Skip to contents

Setup a production chain with JDemetra+

Source: vignettes/process-en.qmd 
#> Installing package into '/home/runner/work/_temp/Library'
#> (as 'lib' is unspecified)

The {rjd3production} package is useful for setting up a production pipeline for seasonally adjusted time series.

Before creating our production pipeline, we need to set up our working environment – our project. The init_env() function creates the following structure:

  • a data/ folder: our raw data
  • a Workspaces/ folder: our workspaces
  • an output/ folder: the output time series, tables and graphs
  • a specs/ folder: workspace-specific specifications (calendar regressors, outliers, etc.)
  • a BQ/ folder: quality reports and decision files
  • a DESCRIPTION file to manage our project’s dependencies
  • a .lintr file for static code analysis (formatting best practices)
  • a README.md file to explain our project
  • a Git project structure
path_project <- tempfile(pattern = "my_sa_project")
init_env(path = path_project)

In this tutorial, we will create a production pipeline using the ABS dataset from the {rjd3toolkit} package. The dataset is also available as the file ABS.csv at /home/runner/work/_temp/Library/rjd3providers/extdata/ABS.csv in the {rjd3providers} package.

library("rjd3toolkit")
path_ABS <- system.file("extdata", "ABS.csv", package = "rjd3providers")
my_data <- ABS[, seq_len(3L)]
colnames(my_data) <- substr(colnames(my_data), start = 2L, stop = 12L)

Selection of calendar regressors

If our time series are sensitive to calendar effects, we can correct for these effects using calendar regressors.

To do this, refer to the td-selection vignette for guidance on how to handle these effects and how to generate the td table containing our selected custom calendar regressors.

td <- select_td(my_data)
#> 
#> Série 0.2.09.10.M en cours... 1/3 
#> Computing spec No_TD ...Done !
#> Computing spec REG1 ...Done !
#> Computing spec REG2 ...Done !
#> Computing spec REG3 ...Done !
#> Computing spec REG5 ...Done !
#> Computing spec REG6 ...Done !
#> Computing spec LY ...Done !
#> Computing spec REG1_LY ...Done !
#> Computing spec REG2_LY ...Done !
#> Computing spec REG3_LY ...Done !
#> Computing spec REG5_LY ...Done !
#> Computing spec REG6_LY ...Done !
#> 
#> Série 0.2.08.10.M en cours... 2/3 
#> Computing spec No_TD ...Done !
#> Computing spec REG1 ...Done !
#> Computing spec REG2 ...Done !
#> Computing spec REG3 ...Done !
#> Computing spec REG5 ...Done !
#> Computing spec REG6 ...Done !
#> Computing spec LY ...Done !
#> Computing spec REG1_LY ...Done !
#> Computing spec REG2_LY ...Done !
#> Computing spec REG3_LY ...Done !
#> Computing spec REG5_LY ...Done !
#> Computing spec REG6_LY ...Done !
#> 
#> Série 0.2.07.10.M en cours... 3/3 
#> Computing spec No_TD ...Done !
#> Computing spec REG1 ...Done !
#> Computing spec REG2 ...Done !
#> Computing spec REG3 ...Done !
#> Computing spec REG5 ...Done !
#> Computing spec REG6 ...Done !
#> Computing spec LY ...Done !
#> Computing spec REG1_LY ...Done !
#> Computing spec REG2_LY ...Done !
#> Computing spec REG3_LY ...Done !
#> Computing spec REG5_LY ...Done !
#> Computing spec REG6_LY ...Done !

Creating a workspace

We will use the {rjd3workspace} package for functions relating to the creation and manipulation of workspaces.

To create a new workspace, you can either create it manually from scratch or from a dataset. If you are using external variables, calendar regressors or a custom calendar, don’t forget to place all of these within a modelling context.

At INSEE, we use the create_insee_context() function to create our contexts:

my_context <- create_insee_context(s = my_data[, 1L])

We will use the {rjd3x13} package to create the X13 specs.

  • Create a workspace from scratch
jws <- jws_new(modelling_context = my_context)
jsap <- jws_sap_new(jws, "Nouveau SAP")
add_sa_item(jsap = jsap, name = "Première série", x = my_data[, 1L], spec = x13_spec())
add_sa_item(jsap = jsap, name = "Seconde série", x = my_data[, 2L], spec = x13_spec())
#... avec autant de commande que de séries
  • Create a workspace from a dataset
jws <- create_ws_from_data(my_data)
set_context(jws, create_insee_context(s = my_data))

If we have any, we need to assign calendar regressors to each series:

jws_compute(jws)
assign_td(td = td, jws = jws)
#> Série 0.2.09.10.M, 1/3
#> Série 0.2.08.10.M, 2/3
#> Série 0.2.07.10.M, 3/3

Don’t forget to update the metadata for our workspace with the path to our raw data:

add_raw_data_path(jws, path_ABS, delimiter = "COMMA")

At last, we can save our workspace!

path_ws <- file.path(path_project, "Workspaces", "workspace_travail", "my_ws.xml")
save_workspace(jws, path_ws, replace = TRUE)

Call from the cruncher

In an R production pipeline, the cruncher plays a vital role as it enables:

  1. Updating the raw data from the data file
  2. Updating the seasonal adjustment model (according to a refresh policy)
  3. Production of outputs