Data Sources


There are a wide variety of way to make data available to OJS:

  • Read CSV, JSON, SQLite, and more using the FileAttachments API.

  • Use the ojs_define() function to make data processed in Python or R available to {ojs} cells.

  • Make calls to Web APIs for online services and data stores.

We’ll explore all of these techniques below.

File Attachments

Use the FileAttachment function from the standard library to read data from a file. For example, here we read and plot a CSV of NOAA’s Monthly CO2 concentration data from Mauna Loa:

data = {
  const co2data = await FileAttachment("co2_mm.csv")
    .csv({ typed: true } );
  return => { 
    d["decimal date"] = Number(d["decimal date"]);
    d.average = Number(d.average); 
    return d; 
  marks: [
      { x: "decimal date", y: "average"}, 
      { stroke: "black" }

Note that we specified the typed: true option to the csv() function. When this option is specified d3.autoType is used to automatically detect numbers, dates, etc. and convert them to the correct JavaScript types. This is highly recommend when you know that your data is compatible with automatic type detection.

Here are the methods available for structured data formats:

Method Description
csv Comma separated values
tsv Tab separated values
json JSON (JavaScript objects)
sqlite SQLite database client
arrow Apache Arrow IPC file (uncompressed)

There are also methods to get the raw data as a blob, text, image, or stream.

Note that if you are using the arrow() method the Apache Arrow IPC file (Feather V2 file) should be written uncompressed. For example:

  compression = "uncompressed"

data = FileAttachment("data.arrow").arrow()

Python and R

The data you want to use with OJS might not always be available in raw form. Often you’ll need to read and preprocess the raw data using Python or R. You can perform this preprocessing during document render (in an {r} or {python} code cell) and then make it available to {ojs} cells via the ojs_define() function.

Here’s an example. We’ll read the same data into R, do some grouping and summarization, then make it available to OJS using ojs_define:

#| output: false


co2 = read_csv("co2_mm.csv")  %>% 
  group_by(year) %>% 
  summarize(max = max(average))

ojs_define(co2data = co2)

Note that we could have done the same thing using Python (the ojs_define function is available in any document that uses R or Python).

Now we plot the data using Observable Plot:

yearlyChart = Plot.plot({
  marks: [
      {x: "year", y: "max"}, 
      { stroke: "black" }

See the NOAA C02 example for the full source code.


You’ll note one additional twist in the OJS code above: we call the transpose function on our co2data before plotting it. The transpose function is built in to Quarto’s OJS engine, and will convert column-oriented datasets (like the ones used in Python and R) into the row-oriented datasets used by many JavaScript plotting libraries (including Plot).

For example, the following JSON data emitted from R or Python:

  "year": [1958, 1959, 1960],
  "max":  [317.51, 318.29, 320.04]

Is converted to the following via the call to transpose:

  { "year": 1959, "max": 317.51 },
  { "year": 1960, "max": 318.29 },
  { "year": 1960, "max": 320.04 }

Check the documentation for whatever plotting library you are using from OJS to see whether a call to transpose is required.

Web APIs

You can use the d3.json() function to read JSON data from web services and data sources. Here we query the GitHub API for data on contributions to the Python pandas package:

d3 = require('d3')

contributors = await d3.json(

commits = => {
  const author =;
  return {
    name: author.login,
    title: author.login,
    group: author.type,

View the data sorted by number of commits:

Inputs.table(commits, { sort: "value", reverse: true })

See the GitHub API example for the full source code.