pyworldatlas 0.3.1

Offline country profiles, coordinate tools, and geography learning utilities

PyWorldAtlas

A compact, source-aware world atlas for Python that works completely offline.

PyWorldAtlas makes real geographic data feel like ordinary Python. Look up a country by name or code, inspect immutable country and capital objects, explore major cities, calculate geographic relationships, and build reproducible learning material—without an API key, runtime download, database server, or third-party dependency.

from pyworldatlas import Atlas

with Atlas() as atlas:
    japan = atlas.country("Japan")

    print(japan.capital.name)                  # Tokyo
    print(japan.capital.coordinates.as_tuple())
    print(atlas["DO"].name)                    # Dominican Republic
    print("France" in atlas)                   # True

Dataset coverage

The bundled dataset contains every country and area in the captured UN M49 scope, cross-checked against GeoNames country metadata. Version 0.3.0 added a reviewed land-border graph to the profile, coordinate, capital, and populated-place records established in earlier releases. Version 0.3.1 makes that graph easier to query, teach, and explain.

Current dataset	Coverage
Countries and areas	248
Primary capitals	241 / 248
Capital coordinates	241 / 241
Populated-place records	6,265, including retained capitals
Reviewed land borders	319 undirected relationships
Countries and areas without an accepted land border	85
Runtime dependencies	0
Bundled databases	1 SQLite file

The 0.3.1 checkout includes richer country profiles, dependency-free coordinate calculations, flag emoji, discovery cards, reproducible sampling, structured flashcards, reviewed neighbors, and shortest land-border paths. Boundary geometry, historical statistics, national leaders, interactive learning applications, and exports remain later work.

Installation

Install the latest published release:

python -m pip install --upgrade pyworldatlas

Install the current source checkout and its separate data builder when contributing:

python -m pip install -e . -e pipeline

You can also test the exact local wheel after running the release build:

python -m pip install --no-index --no-deps dist/pyworldatlas-0.3.1-py3-none-any.whl

The package runtime supports Python 3.10 through 3.14 during the 0.x release series. Python versions are only claimed as release-supported after CI passes.

What works in this checkout

Capability	Example
Exact lookup	atlas.country("Japan")
Standard identifiers	atlas.country("JP"), atlas.country("JPN"), atlas.country("392")
Familiar aliases	atlas.country("USA"), atlas.country("Holy See")
Collection behavior	atlas["DO"], "France" in atlas, len(atlas)
Ranked search	atlas.search_countries("united")
Geographic filtering	atlas.countries(continent="Americas")
Capital records	country.capital, .coordinates, .timezone_id
Major cities	atlas.major_cities("Japan", limit=5)
Rich profile	country.population, .currency, .languages, .calling_codes
Flags and calculated facts	country.flag_emoji, .population_density
Discovery cards	country.discovery_card()
Stable country samples	atlas.sample_countries(count=5, seed=42)
Structured flashcards	atlas.flashcards(topic="capitals", count=10, seed=42)
City coordinates	atlas.coordinates("Tokyo", country="JP")
Distance	atlas.distance_between("Tokyo", "Paris", first_country="JP", second_country="FR")
Bearing and midpoint	coordinate.bearing_to(other), .midpoint_to(other)
Land neighbors	atlas.neighbors("France"), atlas.shares_border("ES", "MA")
Border paths	atlas.border_path("Portugal", "China"), atlas.border_crossings(...)
Land connectivity	atlas.has_land_route("Portugal", "China")
Land components	atlas.countries_reachable_by_land("Portugal")
Borderless entities	atlas.countries_with_no_land_borders()
Source inspection	country.sources
Official local names	country.local_names, country.name_in("pt")
Serialization	country.to_dict(), country.to_json()
Version inspection	atlas.dataset_info()

Typed country profiles

Public results are frozen typed dataclasses rather than loosely structured dictionaries:

from pyworldatlas import Atlas

with Atlas() as atlas:
    country = atlas.country("Dominican Republic")

    print(country.name)
    print(country.official_name)
    print(country.flag)
    print(country.flag_emoji)
    print(country.codes.alpha2)
    print(country.codes.alpha3)
    print(country.codes.numeric)
    print(country.continent)
    print(country.region)
    print(country.subregion)
    print(country.area_km2)
    print(country.population)
    print(country.population_density)
    print(country.currency)
    print(country.languages)
    print(country.calling_codes)
    print(country.top_level_domain)
    print(country.observed_timezones)

    if country.capital is not None:
        print(country.capital.name)
        print(country.capital.coordinates.as_tuple())
        print(country.capital.population)
        print(country.capital.timezone_id)

Country discovery and education

from pyworldatlas import Atlas

with Atlas() as atlas:
    japan = atlas.country("Japan")
    card = japan.discovery_card()

    print(card.flag_emoji, card.capital, card.population_density)

    for country in atlas.sample_countries(count=5, continent="Africa", seed=42):
        print(country.flag_emoji, country.name)

    for flashcard in atlas.flashcards(topic="capitals", count=3, seed=42):
        print(flashcard.prompt)
        print(flashcard.answer)

Sampling uses a versioned SHA-256 ranking over stable M49 identifiers, so the same dataset, filters, and seed produce the same ordered lesson across supported Python versions. Flashcards are immutable structured values rather than an interactive game. Supported topics cover capitals, flags, country codes, currencies, calling codes, domains, language codes, regions, local names, population, area, calculated density, reviewed neighbors, and land-border counts.

Latitude, longitude, and distance

from pyworldatlas import Atlas, Coordinate

with Atlas() as atlas:
    tokyo = atlas.city("Tokyo", country="Japan")
    paris = atlas.city("Paris", country="France")

    print(tokyo.coordinates.latitude, tokyo.coordinates.longitude)
    print(atlas.distance_between(tokyo, paris))             # kilometres
    print(atlas.distance_between(tokyo, paris, unit="mi"))  # miles
    print(tokyo.coordinates.bearing_to(paris.coordinates))
    print(tokyo.coordinates.midpoint_to(paris.coordinates))

london = Coordinate(51.5074, -0.1278)
paris_center = Coordinate(48.8566, 2.3522)
print(london.distance_to(paris_center))

Distances use the haversine formula and WGS84 mean Earth radius. They are surface great-circle distances, not road or flight-routing distances.

Search and filter

with Atlas() as atlas:
    for match in atlas.search_countries("united"):
        print(match.country.name, match.matched_name, match.score)

    for country in atlas.countries(continent="Europe"):
        capital = country.capital.name if country.capital else "not available"
        print(country.name, capital)

Search is case- and accent-insensitive. Exact country lookup accepts common names, reviewed aliases, alpha-2, alpha-3, and M49 numeric codes.

Land borders and shortest paths

The land-border graph introduced in 0.3.0 contains 319 reviewed undirected relationships. Neighbor results are alphabetical, and equal-length shortest paths are deterministic.

from pyworldatlas import Atlas

with Atlas() as atlas:
    print([country.name for country in atlas.neighbors("France")])
    print(atlas.shares_border("Spain", "Morocco"))

    path = atlas.border_path("Portugal", "China")
    print(path.crossings)
    print(" -> ".join(path.names))
    print(path.alpha2_codes)

    print(atlas.border_path("Japan", "China"))  # None
    print(atlas.has_land_route("Portugal", "China"))  # True

border_path() uses breadth-first search and returns an immutable, JSON-serializable BorderPathResult. A missing route is None, not an error. Maritime proximity, border geometry, border length, and road routing are not represented.

The structured flashcard API also supports neighbors and border_counts. Neighbor answers come directly from the reviewed graph; border counts are the number of accepted edges attached to the selected country or area.

Small by design

The installed wheel contains only:

• Python source files.
• One generated, read-only SQLite database.
• Standard package metadata.

At runtime PyWorldAtlas does not:

• Contact the internet.
• Require an API key.
• Download or decompress data after installation.
• Write into site-packages.
• Load the complete database during Atlas() initialization.
• Depend on pandas, NumPy, an ORM, a GIS engine, or SQLite extensions.

Data you can trace

The 0.3.1 checkout uses:

• United Nations M49 for canonical identities, standard codes, regions, and subregions.
• GeoNames for capitals, populated places, WGS84 coordinates, population snapshots, currencies, language and calling codes, country-code domains, timezone identifiers, GeoNames IDs, and one input to border review.
• Natural Earth public-domain 1:50m map units as an independent land-border topology check.

The reviewed local-name records use the UNGEGN List of Country Names (E/CONF.105/13/CRP.13) for national official short and formal names. Current coverage is five records across Brazil and Switzerland. The captured source artifact and reviewed rows include checksums and exact entry/page locators.

Raw snapshots are preserved with SHA-256 manifests. The separate builder emits inspectable normalized JSON Lines before generating SQLite. Missing values stay missing; unsourced assumptions are never substituted for country facts.

Flag emoji are derived from alpha-2 codes, population density is a transparent ratio of sourced values, and discovery/learning tools only rearrange existing profile data. They introduce no additional country claims or third-party data.

The two border inputs agree on 315 relationships. Six differences have explicit decisions in build_data/reviewed/border_decisions.csv: four relationships are included and two are excluded. Any unreviewed source difference fails the data build.

Seven areas have no usable primary-capital record in the current snapshot. Their country.capital value is None. GeoNames-only country rows that do not have a matching identity in the captured UN M49 scope are excluded rather than inferred.

See DATA_SOURCES.md, DATA_QUALITY.md, and THIRD_PARTY_NOTICES.md.

Three different versions

with Atlas() as atlas:
    print(atlas.dataset_info())

• Library version describes Python behavior and the public API.
• Schema version describes compatibility with the bundled SQLite structure.
• Dataset version identifies the captured source snapshot.

For this development checkout they are 0.3.1, 3, and 2026.07.21.1.

Documentation and roadmap

• Documentation source for this checkout: docs/source
• Published documentation (updated by the release workflow): https://jcari-dev.github.io/pyworldatlas-documentation/
• Current implementation status: ROADMAP_STATUS.md
• Milestone evidence: MILESTONE_0_1_REPORT.md
• 0.2.1 execution status: RELEASE_0_2_STATUS.md
• 0.3.0 release status: RELEASE_0_3_STATUS.md
• 0.3.1 release status: RELEASE_0_3_1_STATUS.md
• Maintainer release process: RELEASING.md

Version 0.3.1 polishes the reviewed land-border API and its learning tools. Later releases extend boundary geometry, historical statistics, institutions, culture, and exports.

License and attribution

PyWorldAtlas code is available under the MIT License. GeoNames data is provided under CC BY 4.0, and Natural Earth data is public domain. Other source terms and notices are recorded in THIRD_PARTY_NOTICES.md.