rdatagouv 0.1.0 – Aymeric Stamm

rdatagouv is an R client for the public API of data.gouv.fr, the French government’s open data platform. It helps you find a dataset that matches your interests, judge whether it is usable, download it, and later re-fetch the exact same table reproducibly. Requests are built on top of the httr2 package.

Installation

The package is on CRAN under v0.1.0, so you can install it with:

# install.packages("pak") # if you don't have pak yet
pak::pak("rdatagouv")

Alternatively, you can install the development version from GitHub:

pak::pak("astamm/rdatagouv")

Quick start

rdatagouv revolves around a simple workflow: find a dataset, judge whether it is usable, fetch it into a table, and later re-fetch the exact same table reproducibly. The examples below hit the live data.gouv.fr API and show real results.

1. Find

As illustrated in Table 1, dg_find_datasets() searches the catalog by keyword and returns one row per dataset, with metadata (quality_score, views, license, …) to help you choose:

hits <- dg_find_datasets(q = "vélo", n = 5)
hits |> 
  dplyr::select(title, id, organization, quality_score) |>
  gt::gt() |> 
  gt::cols_label(
    title = "Title",
    id = "Dataset ID",
    organization = "Producer",
    quality_score = "Quality Score"
  ) |> 
  gt::fmt_percent(columns = "quality_score", decimals = 0) |>
  gt::opt_stylize(style = 4)

Table 1: First 5 datasets matching the keyword vélo (bike) on data.gouv.fr

Title	Dataset ID	Producer	Quality Score
Statistiques de subventions d’achat de vélo en Île-de-France	63a3d64d25cb230180a80dca	ile-de-france-mobilites	89%
Fréquentation mesurée dans les Parkings Vélos IDFM	63a3d64ea0830988959dde9d	ile-de-france-mobilites	89%
Nombre de places de stationnement vélo	67cad6cecc9d655511810ad9	ecolab-1	89%
Vélib - Vélos et bornes - Disponibilité temps réel	5a4e60f2b595080ee8014056	ville-de-paris	89%
Plan Vélo 2021-2026	6271fa96a25bdfbfbaf11a3b	ville-de-paris	78%

You can also discover producers with dg_find_organization() or curated themes with dg_find_topics(), and narrow a search to one of them via the organization / topic arguments.

2. Judge

Before pulling, glimpse a dataset’s health and engagement in one call, or read its documented columns with dg_schema():

g <- dg_glimpse("6a6be5976a05df136d48fb7a")
g$quality$score

[1] 0.5555556

3. Fetch

As illustrated in Table 2, dg_pull_dataset() downloads the dataset’s first tabular resource and returns a single tibble, with a stable table id attached as an attribute:

tbl <- dg_pull_dataset("6397c0ff56d3963118a18345")     # C.vélo bike stations
tbl |> 
  dplyr::select(station_id, name, physical_configuration, lat, lon, altitude) |>
  dplyr::slice_head(n = 5) |>
  gt::gt() |> 
  gt::cols_label(
    station_id = "Station ID",
    name = "Station Name",
    physical_configuration = "Physical Configuration",
    lat = "Latitude",
    lon = "Longitude",
    altitude = "Altitude"
  ) |> 
  gt::opt_stylize(style = 4)

Table 2: First 5 rows of the C.vélo bike stations dataset

Station ID	Station Name	Physical Configuration	Latitude	Longitude	Altitude
4	04 UCA - Campus Cézeaux	REGULAR	NA	NA	NA
7	07 - Delille	REGULAR	NA	NA	NA
8	08A - Gaillard	REGULAR	NA	NA	NA
9	09 - Chamalières Mairie	REGULAR	NA	NA	NA
14	14 - Les Carmes	REGULAR	NA	NA	NA

At any time, you can retrieve the table’s stable id with dg_table_id():

dg_table_id(tbl)

[1] "https://www.data.gouv.fr/datasets/6397c0ff56d3963118a18345#01f5b3da-8d58-42c6-a07d-202538ad6672"

4. Re-fetch reproducibly

Save that id, and dg_refetch() gets back the exact same table later — no matter how titles or file names drift on the platform:

again <- dg_refetch(tbl)
identical(again, tbl)

[1] TRUE

5. Summarise

As illustrated in Table 3, dg_summary() computes metrics for one table (size, columns, rows, missing-value rate), and dg_summarise() does the same for many at once — even an entire dg_find_datasets() result:

dg_summarise(datasets = list(iris = iris, mtcars = mtcars)) |> 
  gt::gt() |> 
  gt::cols_label(
    dataset = "Dataset",
    size_kb = "Size (KB)",
    n_vars = "Number of Variables",
    n_numeric = "Numeric Variables",
    n_non_numeric = "Non-numeric Variables",
    n_rows = "Number of Rows",
    prop_missing = "Missing Rate"
  ) |> 
  gt::fmt_number(columns = c("size_kb", "n_vars", "n_numeric", "n_non_numeric", "n_rows"), decimals = 0) |>
  gt::fmt_percent(columns = "prop_missing", decimals = 1) |>
  gt::opt_stylize(style = 4)

Table 3: Summary of the iris and mtcars datasets

Dataset	Size (KB)	Number of Variables	Numeric Variables	Non-numeric Variables	Number of Rows	Missing Rate
iris	7	5	4	1	150	0.0%
mtcars	7	11	11	0	32	0.0%

For a deeper tour — including column-type control and handling parsing problems — see the vignette.

Supported formats

dg_pull_dataset() downloads the first tabular resource of a dataset among CSV, CSV.GZ, XLS, XLSX, PARQUET, TSV, TXT and JSON, and returns the parsed table as a single tibble. A ZIP resource is unpacked and its first parseable file is returned by default; all_files = TRUE keeps every contained file in one of these formats as a named list — one element per file. The delimiter of CSV/TXT resources is auto-detected (comma, semicolon, tab, pipe, …), so both standard and European-style (semicolon/comma-decimal) files are handled without special configuration. Column types are inferred by vroom, but seeded by default from data.gouv’s own csv-detective profile (use_tabular_types = TRUE) and you can force specific columns with col_types = c(col = "Date") (e.g. to make a mostly-padded ISO date column with a few non-padded stragglers read as Date, turning the stragglers into NA; explicit col_types always win). The profile is a best-effort signal — it only exists for single-file resources indexed by the tabular service, and a missing profile (or a ZIP member) falls back to type inference; pass use_tabular_types = FALSE to disable seeding entirely. Any parsing issues are attached to the returned table as an rdatagouv_problems attribute — read them with dg_problems(tbl) — rather than a noisy per-cell warning.

The discovery catalog (dg_find_datasets()) is restricted to the official tabular formats data.gouv.fr itself indexes (csv, csv.gz, xls, xlsx, parquet), so every listed dataset is in principle openable as a table. Use the format argument to narrow the catalog to datasets with a resource in a specific format (the v2 search API matches several formats as a server-side union). The exact n_resources/formats/has_table/has_schema resource columns are populated by passing resources = TRUE. Direct pulls additionally accept tsv, txt and json resources.

Disclaimer

This package is not affiliated with or endorsed by data.gouv.fr. It is an independent project, and the authors are not responsible for the content of the datasets it indexes. This first version is a proof of concept and may contain bugs. Please report any issues on the GitHub repository.

This package is the result of a joint effort initiated during the French Finist’R bootcamp which was held in Roscoff in August 2026. It served as a complex task to benchmark R package development assisted by AI. The refinements that came up during the bootcamp were made with the assistance of deepseek-v4-flash as provided by Albert API, the French government’s AI platform. The author is grateful to the Finist’R organizers and the Albert API team for their support and guidance.