Using DataPackageR

2024-09-17

Purpose

This vignette demonstrates how to use DataPackageR to build a data package. DataPackageR aims to simplify data package construction. It provides mechanisms for reproducibly preprocessing and tidying raw data into into documented, versioned, and packaged analysis-ready data sets. Long-running or computationally intensive data processing can be decoupled from the usual R CMD build process while maintaining data lineage.

For demonstration purposes, in this vignette we will subset and package the mtcars data set.

Set up a new data package.

We will set up a new data package based on the mtcars example in the README. The datapackage_skeleton() API is used to set up a new package. The user needs to provide:

library(DataPackageR)

# Let's reproducibly package the cars in the mtcars dataset with speed
# > 20. Our dataset will be called `cars_over_20`.

# Get the code file that turns the raw data to our packaged and
# processed analysis-ready dataset.
processing_code <-
    system.file("extdata", "tests", "subsetCars.Rmd", package = "DataPackageR")

# Create the package framework.
DataPackageR::datapackage_skeleton(name = "mtcars20",
                                   force = TRUE,
                                   code_files = processing_code,
                                   r_object_names = "cars_over_20",
                                   path = tempdir()
                                   #dependencies argument is empty
                                   #raw_data_dir argument is empty.
                                   )
✔ Creating '/tmp/RtmpLOx2Uq/mtcars20/'.
✔ Setting active project to "/tmp/RtmpLOx2Uq/mtcars20".
✔ Creating 'R/'.
✔ Writing 'DESCRIPTION'.
✔ Writing 'NAMESPACE'.
✔ Setting active project to "<no active project>".
✔ Setting active project to "/tmp/RtmpLOx2Uq/mtcars20".
✔ Creating 'data-raw/'.
✔ Creating 'data/'.
✔ Creating 'inst/extdata/'.

What’s in the package skeleton structure?

The process above has created a DataPackageR source tree named “mtcars20” in a temporary directory. For a real use case, you would pick a path on your file system where you could then initialize a new GitHub repository for the package.

The contents of mtcars20 are:

                levelName
1  mtcars20              
2   ¦--DESCRIPTION       
3   ¦--R                 
4   ¦--Read-and-delete-me
5   ¦--data              
6   ¦--data-raw          
7   ¦   °--subsetCars.Rmd
8   ¦--datapackager.yml  
9   °--inst              
10      °--extdata       

You should fill out the DESCRIPTION file to describe your data package. It contains a new DataVersion string that will be automatically incremented when the data package is built if the packaged data has changed.

The user-provided code files reside in data-raw. They are executed during the data package build process.

A note about the YAML config file.

A datapackager.yml file is used to configure and control the build process.

The contents are:

configuration:
  files:
    subsetCars.Rmd:
      enabled: yes
  objects: cars_over_20
  render_root:
    tmp: '8402'

The two main pieces of information in the configuration are a list of the files to be processed and the data sets the package will store.

This example packages an R data set named cars_over_20 (the name was passed to datapackage_skeleton()), which is created by the subsetCars.Rmd file.

The objects must be listed in the yaml configuration file. datapackage_skeleton() ensures this is done for you automatically.

DataPackageR provides an API for modifying this file, so it does not need to be done by hand.

Further information on the contents of the YAML configuration file, and the API are in the YAML Configuration Details vignette.

Where do I put my raw datasets?

Raw data (provided the size is not prohibitive) can be placed in inst/extdata.

The datapackage_skeleton() API has the raw_data_dir argument, which will copy the contents of raw_data_dir (and its subdirectories) into inst/extdata automatically.

In this example we are reading the mtcars data set that is already in memory, rather than from the file system.

An API to read raw data sets from within an R or Rmd processing script.

As stated in the README, in order for your processing scripts to be portable, you should not use absolute paths to files. DataPackageR provides an API to point to the data package root directory and the inst/extdata and data subdirectories. These are useful for constructing portable paths in your code to read files from these locations.

For example, to construct a path to a file named “mydata.csv” located in inst/extdata in your data package source tree:

Similarly:

Raw data sets that are stored externally (outside the data package source tree) can be constructed relative to the project_path().

YAML header metadata for R files and Rmd files.

If your processing scripts are Rmd files, the usual yaml header for rmarkdown documents should be present.

If your processing scripts are R files, you can still include a yaml header, but it should be commented with #' and it should be at the top of your R file. For example, a test R file in the DataPackageR package looks as follows:

#'---
#\'title: Sample report from R script
#'author: Greg Finak
#'date: August 1, 2018
#'---
data <- runif(100)

This will be converted to an Rmd file with a proper yaml header, which will then be turned into a vignette and indexed in the built package.

Build the data package.

Once the skeleton framework is set up, run the preprocessing code to build cars_over_20, and reproducibly enclose it in a package.

DataPackageR::package_build(file.path(tempdir(),"mtcars20"))

1 data set(s) created by subsetCars.Rmd
• cars_over_20
☘ Built all datasets!
Non-interactive NEWS.md file update.
* Added: cars_over_20
✔ Creating 'vignettes/'.
✔ Creating 'inst/doc/'.
ℹ Loading mtcars20
Writing 'NAMESPACE'
Writing 'mtcars20.Rd'
Writing 'cars_over_20.Rd'
── R CMD build ─────────────────────────────────────────────────────────────────
* checking for file ‘/tmp/RtmpLOx2Uq/mtcars20/DESCRIPTION’ ... OK
* preparing ‘mtcars20’:
* checking DESCRIPTION meta-information ... OK
* checking for LF line-endings in source and make files and shell scripts
* checking for empty or unneeded directories
* looking to see if a ‘data/datalist’ file should be added
* building ‘mtcars20_1.0.tar.gz’

Next Steps 
1. Update your package documentation.
   - Edit the documentation.R file in the package sourcedata-rawsubdirectory and update the roxygen markup. 
   - Rebuild the package documentation with document(). 
2. Add your package to source control.
   - Call git init . in the package source root directory. 
   - git add the package files. 
   - git commit your new package. 
   - Set up a github repository for your pacakge. 
   - Add the github repository as a remote of your local package repository. 
   - git push your local repository to gitub. 
[1] "/tmp/RtmpLOx2Uq/mtcars20_1.0.tar.gz"

Documenting your data set changes in NEWS.

When you build a package in interactive mode, you will be prompted to input text describing the changes to your data package (one line).

These will appear in the NEWS.md file in the following format:

DataVersion: xx.yy.zz
========
A description of your changes to the package

[The rest of the file]

Logging the build process.

DataPackageR uses the futile.logger package to log progress.

If there are errors in the processing, the script will notify you via logging to console and to /private/tmp/Test/inst/extdata/Logfiles/processing.log. Errors should be corrected and the build repeated.

If everything goes smoothly, you will have a new package built in the parent directory.

In this case we have a new package: mtcars20_1.0.tar.gz.

A note about the package source directory after building.

The package source directory changes after the first build.

                         levelName
1  mtcars20                       
2   ¦--DATADIGEST                 
3   ¦--DESCRIPTION                
4   ¦--NAMESPACE                  
5   ¦--NEWS.md                    
6   ¦--R                          
7   ¦   °--mtcars20.R             
8   ¦--Read-and-delete-me         
9   ¦--data                       
10  ¦   °--cars_over_20.rda       
11  ¦--data-raw                   
12  ¦   ¦--documentation.R        
13  ¦   ¦--subsetCars.R           
14  ¦   °--subsetCars.Rmd         
15  ¦--datapackager.yml           
16  ¦--inst                       
17  ¦   ¦--doc                    
18  ¦   ¦   ¦--subsetCars.Rmd     
19  ¦   ¦   °--subsetCars.html    
20  ¦   °--extdata                
21  ¦       °--Logfiles           
22  ¦           ¦--processing.log 
23  ¦           °--subsetCars.html
24  ¦--man                        
25  ¦   ¦--cars_over_20.Rd        
26  ¦   °--mtcars20.Rd            
27  °--vignettes                  
28      °--subsetCars.Rmd         

Update the auto-generated documentation.

After the first build, the R directory contains mtcars.R that has auto-generated roxygen2 markup documentation for the data package and for the cars_over20 packaged data.

The processed Rd files can be found in man.

The auto-generated documentation source is in the documentation.R file in data-raw.

You should update this file to properly document your objects. Then rebuild the documentation:

DataPackageR::document(file.path(tempdir(),"mtcars20"))
ℹ Loading mtcars20
[1] TRUE

Updating documentation does not reprocess the data.

Once the the documentation is updated in R/mtcars.R, then run package_build() again.

Why not just use R CMD build?

If the processing script is time consuming or the data set is particularly large, then R CMD build would run the code each time the package is installed. In such cases, raw data may not be available, or the environment to do the data processing may not be set up for each user of the data. DataPackageR decouples data processing from package building/installation for data consumers.

Installing and using the new data package.

Accessing vignettes, data sets, and data set documentation.

The package source also contains files in the vignettes and inst/doc directories that provide a log of the data processing.

When the package is installed, these will be accessible via the vignette() API.

The vignette will detail the processing performed by the subsetCars.Rmd processing script.

The data set documentation will be accessible via ?cars_over_20, and the data sets via data().

# Create a temporary library to install into.
dir.create(file.path(tempdir(),"lib"))

# Let's install the package we just created.
# This can also be done with with `install = TRUE` in package_build() or document().

install.packages(file.path(tempdir(),"mtcars20_1.0.tar.gz"),
                 type = "source", repos = NULL,
                 lib = file.path(tempdir(),"lib"))
lns <- loadNamespace
if (!"package:mtcars20"%in%search())
  attachNamespace(lns('mtcars20',lib.loc = file.path(tempdir(),"lib"))) #use library() in your code
data("cars_over_20") # load the data

cars_over_20 # now we can use it.
   speed dist
44    22   66
45    23   54
46    24   70
47    24   92
48    24   93
49    24  120
50    25   85
?cars_over_20 # See the documentation you wrote in data-raw/documentation.R.

vignettes <- vignette(package = "mtcars20", lib.loc = file.path(tempdir(),"lib"))
vignettes$results
      Package    LibPath               Item        
Topic "mtcars20" "/tmp/RtmpLOx2Uq/lib" "subsetCars"
      Title                                            
Topic "A Test Document for DataPackageR (source, html)"

Using the DataVersion.

Your downstream data analysis can depend on a specific version of the data in your data package by testing the DataVersion string in the DESCRIPTION file.

We provide an API for this:

# We can easily check the version of the data.
DataPackageR::data_version("mtcars20", lib.loc = file.path(tempdir(),"lib"))
[1] '0.1.0'

# You can use an assert to check the data version in  reports and
# analyses that use the packaged data.
assert_data_version(data_package_name = "mtcars20",
                    version_string = "0.1.0", acceptable = "equal",
                    lib.loc = file.path(tempdir(),"lib"))  #If this fails, execution stops
                                           #and provides an informative error.

Migrating old data packages.

Version 1.12.0 has moved away from controlling the build process using datasets.R and an additional masterfile argument.

The build process is now controlled via a datapackager.yml configuration file located in the package root directory. See YAML Configuration Details.

Create a datapackager.yml file.

You can migrate an old package by constructing such a config file using the construct_yml_config() API.

# Assume I have file1.Rmd and file2.R located in /data-raw, and these
# create 'object1' and 'object2' respectively.

config <- construct_yml_config(code = c("file1.Rmd", "file2.R"),
                               data = c("object1", "object2"))
cat(yaml::as.yaml(config))
configuration:
  files:
    file1.Rmd:
      enabled: yes
    file2.R:
      enabled: yes
  objects:
  - object1
  - object2
  render_root:
    tmp: '485643'

config is a newly constructed yaml configuration object. It can be written to the package directory:

path_to_package <- tempdir() # e.g., if tempdir() was the root of our package.
yml_write(config, path = path_to_package)

Now the package at path_to_package will build with version 1.12.0 or greater.

Reading data sets from Rmd files.

In versions prior to 1.12.1 we would read data sets from inst/extdata in an Rmd script using paths relative to data-raw in the data package source tree.

For example:

The old way.

# read 'myfile.csv' from inst/extdata relative to data-raw where the Rmd is rendered.
read.csv(file.path("../inst/extdata","myfile.csv"))

Now Rmd and R scripts are processed in render_root defined in the yaml config.

To read a raw data set we can get the path to the package source directory using an API call:

The new way.

# DataPackageR::project_extdata_path() returns the path to the data package inst/extdata subdirectory directory.
# DataPackageR::project_path() returns the path to the data package root directory.
# DataPackageR::project_data_path() returns the path to the data package data subdirectory directory.
read.csv(DataPackageR::project_extdata_path("myfile.csv"))

Partial builds.

We can also perform partial builds of a subset of files in a package by toggling the enabled key in the yaml config file.

This can be done with the following API:

config <- yml_disable_compile(config,filenames = "file2.R")
yml_write(config, path = path_to_package) # write modified yml to the package.
configuration:
  files:
    file1.Rmd:
      enabled: yes
    file2.R:
      enabled: no
  objects:
  - object1
  - object2
  render_root:
    tmp: '485643'

Note that the modified configuration needs to be written back to the package source directory in order for the changes to take effect.

The consequence of toggling a file to enable: no is that it will be skipped when the package is rebuilt, but the data will still be retained in the package, and the documentation will not be altered.

This is useful in situations where we have multiple data sets, and we want to re-run one script to update a specific data set, but not the other scripts because they may be too time consuming.

Multi-script pipelines.

We may have situations where we have multi-script pipelines. There are two ways to share data among scripts.

  1. file system artifacts
  2. data objects passed to subsequent scripts

File system artifacts.

The yaml configuration property render_root specifies the working directory where scripts will be rendered.

If a script writes files to the working directory, that is where files will appear. These can be read by subsequent scripts.

Passing data objects to subsequent scripts.

A script can access a data object designated to be packaged by previously ran scripts using datapackager_object_read().

For example, script2.Rmd will run after script1.Rmd. script2.Rmd needs to access a data object that has been designated to be packaged named dataset1, which was created by script1.Rmd. This data set can be accessed by script2.Rmd using the following expression:

dataset1 <- DataPackageR::datapackager_object_read("dataset1").

Passing of data objects among scripts can be turned off via:

package_build(deps = FALSE)

Next steps.

We recommend the following once your package is created.

Place your package under source control.

You now have a data package source tree.

This will let you version control your data processing code, and will provide a mechanism for sharing your package with others.

For more details on using git and GitHub with R, there is an excellent guide provided by Jenny Bryan: Happy git and GitHub for the useR and Hadley Wickham’s book on R packages.

Additional Details.

Fingerprints of stored data objects.

DataPackageR calculates an md5 checksum of each data object it stores, and keeps track of them in a file called DATADIGEST.

DATADIGEST

The DATADIGEST file contains the following:

DataVersion: 0.1.0
cars_over_20: 3ccb5b0aaa74fe7cfc0d3ca6ab0b5cf3

DESCRIPTION

The description file has the new DataVersion string.

Package: mtcars20
Title: What the Package Does (One Line, Title Case)
Version: 1.0
Authors@R: 
    person("First", "Last", , "first.last@example.com", role = c("aut", "cre"))
Description: What the package does (one paragraph).
License: `use_mit_license()`, `use_gpl3_license()` or friends to pick a
    license
Encoding: UTF-8
Roxygen: list(markdown = TRUE)
RoxygenNote: 7.3.2
DataVersion: 0.1.0
Depends: 
    R (>= 3.5.0)
Date: 2024-09-17
Suggests: 
    knitr,
    rmarkdown
VignetteBuilder: knitr