720.15
## [22,] 28931.70 30875.09
## [23,] 33512.87 35471.87
## [24,] 30140.86 32115.33
## [25,] 28869.58 30993.51
## [26,] 31403.97 33539.96
## [27,] 36093.19 38246.66
## [28,] 41516.39 43689.85
## [29,] 20265.81 22460.26
## [30,] 25642.77 27858.52
## [31,] 28167.67 30404.75
## [32,] 30148.37 32406.66
## [33,] 28493.33 30772.68
## [34,] 29642.31 31942.56
## [35,] 34233.13 36554.08
## [36,] 30870.66 33212.13
## [37,] 29596.42 32085.79
## [38,] 32139.67 34645.79
## [39,] 36838.33 39366.93
## [40,] 42271.07 44824.70
## [41,] 21029.97 23609.61
## [42,] 26416.32 29022.24
## [43,] 28950.51 31582.68
## [44,] 30940.39 33598.63
## [45,] 29294.43 31978.54
## [46,] 30452.40 33162.14
## [47,] 35052.10 37787.25
## [48,] 31698.41 34458.73
# Extract the (default) lower 80% and 95% prediction intervals
fc$lower
##            80%       95%
##  [1,] 22270.74 20694.346
##  [2,] 22902.75 21598.537
##  [3,] 29048.62 27389.280
##  [4,] 31109.96 29327.301
##  [5,] 13051.76 11457.589
##  [6,] 17863.48 16757.327
##  [7,] 20822.76 19208.789
##  [8,] 20909.59 19696.301
##  [9,] 21027.19 19392.591
## [10,] 20254.10 19071.600
## [11,] 25891.51 24375.319
## [12,] 23228.55 21563.174
## [13,] 21393.44 19579.060
## [14,] 21816.55 20527.275
## [15,] 27740.78 26096.665
## [16,] 29780.03 28009.915
## [17,] 12562.57 10698.506
## [18,] 17174.51 15989.302
## [19,] 20323.72 18427.796
## [20,] 20181.36 18967.695
## [21,] 20509.53 18581.853
## [22,] 19616.42 18429.956
## [23,] 25116.13 23592.762
## [24,] 22681.14 20706.670
## [25,] 20845.16 18721.224
## [26,] 21253.10 19953.022
## [27,] 27058.33 25398.686
## [28,] 29081.91 27293.298
## [29,] 11974.98  9780.527
## [30,] 16808.42 15055.683
## [31,] 19715.78 17478.701
## [32,] 19789.54 18559.089
## [33,] 19881.71 17602.349
## [34,] 19268.84 18064.344
## [35,] 24690.68 23143.212
## [36,] 22024.37 19682.903
## [37,] 20191.35 17701.976
## [38,] 20937.45 19614.730
## [39,] 26673.26 24756.450
## [40,] 28685.08 26863.729
## [41,] 11283.85  8704.202
## [42,] 16570.92 13964.995
## [43,] 19005.97 16373.807
## [44,] 19559.85 18239.075
## [45,] 19153.63 16469.523
## [46,] 19061.74 17504.990
## [47,] 24435.04 21983.305
## [48,] 21269.65 18509.331
# Produce a forecast with prediction intervals at the 70%, 80%, 90%, and 95% levels
fc2 <- forecast(hm2, h = 48, level = c(70, 80, 90, 95))

Citation

To cite package ‘forecastHybrid’ in publications use:

David Shaub and Peter Ellis (2020). forecastHybrid: Convenient Functions for Ensemble Time Series Forecasts. https://gitlab.com/dashaub/forecastHybrid, https://github.com/ellisp/forecastHybrid.

A BibTeX entry for LaTeX users is

  @Manual{,
    title = {forecastHybrid: Convenient Functions for Ensemble Time Series Forecasts},
    author = {David Shaub and Peter Ellis},
    year = {2020},
    note = {https://gitlab.com/dashaub/forecastHybrid,
https://github.com/ellisp/forecastHybrid},
  }

License

(c) 2016-2020 David Shaub and Peter Ellis.

This package is free software released under the GPL-3 license.

About

Convenient functions for ensemble forecasts in R combining approaches from the {forecast} package

Resources

License

Stars

Watchers

Forks

Packages

No packages published