owid-catalog 0.4.2

Core data types used by OWID for managing data.

owid-catalog

A Pythonic API for working with OWID's data catalog.

Status: experimental, APIs likely to change

Overview

Our World in Data is building a new data catalog, with the goal of our datasets being reproducible and transparent to the general public. That project is our etl, which going forward will contain the recipes for all the datasets we republish.

This library allows you to query our data catalog programmatically, and get back data in the form of Pandas data frames, perfect for data pipelines or Jupyter notebook explorations.

graph TB

etl -->|reads| snapshot[upstream datasets]
etl -->|generates| s3[data catalog]
catalog[owid-catalog-py] -->|queries| s3

We would love feedback on how we can make this library and overall data catalog better. Feel free to send us an email at [e-mail], or start a discussion on Github.

Quickstart

Install with pip install owid-catalog. Then you can get data in two different ways.

Charts catalog

This API attempts to give you exactly the data you in a chart on our site.

from owid.catalog import charts

# get the data for one chart by URL
df = charts.get_data('https://ourworldindata.org/grapher/life-expectancy')

Notice that the last part of the URL is the chart's slug, its identifier, in this case life-expectancy. Using the slug alone also works.

df = charts.get_data('life-expectancy')

Data science API

We also curate much more data than is available on our site. To access that in efficient binary (Feather) format, use our data science API.

This API is designed for use in Jupyter notebooks.

from owid import catalog

# look for Covid-19 data, return a data frame of matches
catalog.find('covid')

# load Covid-19 data from the Our World in Data namespace as a data frame
df = catalog.find('covid', namespace='owid').load()

There many be multiple versions of the same dataset in a catalog, each will have a unique path. To easily load the same dataset again, you should record its path and load it this way:

from owid import catalog

path = 'garden/ihme_gbd/2023-05-15/gbd_mental_health_prevalence_rate/gbd_mental_health_prevalence_rate'

rc = catalog.RemoteCatalog()
df = rc[path]

Development

You need Python 3.10+, uv and make installed. Clone the repo, then you can simply run:

# run all unit tests and CI checks
make test

# watch for changes, then run all checks
make watch

Changelog

• v0.4.0
  • Highlights
    • Support for Python 3.10-3.13 (was 3.11-3.13)
    • Drop support for Python 3.9 (breaking change)
  • Others
    • Deprecate Walden.
    • Dependencies: Change rdata for pyreadr.
    • Support: indicator dimensions.
    • Support: MDIMs.
    • Switched from Poetry to UV package manager.
    • New decorator @keep_metadata to propagate metadata in pandas functions.
  • Fixes: Table.apply, groupby.apply, metadata propagation, type hinting, etc.
• v0.3.11
  • Add support for Python 3.12 in pypackage.toml
• v0.3.10
  • Add experimental chart data API in owid.catalog.charts
• v0.3.9
  • Switch from isort & black & fake8 to ruff
• v0.3.8
  • Pin dataclasses-json==0.5.8 to fix error with python3.9
• v0.3.7
  • Fix bugs.
  • Improve metadata propagation.
  • Improve metadata YAML file handling, to have common definitions.
  • Remove DatasetMeta.origins.
• v0.3.6
  • Fixed tons of bugs
  • processing.py module with pandas-like functions that propagate metadata
  • Support for Dynamic YAML files
  • Support for R2 alongside S3
• v0.3.5
  • Remove catalog.frames; use owid-repack package instead
  • Relax dependency constraints
  • Add optional channel argument to DatasetMeta
  • Stop supporting metadata in Parquet format, load JSON sidecar instead
  • Fix errors when creating new Table columns
• v0.3.4
  • Bump pyarrow dependency to enable Python 3.11 support
• v0.3.3
  • Add more arguments to Table.__init__ that are often used in ETL
  • Add Dataset.update_metadata function for updating metadata from YAML file
  • Python 3.11 support via update of pyarrow dependency
• v0.3.2
  • Fix a bug in Catalog.__getitem__()
  • Replace mypy type checker by pyright
• v0.3.1
  • Sort imports with isort
  • Change black line length to 120
  • Add grapher channel
  • Support path-based indexing into catalogs
• v0.3.0
  • Update OWID_CATALOG_VERSION to 3
  • Support multiple formats per table
  • Support reading and writing parquet files with embedded metadata
  • Optional repack argument when adding tables to dataset
  • Underscore |
  • Get version field from DatasetMeta init
  • Resolve collisions of underscore_table function
  • Convert version to str and load json dimensions
• v0.2.9
  • Allow multiple channels in catalog.find function
• v0.2.8
  • Update OWID_CATALOG_VERSION to 2
• v0.2.7
  • Split datasets into channels (garden, meadow, open_numbers, ...) and make garden default one
  • Add .find_latest method to Catalog
• v0.2.6
  • Add flag is_public for public/private datasets
  • Enforce snake_case for table, dataset and variable short names
  • Add fields published_by and published_at to Source
  • Added a list of supported and unsupported operations on columns
  • Updated pyarrow
• v0.2.5
  • Fix ability to load remote CSV tables
• v0.2.4
  • Update the default catalog URL to use a CDN
• v0.2.3
  • Fix methods for finding and loading data from a LocalCatalog
• v0.2.2
  • Repack frames to compact dtypes on Table.to_feather()
• v0.2.1
  • Fix key typo used in version check
• v0.2.0
  • Copy dataset metadata into tables, to make tables more traceable
  • Add API versioning, and a requirement to update if your version of this library is too old
• v0.1.1
  • Add support for Python 3.8
• v0.1.0
  • Initial release, including searching and fetching data from a remote catalog