Bulk history

Long time ranges of the 20-second STIB feeds as Parquet files, rather than one snapshot per call.

Endpoints

Parquet files for a time range

/parquetized

PARQUET

The URLs of the Parquet files holding every snapshot of a component over a time range, oldest first. Each file is a plain Parquet file (snappy), one row per vehicle and poll, with the snapshot's fields plus a date column (UTC); download it directly from its URL. Available for stib_vehicle_trip (endpoint stib/vehicle-trip, with lon/lat) and stib_vehicle_distance (endpoint stib/vehicle-distance). Files cover one UTC day each, and one hour each for the current day. With keys set to {"lineId": "71"}, one file per week (Monday to Monday, UTC) holds a single line; days not yet grouped into a week are returned as daily files of all lines. Files can extend past the requested range: filter on date.

import requests
import pandas as pd

from datetime import datetime, timedelta, timezone

end = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
params = {
    "component": "stib_vehicle_trip",
    "start_timestamp": (end - timedelta(days=2)).isoformat(),   # ISO accepted; epoch also works
    "end_timestamp": end.isoformat(),
}

urls = requests.get("https://api.mobilitytwin.brussels/parquetized", params=params).json()["results"]
print(f"{len(urls)} files")   # one per day here

# Each file is a plain Parquet file: read it straight from its URL.
df = pd.concat([pd.read_parquet(url) for url in urls], ignore_index=True)
df = df[(df["date"] >= params["start_timestamp"][:19]) & (df["date"] < params["end_timestamp"][:19])]
print(f"{len(df):,} vehicle positions, {df['date'].nunique():,} polls")

# Mean delay per line over the two days, for vehicles with a trip
print(df.dropna(subset=["delay"]).groupby("lineId")["delay"].mean().sort_values().tail(10))