stops is an open dataset of public transport stops in Slovakia — stops, stations and their individual platforms, with coordinates, administrative location (region, district, municipality), transport modes, amenities and, where a match was found, the stop's identifier in the official ISCP registry. It is built by the pipeline in this repository from open sources and published in CSV, GeoJSON, GTFS and SQLite formats as part of the gtfs.sk project.
You don't need to run anything — the dataset is published as GitHub Releases.
- Latest release —
all export files, a combined
stops-<version>.zipbundle andSHA256SUMSwith checksums for every asset. - Releases are monthly snapshots of the source data, versioned with CalVer (see Releases and versioning).
A build produces seven files in data/export/:
| File | Contents |
|---|---|
stops.csv |
Stops only, curated columns (12) |
platforms.csv |
Platforms only, curated columns (14) |
stops-and-platforms.csv |
Stops and platforms in one table, curated columns |
stops.geojson |
Stops and platforms as GeoJSON features |
stops.gtfs.zip |
ZIP containing a GTFS stops.txt |
stops.sqlite |
Dedicated SQLite export: stops, platforms, the ISCP registry tables and the match links |
iscp-register.zip |
The ISCP registry on its own: iscp-stops.csv, iscp-platforms.csv, iscp-stops.geojson, iscp.sqlite |
Use stops.csv for a list of stops, platforms.csv for boarding
locations, stops-and-platforms.csv when you want both as one table, and
stops.sqlite when you need all tables and relationships. The SQLite
export is built with an explicit, versioned schema — it is not a copy of
the pipeline's working database.
Note
stops.gtfs.zip contains only stops.txt — it is not a complete
GTFS Schedule feed (there are no routes, trips or stop times).
The full column-by-column reference for every file is in docs/exports.md.
The dataset has two record types:
- A stop is a stop or station. Whether a record is a station is a property of the record itself (its kind), not a separate flag.
- A platform is an individual boarding location belonging to a stop. A stop can have zero or more platforms.
Identifiers:
id— the project's own stable identifier of a stop or platform row.iscp_id— the identifier of the record in the ISCP registry, where a match was found. It is distinct fromid. Matches belong to stops; platforms are linked through their parent stop.- In every public export, a platform points to its parent stop through
platform_parent_stop_id(the internal database model calls the same relationshipstop_id, and GTFSstops.txtusesparent_station, as the GTFS specification requires).
The ISCP registry itself is stored in its own tables (iscp_stops,
iscp_platforms) and published separately as iscp-register.zip; the
links between the registry and the canonical stops, including match
quality and distance, are in iscp_matches.
See docs/exports.md for the details.
pipeline/crawl → pipeline/clean → pipeline/patch → pipeline/export
(OSM, ISCP) (cluster, dedup) (ISCP matches, (7 files)
names)
- OpenStreetMap — the base source. The pipeline downloads the Slovakia
extract from Geofabrik, filters it down to public transport features
with
osmium, and maps them to stops and platforms. - ISCP registry (iscp.sk) — the official
Slovak register of bus stops. The crawler stores it faithfully in its
own tables (all source fields, every GPS coordinate pair); a later
patch step matches canonical stops against it to attach
iscp_ids, and stop names are normalised using the official registry names. - GKÚ ZBGIS administrative boundaries — used to assign each record its country, region, district and municipality.
After crawling, the pipeline clusters nearby records and deduplicates
platforms (pipeline/clean/), then applies the patches
(pipeline/patch/), and finally writes the export files
(pipeline/export/).
The main exports contain the processed dataset. The ISCP registry is published separately and unchanged (as a normalized register) in
iscp-register.zip.
Prerequisites:
- Python 3.12
osmium-tool— the build shells out to theosmiumcommand-line tool to filter the OSM extract. The Pythonosmiumpackage inrequirements.txtonly provides bindings, not this binary.- macOS:
brew install osmium-tool - Debian/Ubuntu:
sudo apt-get install osmium-tool
- macOS:
git clone https://github.com/gtfs-sk/stops.git
cd stops
python3.12 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python build.pyThe build downloads the OSM extract, the ZBGIS boundaries and the ISCP
registry, runs the whole pipeline, and writes the results to
data/export/ (the working database is data/stops.sqlite). Generated
data lives in data/, which is gitignored — published data only ever
ships in GitHub Releases.
Check dependencies are pinned in requirements-dev.txt:
python3.12 -m venv .venv
.venv/bin/pip install -r requirements.txt -r requirements-dev.txt
.venv/bin/ruff check .
.venv/bin/ruff format --check build.py models pipeline scripts
.venv/bin/pyright
.venv/bin/python -m compileall -q build.py models pipeline scripts
.venv/bin/python scripts/smoke_test.pyThe smoke test runs every exporter against an empty database and
validates the output — no source data is downloaded. The same checks run
in CI on every push to main and every pull request
(.github/workflows/check.yml), and the release workflow runs them again
before building, so a release is never published from a red tree.
Releases are built by .github/workflows/release.yml:
- automatically once a month (on the 1st at 03:00 UTC), or manually
via
workflow_dispatch, - versioned with CalVer: the tag is the snapshot date of the data in
the Europe/Bratislava timezone,
vYYYY.MM.DD(for examplev2026.10.23). A re-run on a day that already has a release gets a numeric suffix (.2,.3, …). There is deliberately no SemVer — the version describes the data snapshot, not the code. - Before anything is published, the exports are validated
(
scripts/validate_exports.py): all seven files must exist, the CSV headers must match the documented schemas, the SQLite export must use the public schema, and row counts must agree across the CSV, GeoJSON, GTFS, SQLite and ISCP register exports.
Release notes list the stop and platform counts, the share of stops with an ISCP id, the number of ISCP registry records, and the commit the release was built from.
├── build.py # orchestrator: crawl → clean → patch → export
├── models/ # SQLAlchemy models (stops, platforms, ISCP registry)
├── pipeline/
│ ├── crawl/ # data collectors (OSM, ISCP registry)
│ ├── clean/ # clustering and platform deduplication
│ ├── patch/ # ISCP matching, stop-name cleanup
│ └── export/ # CSV, GeoJSON, GTFS, SQLite and ISCP exporters
├── scripts/ # smoke test and export validator
├── www/ # static web explorer for the dataset
├── docs/ # exports reference (docs/exports.md)
└── data/ # generated (gitignored): cache, working DB, exports
One file, two parts — see LICENSE:
- Code (Part 1): MIT
- Data (Part 2): Open Database License (ODbL) 1.0
When you use the data, credit the project, for example:
Contains data from gtfs.sk, licensed under ODbL 1.0.
The data itself comes from sources with their own attribution requirements, in particular © OpenStreetMap contributors (ODbL).