Data wrangling involves cleaning and restructuring raw data into a more organized and usable form. The special characteristics of time series means that it is helpful to make use of a special data structure to store time series data.
A time series comprises a series of measurements along with information about when those measurements were taken (the time index). The tsibble package implements a convenient data structure, the tsibble, for storing time series data. To illustrate how they work, we create a simple example as follows.
example1 <- tsibble(
1 year = 2015:2019,
2 y = c(123, 39, 78, 52, 110),
3 index = year
)
str(example1)
# output:
# tbl_ts [5 × 2] (S3: tbl_ts/tbl_df/tbl/data.frame)
# ...y is the vector of values in the time series.
year column that was just created.
Applying str() to the created object shows that it has class tbl_ts (tsibble). It also shows that it also inherits from tbl_df (tibble) and base R data.frame. This means that all dplyr verbs (such as filter, select, and mutate) work as per normal. However, there are now 3 types of columns in a tsibble:
We discuss the second and third types in more detail.
The index column is what introduces the temporal component of the data. It is what associates the measurements in each row with a specific time point. The time point should be from a time class in R. The time points could correspond to yearly, monthly, weekly, daily and sub-daily intervals. Consider the examples below, noting that the annotation “[1Y]”, “[1M]”, etc. tells us the interval between observations of each time series.
aus_airpassengers# A tsibble: 47 x 2 [1Y]
Year Passengers
<dbl> <dbl>
1 1970 7.32
2 1971 7.33
3 1972 7.80
4 1973 9.38
5 1974 10.7
6 1975 11.1
7 1976 10.9
8 1977 11.3
9 1978 12.1
10 1979 13.0
# ℹ 37 more rowsTotal annual air passengers (in millions) including domestic and international aircraft passengers of air carriers registered in Australia.
us_employment# A tsibble: 143,412 x 4 [1M]
# Key: Series_ID [148]
Month Series_ID Title Employed
<mth> <chr> <chr> <dbl>
1 1939 Jan CEU0500000001 Total Private 25338
2 1939 Feb CEU0500000001 Total Private 25447
3 1939 Mar CEU0500000001 Total Private 25833
4 1939 Apr CEU0500000001 Total Private 25801
5 1939 May CEU0500000001 Total Private 26113
6 1939 Jun CEU0500000001 Total Private 26485
7 1939 Jul CEU0500000001 Total Private 26481
8 1939 Aug CEU0500000001 Total Private 26848
9 1939 Sep CEU0500000001 Total Private 27468
10 1939 Oct CEU0500000001 Total Private 27830
# ℹ 143,402 more rowsUS employment data from January 1939 to June 2019. Each ‘Series_ID’ represents different sectors of the economy.
aus_arrivals# A tsibble: 508 x 3 [1Q]
# Key: Origin [4]
Quarter Origin Arrivals
<qtr> <chr> <int>
1 1981 Q1 Japan 14763
2 1981 Q2 Japan 9321
3 1981 Q3 Japan 10166
4 1981 Q4 Japan 19509
5 1982 Q1 Japan 17117
6 1982 Q2 Japan 10617
7 1982 Q3 Japan 11737
8 1982 Q4 Japan 20961
9 1983 Q1 Japan 20671
10 1983 Q2 Japan 12235
# ℹ 498 more rowsQuarterly international arrivals to Australia from Japan, New Zealand, UK and the US. 1981Q1 - 2012Q3.
us_gasoline# A tsibble: 1,355 x 2 [1W]
Week Barrels
<week> <dbl>
1 1991 W06 6.62
2 1991 W07 6.43
3 1991 W08 6.58
4 1991 W09 7.22
5 1991 W10 6.88
6 1991 W11 6.95
7 1991 W12 7.33
8 1991 W13 6.78
9 1991 W14 7.50
10 1991 W15 6.92
# ℹ 1,345 more rowsMillion barrels per day, beginning Week 6, 1991, ending Week 3, 2017.
The easiest way to create such a column is to use one of the convenience functions from the lubridate package in R.
| Frequency | Function |
|---|---|
| Annual | Use integers in R |
| Quarterly | yearquarter() |
| Monthly | yearmonth() |
| Weekly | yearweek() |
| Daily | as_date(),ymd() |
| Sub-daily | as_datetime(), ymd_hms() |
A tsibble can store multiple time series in a single tsibble. It makes sense to do this when the timepoints overlap, and we wish to compare and contrast different time series. Consider the dataset us_employment, containing US employment data from January 1939 to June 2019.
us_employment
# Output:
# A tsibble: 143,412 x 4 [1M]
# # Key: Series_ID [148]
# Month Series_ID Title Employed
# <mth> <chr> <chr> <dbl>The printed output states that the tsibble’s key is determined by a single column: Series_ID Inspecting the metadata by running help(us_employment) tells us that each ‘Series_ID’ represents different sectors of the economy. There can of course be multiple key columns, such as in olympic_running, which records the fastest running time for each Olympic event.
olympic_running
# Output:
# # A tsibble: 312 x 4 [4Y]
# # Key: Length, Sex [14]
# Year Length Sex Time
# <int> <int> <chr> <dbl>The unique combinations of key columns is the number of time series in the tsibble.
Time series data in the wild does not automatically occur as tsibbles. They are instead stored as csv files or in other formats, and hence have to be converted to tsibble format. The function as_tsibble() converts a tibble object into a tsibble. It takes two arguments, index and key, which are used to specify the index and key columns respectively. This often has to be preceded by several dplyr function calls on the tibble to make sure it has the appropriate format. For an example on how to do this, refer to Chapter 2.1 in Hyndman and Athanasopoulos (2018).