The goal of EasyFORcasting or short efor is to make it easier if you have to forecast multiple timeseries. The package supports you in creating forecasts with different methods implemented in the forecast, forecastHyrid, smooth and prophet package. Furthermore it provides some functions for evaluating these forecasts.
Following you see a possible workflow:
You can install the released version of efor from github with:
#devtools::install_github("flostracke/efor")First we load some packages for this example. The efor package contains some fictional timeseries data for demonstrating the apporaches:
library(tidyverse)
library(tsibble) # for creating nicer representation of monthly data
library(efor)
library(furrr) # for running the forecasting in parallel
library(forecast) #provides forecast mehotds
library(prophet) # provides forecast mehod
sales_data <- sales_monthly %>%
mutate(date = yearmonth(date))
sales_data
#> # A tibble: 216 x 3
#> date y iterate
#> <mth> <dbl> <chr>
#> 1 2012 Jan 1179 Article_A
#> 2 2012 Feb 516 Article_A
#> 3 2012 Mrz 381 Article_A
#> 4 2012 Apr 171 Article_A
#> 5 2012 Mai 264 Article_A
#> 6 2012 Jun 135 Article_A
#> 7 2012 Jul 225 Article_A
#> 8 2012 Aug 66 Article_A
#> 9 2012 Sep 123 Article_A
#> 10 2012 Okt 54 Article_A
#> # ... with 206 more rowsWe have some sales data for four articles. We want to create forecasts for all these articles. The efor package makes this quite easy, because it provides functionality to create forecasts for multiple articles with just one function call
The idea is that all the data has to be organised in one dataframe with the following columns:
- date: A date column
- iterate: the grouping variable. In this example it is the articlenumber
- y: the value you want to forecast
The dataframe sales_data is already meeting these requirements. You can verify the correct structure with the following function call, otherweise there would be an error:
check_input_data(sales_data)Before we start forecasts let’s quickly create a plot of the 4 different articles we want to forecast:
ggplot(sales_data, aes(x = date, y = y)) +
geom_line() +
geom_point() +
facet_wrap(~iterate) +
ggtitle("The original series") +
theme_minimal() 
We split the dataset in a train and test set. All observations from the year 2016 go into the test set. We want to create forecasts for the next 4 months of the testset and evaluate the performance of different methods.
train_data <- sales_data %>%
filter(date < "2016-01-01")
test_data <- sales_data %>%
filter(date >= "2016-01-01")Now we can apply the the auto.arima function to the dataset and create the forecasts. All the methods from the forecast package can be run in parallel.
forecasts_ar <- tf_grouped_forecasts(
train_data, # used training dataset
n_pred = 6, # number of predictions
func = auto.arima, # used forecasting method
parallel = TRUE # for runiing in parallel
)
forecasts_ar
#> # A tibble: 24 x 6
#> date iterate key y y_lo_95 y_hi_95
#> <mth> <chr> <chr> <dbl> <dbl> <dbl>
#> 1 2016 Jan Article_A auto.arima 1266. 1051. 1481.
#> 2 2016 Feb Article_A auto.arima 425. 210. 640.
#> 3 2016 Mrz Article_A auto.arima 476. 260. 691.
#> 4 2016 Apr Article_A auto.arima 367. 152. 582.
#> 5 2016 Mai Article_A auto.arima 319. 103. 534.
#> 6 2016 Jun Article_A auto.arima 255. 39.8 470.
#> 7 2016 Jan Article_B auto.arima 723 500. 946.
#> 8 2016 Feb Article_B auto.arima 501 278. 724.
#> 9 2016 Mrz Article_B auto.arima 642 419. 865.
#> 10 2016 Apr Article_B auto.arima 306 83.0 529.
#> # ... with 14 more rowsWith the same syntax you can create forecasts utilizing the prophet package: Please note that there we disable the parallel function for prophet, because there is some bug right now.
forecasts_prophet <- tf_grouped_forecasts(
train_data, # used training dataset
n_pred = 6, # number of predictions
func = prophet, # used forecasting method
parallel = FALSE #disabling parallel for prohet
)
forecasts_prophet
#> # A tibble: 24 x 6
#> date iterate key y y_lo_95 y_hi_95
#> <mth> <chr> <chr> <dbl> <dbl> <dbl>
#> 1 2016 Jan Article_A prophet 1224. 1072. 1365.
#> 2 2016 Feb Article_A prophet 589. 453. 744.
#> 3 2016 Mrz Article_A prophet 545. 409. 679.
#> 4 2016 Apr Article_A prophet 334. 196. 470.
#> 5 2016 Mai Article_A prophet 384. 240. 518.
#> 6 2016 Jun Article_A prophet 258. 114. 399.
#> 7 2016 Jan Article_B prophet 835. 691. 972.
#> 8 2016 Feb Article_B prophet 635. 494. 771.
#> 9 2016 Mrz Article_B prophet 759. 618. 899.
#> 10 2016 Apr Article_B prophet 355. 216. 496.
#> # ... with 14 more rowsIn order to create some plots and evaluate the performance we combine the forecasts into one dataset.
forecasts <- bind_rows(forecasts_ar, forecasts_prophet) %>%
mutate(date = yearmonth(date)) #reformat the date because of a bug in bind_rowsThe package brings also a function which makes it quite easy to access the performance (right now thhe mae, rmse and rsquared is calculated) of all the forecasting methods in the passed prediction dataframe:
tf_calc_metrics(forecasts, test_data) %>%
spread(metric, value)
#> # A tibble: 2 x 6
#> key mae mape mase rmse rsq
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 auto.arima 129. 25.4 0.469 192. 0.865
#> 2 prophet 135. 31.4 0.491 188. 0.869Also it is possible to access the performance of each article:
tf_calc_metrics(forecasts, test_data, detailed = TRUE)
#> # A tibble: 40 x 4
#> key metric value iterate
#> <chr> <chr> <dbl> <chr>
#> 1 auto.arima mae 60 Article_B
#> 2 prophet mae 83.8 Article_C
#> 3 prophet mae 84.5 Article_B
#> 4 auto.arima mae 115. Article_C
#> 5 auto.arima mae 154. Article_A
#> 6 prophet mae 186. Article_D
#> 7 prophet mae 187. Article_A
#> 8 auto.arima mae 189. Article_D
#> 9 auto.arima rmse 73.8 Article_B
#> 10 prophet rmse 94.0 Article_C
#> # ... with 30 more rowsFinally we create a quick graph visualising the results of the forecasts.
train_data_plot <- train_data %>%
mutate(key = "train")
test_data_plot <- test_data %>%
mutate(key = "test")
bind_rows(train_data_plot, test_data_plot) %>%
bind_rows(forecasts) %>%
filter(key %in% c("auto.arima", "train", "test")) %>%
mutate(date = yearmonth(date)) %>%
ggplot(aes(x = date, y = y, color = key)) +
geom_point() +
geom_line() +
facet_wrap(~iterate) +
ggtitle("Forecasted values for each article") +
ylab("Sales amount") +
theme_minimal()