CSV on the Web R Package (csvwr)

Read and write csv tables annotated with metadata according to the "CSV on the Web" standard (CSVW).

The csvw model for tabular data describes how to annotate a group of csv tables to ensure they are interpreted correctly.

This package uses the csvw metadata schema to find tables, identify column names and cast values to the correct types.

The aim is to reduce the amount of manual work needed to parse and prepare data before it can be used in analysis.

Usage

Reading CSVW

You can use csvwr to read a csv table with json annotations into a data frame:

library(csvwr)

# Parse a csv table using json metadata :
csvw <- read_csvw("data.csv", "metadata.json")

# To extract the parsed table (with syntactic variable names and typed-columns):
csvw$tables[[1]]$dataframe

Alternatively, you can jump straight to the parsed table in one call:

read_csvw_dataframe("data.csv", "metadata.json")

Writing CSVW

You can also prepare annotations for a data frame:

# Given a data frame (saved as a csv)
d <- data.frame(x=c("a","b","c"), y=1:3)
write.csv(d, "table.csv", row.names=FALSE)

# Derive a schema
s <- derive_table_schema(d)

# Create metadata (as a list)
m <- create_metadata(tables=list(list(url="table.csv", tableSchema=s)))

# Serialise the metadata to JSON
j <- jsonlite::toJSON(m)

# Write the json to a file
cat(j, file="metadata.json")

For a complete introduction to the library please see the vignette("read-write-csvw").

Installation

You can install the latest release from CRAN:

install.packages("csvwr")

Or for the development version you can use devtools to install csvwr from GitHub:

install.packages("devtools")
devtools::install_github("Robsteranium/csvwr")

Contributing

Roadmap

Broadly speaking, the objectives are as follows:

parse csvw, creating dataframes with specified names and types (mostly implemented)
connecting associated csv tables and json files according to the conventions set out in the csvw standard (partly implemented)
support for validating a table according to a metadata document (a little implemented)
support for multiple tables (mostly implemented)
tools for writing csvw metadata, given an R data frame (partly implemented)
vignettes and documentation (mostly implemented)
scripts for running the most useful tools from the command line (not yet implemented)

It's not an urgent objective for the library to perform csv2rdf or csv2json translation although some support for csv2json is provided as this is used to test that the parsing is done correctly.

In terms of the csvw test cases provided by the standard, the following areas need to be addressed (in rough priority order):

datatypes (most of simple datatypes and some complex ones are supported, but there are more types and constraints too)
validations (there are a lot of these 😊)
propagation of inherited properties
http retrieval (readr::read_csv (and indeed utils::read.csv) accepts URIs, but the spec also involves link, dialect, and content-type headers)
referential integrity (a foundation for this is in place)
json nesting

Testing

The project currently incorporates two main parts of the csvw test suite:

In each case, we're running only that subset of test entries that can be expected to pass given that part of the standard that has thus far been implemented. Some entries will be skipped (either permanently or) while other priorities are implemented.

You can find out what needs to be implemented next by widening the subset to include the next entry.

During development, you may find it convenient to recreate one of the test entries for exploration. There is a convenience function in tests/csvw-tests-helpers.R. This isn't exported by the package so you'll need to evaluate it explicitly. You can then use it as follows:

run_entry_in_dev(16) # index number in the list of entries
run_entry_in_dev(id="manifest-json#test023") # identifier for the test

There are also some more in-depth unit tests written for this library.

We use GitHub actions to test the package against multiple architectures and the current, previous and development versions of R. If you need to test against the R-devel locally then you can use the r-devel.Dockerfile:

docker build -f r-devel.Dockerfile . --tag csvw-devel
docker run --rm "csvw-devel"

Workflow

You can use devtools::load_all() (CTRL + SHIFT + L in RStudio) to load updates and testthat::test_local() (CTRL + SHIFT + T) to run the tests.

In order to check the vignettes, you need to do devtools::install(build_vignettes=T). Then you can open e.g. vignette("read-write-csvw").

License

GPL-3

To discuss other licensing terms, please get in contact.

Other CSVW tools

There's another R implementation of csvw in the package rcsvw.

If you're interested in csvw more generally, then the RDF::Tabular ruby gem provides one of the more robust and comprehensive implementations, supporting both translation and validation.

If you're specifically interested in validation, take a look at the ODI's csvlint which implements csvw and also the OKFN's frictionless data table schemas.

If you want rdf translation, then you might like to check out Swirrl's csv2rdf and also table2qb which generates csvw annotations from csv files to describe RDF Data Cubes.

Robsteranium / csvwr