Karry001/movemelbourne-gtfsr

MoveMelbourne: Melbourne train & tram GTFS-Realtime snapshots

Self-collected, minute-by-minute snapshots of Victoria's GTFS-Realtime feeds for Metro Trains and Yarra Trams, plus every weekly GTFS Schedule release in force during the collection. GTFS-Realtime is live-only (there is no public Melbourne archive), so this dataset preserves what the feeds said, to measure punctuality and reliability after the fact.

Built for the MoveMelbourne project: https://github.com/KarthikReddy-Tanigundala/movemelbourne (method, data model and limitations).

Source and licence

Contains data from the Department of Transport and Planning (Victoria): GTFS Schedule and GTFS-Realtime, published under Creative Commons Attribution 4.0 via the Transport Victoria Open Data Portal. This dataset is a reformatted copy (protobuf flattened to Parquet) and keeps the same licence. It is not an official DTP product; figures derived from it are independent estimates, not official performance results.

Collection

• Feeds: metro and tram × trip-updates, vehicle-positions, service-alerts (https://api.opendata.transport.vic.gov.au/opendata/public-transport/gtfs/realtime/v1/{mode}/{feed}).
• Cadence: each feed is polled every 60 s in 29-minute runs every 30 minutes. A snapshot whose header.timestamp matches the previous one (DTP caches ~30 s) is counted but not stored.
• Collectors: a local bridge collector (from 22 Oct 2026 00:35 AEDT) and GitHub Actions (from 25 Oct 2026). Overlapping snapshots from both are expected; deduplicate on (mode, feed_ts, trip_id, stop_sequence).
• Study window: planned 22 Oct – 19 Nov 2026 (includes Melbourne Cup Day, 3 Nov).
• Gaps: the local collector missed many hours on 22–25 Oct (laptop sleep and power loss). Every run writes a row to raw/coverage, so gaps are measurable rather than hidden.

Layout

raw/realtime/trip_updates/mode={metro|tram}/dt=YYYY-MM-DD/run_<run_id>[_cNNN].parquet
raw/realtime/vehicle_positions/...      raw/realtime/service_alerts/...
raw/coverage/dt=YYYY-MM-DD/run_<run_id>.parquet          one row per run x mode x feed
raw/pb_samples/...                                       first raw protobuf per feed per run (gzip)
raw/schedule/feed_version=<date>_<sha8>/mode_branch={2|3}/<table>.parquet   GTFS Schedule, all strings
raw/schedule_versions/feed_version=<v>.parquet           release metadata (sha256, calendar span)

dt is the Melbourne-local date of the run start; _cNNN are 5-minute checkpoint chunks.

Main columns

Table	Grain	Key columns
trip_updates	stop_time_update × snapshot (one stop-less row for trips with none, e.g. CANCELED)	feed_ts, poll_ts_utc, trip_id, route_id, start_date, trip_schedule_relationship, stop_sequence, stop_id, arrival_time, arrival_delay, departure_time, departure_delay, stu_schedule_relationship, extra_json
vehicle_positions	vehicle × snapshot	feed_ts, vehicle_id, trip_id, start_date, latitude, longitude, bearing, vehicle_ts
service_alerts	alert × informed entity × active period × snapshot	alert_id, active_start, active_end, cause, effect, header_text, description_text, route_id, stop_id, direction_id
coverage	run × mode × feed	run_id, collector, run_start_utc, run_end_utc, polls_attempted, polls_succeeded, polls_duplicate, auth_method, rows_written

Times are POSIX epoch seconds (UTC). Fields DTP adds later land in extra_json.

Feed behaviour worth knowing

Measured on the live feeds (details in the repo's docs/feed_behaviour.md):

• Metro trip updates list all remaining stops and keep a passed stop for 1–4 minutes. Metro times are whole minutes. Tram trip updates list only the current and next one or two stops, to the second.
• Vehicle positions carry no current_status, stop_id or speed; about a quarter of metro positions are more than 3 minutes old.
• The metro trip-updates feed excludes replacement buses, service deviations and platforms.

Loading

import duckdb
from huggingface_hub import snapshot_download

path = snapshot_download("Karry001/movemelbourne-gtfsr", repo_type="dataset",
                         allow_patterns=["raw/realtime/trip_updates/**", "raw/coverage/**"])
duckdb.sql(f"""
    select mode, count(distinct feed_ts) as snapshots, count(*) as rows
    from read_parquet('{path}/raw/realtime/trip_updates/**/*.parquet', union_by_name = true)
    group by mode
""").show()

Citation

Tanigundala, K. R. (2026). MoveMelbourne: Melbourne train & tram GTFS-Realtime snapshots [Data set]. Hugging Face. https://huggingface.co/datasets/Karry001/movemelbourne-gtfsr. Contains data from the Department of Transport and Planning (Victoria), CC BY 4.0.