knitr::opts_chunk$set( collapse = TRUE, comment = "", eval = TRUE )
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 maintinaing data lineage.
For demonstration purposes, in this vignette we will subset and package the mtcars
data set.
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. )
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 filesystem where you could then initialize a new github repository for the package.
The contents of mtcars20
are:
library(data.tree) df <- data.frame( pathString = file.path( "mtcars20", list.files( file.path(tempdir(), "mtcars20"), include.dirs = TRUE, recursive = TRUE ) ) ) as.Node(df)
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 datapackager.yml
file is used to configure and control the build process.
The contents are:
cat(yaml::as.yaml(yaml::yaml.load_file(file.path(tempdir(),"mtcars20","datapackager.yml"))))
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.
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.
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:
DataPackageR::project_extdata_path("mydata.csv")
in your R
or Rmd
file. This would return: e.g., r file.path(tempdir(),"mtcars20","inst","extdata","mydata.csv")
Similarly:
DataPackageR::project_path()
constructs a path to the data package root directory. (e.g., r file.path(tempdir(),"mtcars20")
)DataPackageR::project_data_path()
constructs a path to the data package data
subdirectory. (e.g., r file.path(tempdir(),"mtcars20","data")
)Raw data sets that are stored externally (outside the data package source tree) can be constructed relative to the project_path()
.
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.
Once the skeleton framework is set up, run the preprocessing code to build cars_over_20
, and reproducibly enclose it in a package.
dir.create(file.path(tempdir(),"lib")) DataPackageR:::package_build( file.path(tempdir(),"mtcars20"), install = TRUE, lib = file.path(tempdir(),"lib") )
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]
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
.
The package source directory changes after the first build.
df <- data.frame( pathString = file.path( "mtcars20", list.files( file.path(tempdir(), "mtcars20"), include.dirs = TRUE, recursive = TRUE ) ) ) as.Node(df)
After the first build, the R
directory contains mtcars.R
that has autogenerated roxygen2
markup documentation for the data package and for the cars_over20
packaged data.
The processed Rd
files can be found in man
.
The autogenerated 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:
dir.create(file.path(tempdir(),"lib")) # a temporary library directory document(file.path(tempdir(),"mtcars20"), lib = file.path(tempdir(),"lib"))
Updating documentation does not reprocess the data.
Once the the documentation is updated in R/mtcars.R
, then run package_build()
again.
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.
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 use the package we just created. 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. ?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
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")) # 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.
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.
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))
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.
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:
# 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:
# 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"))
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. cat(yaml::as.yaml(config))
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.
We may have situations where we have mutli-script pipelines. There are two ways to share data among scripts.
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.
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 amongst scripts can be turned off via:
package_build(deps = FALSE)
We recommend the following once your package is created.
You now have a data package source tree.
git init
in the package source root to initialize a new git repository.github
. see step 7This 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.
DataPackageR calculates an md5 checksum of each data object it stores, and keeps track of them in a file
called DATADIGEST
.
DATADIGEST
.DataVersion
string has been incremented in the DESCRIPTION
file.The DATADIGEST
file contains the following:
cat(readLines(file.path(tempdir(),"mtcars20","DATADIGEST")),sep="\n")
The description file has the new DataVersion
string.
cat(readLines(file.path(tempdir(),"mtcars20","DESCRIPTION")),sep="\n")
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.