M5comp

M5 Competition Data

Installation

You can install the package M5comp with:

devtools::install_github("xqnwang/M5comp")

Aggregation

Number of M5 series per aggregation level.

Level id	Aggregation level	Number of series
1	Unit sales of all products, aggregated for all stores/states	1
2	Unit sales of all products, aggregated for each State	3
3	Unit sales of all products, aggregated for each store	10
4	Unit sales of all products, aggregated for each category	3
5	Unit sales of all products, aggregated for each department	7
6	Unit sales of all products, aggregated for each State and category	9
7	Unit sales of all products, aggregated for each State and department	21
8	Unit sales of all products, aggregated for each store and category	30
9	Unit sales of all products, aggregated for each store and department	70
10	Unit sales of product x, aggregated for all stores/states	3049
11	Unit sales of product x, aggregated for each State	9147
12	Unit sales of product x, aggregated for each store	30490
Total	-	42840

The following table shows more simply and clearly how the aggregated time series are generated. For example, series of level 11 are aggregated by series with the same item.id and state.id.

Level id	id	item.id	dept.id	cat.id	store.id	state.id
(example)	(HOBBIES_1_001_CA_1)	(HOBBIES_1_001)	(HOBBIES_1)	(HOBBIES)	(CA_1)	(CA)
1
2						✓
3					✓
4				✓
5			✓
6				✓		✓
7			✓			✓
8				✓	✓
9			✓		✓
10		✓
11		✓				✓
12	✓

Example

The package M5comp contains two datasets: M5 and calendar.

M5

The M5 is a list of 42840 series, including 30490 bottom-level time series and 12350 aggregated time series. Each element of this list is also a list with some components, like the time series of the training periods and the level id of the series. The components of the bottom-level series and the aggregated series are partially different.

# extract bottom-level series (level 12)
library(M5comp)
data(M5)
M5_bottom <- Filter(function(l) l$level == 12, M5)
# the structure of the bottom-level series
str(M5_bottom[[1]])
#> List of 13
#>  $ id       : chr "HOBBIES_1_001_CA_1"
#>  $ level    : num 12
#>  $ item.id  : chr "HOBBIES_1_001"
#>  $ dept.id  : chr "HOBBIES_1"
#>  $ cat.id   : chr "HOBBIES"
#>  $ store.id : chr "CA_1"
#>  $ state.id : chr "CA"
#>  $ n        : int 1913
#>  $ h        : num 28
#>  $ x        : Time-Series [1:1913] from 1 to 1913: 0 0 0 0 0 0 0 0 0 0 ...
#>  $ x.price  : num [1:1913] NA NA NA NA NA NA NA NA NA NA ...
#>  $ xx.price : num [1:28] 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 ...
#>  $ xxx.price: num [1:28] 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 8.38 ...

# extract extract level 9 series
M5_l9 <- Filter(function(l) l$level == 9, M5)
# the structure of the aggregated series
str(M5_l9[[1]])
#> List of 9
#>  $ level   : num 9
#>  $ agg.by  : Named chr [1:2] "FOODS_1" "CA_1"
#>   ..- attr(*, "names")= chr [1:2] "dept.id" "store.id"
#>  $ dept.id : chr "FOODS_1"
#>  $ cat.id  : chr "FOODS"
#>  $ store.id: chr "CA_1"
#>  $ state.id: chr "CA"
#>  $ n       : int 1913
#>  $ h       : num 28
#>  $ x       : Time-Series [1:1913] from 1 to 1913: 297 284 214 175 182 191 224 263 245 176 ...
# which levels the series is aggregated on
M5_l9[[1]]$agg.by
#>   dept.id  store.id 
#> "FOODS_1"    "CA_1"

calendar

The calendar data contains information about the dates the products are sold. The calendar object is a dataframe with some components, such as date, weekday and event_name_1.

# the structure of the calendar data
data(calendar)
str(calendar)
#> 'data.frame':    1969 obs. of  14 variables:
#>  $ date        : Date, format: "2011-01-29" "2011-01-30" ...
#>  $ wm_yr_wk    : Factor w/ 282 levels "11101","11102",..: 1 1 1 1 1 1 1 2 2 2 ...
#>  $ weekday     : Factor w/ 7 levels "Monday","Tuesday",..: 6 7 1 2 3 4 5 6 7 1 ...
#>  $ wday        : Factor w/ 7 levels "1","2","3","4",..: 1 2 3 4 5 6 7 1 2 3 ...
#>  $ month       : Factor w/ 12 levels "1","2","3","4",..: 1 1 1 2 2 2 2 2 2 2 ...
#>  $ year        : Factor w/ 6 levels "2011","2012",..: 1 1 1 1 1 1 1 1 1 1 ...
#>  $ d           : Factor w/ 1969 levels "d_1","d_2","d_3",..: 1 2 3 4 5 6 7 8 9 10 ...
#>  $ event_name_1: Factor w/ 31 levels "","Chanukah End",..: 1 1 1 1 1 1 1 1 28 1 ...
#>  $ event_type_1: Factor w/ 5 levels "","Cultural",..: 1 1 1 1 1 1 1 1 5 1 ...
#>  $ event_name_2: Factor w/ 5 levels "","Cinco De Mayo",..: 1 1 1 1 1 1 1 1 1 1 ...
#>  $ event_type_2: Factor w/ 3 levels "","Cultural",..: 1 1 1 1 1 1 1 1 1 1 ...
#>  $ snap_CA     : Factor w/ 2 levels "0","1": 1 1 1 2 2 2 2 2 2 2 ...
#>  $ snap_TX     : Factor w/ 2 levels "0","1": 1 1 1 2 1 2 1 2 2 2 ...
#>  $ snap_WI     : Factor w/ 2 levels "0","1": 1 1 1 1 2 2 1 2 2 1 ...

# plot one of the series with date
library(ggplot2)
library(xts)
library(magrittr)
xts(M5[[1]]$x, order.by = calendar$date[1:M5[[1]]$n]) %>% 
   autoplot() + xlab("Time") + ylab("Sales")

# help pages
?M5
?calendar

License

This package is free and open source software, licensed under GPL-3