gtfs2nx

Create routable NetworkX graph with realistic transfer times from GTFS feeds. 🚌 🚆 🚡

How

• Cleaning & preprocessing
  • Remove trips with unrealistic travel times and speeds
  • Fix trips contradictory stop sequences and departure times
• Enable routability
  • Ensure each node belongs only to single route
  • Calculate average segment travel times
  • Calculate average stop headway and service frequency
  • Add edges for walking transfer between routes with realistic transfer time (walking time + headway / 2)

Install

pip install gtfs2nx

Usage

import gtfs2nx as gx

G = gx.transit_graph('path/to/GTFS-feed.zip')

Customize transit network:

G = gx.transit_graph(
    gtfs_paths='path/to/GTFS-feed.zip',
    time_window=('06:00', '08:00'),
    route_types=[700], # only buses
    walk_transfer_max_distance=400, # allow transfers with long walking distance
)

Combine several feeds and pick a specific day:

G = gx.transit_graph(
    gtfs_paths=['switzerland.zip', 'germany.zip'],
    feed_ids=['ch', 'de'],      # namespaces IDs so the feeds cannot collide
    date='2026-07-04',          # or one date per feed, if their validity periods differ
    max_headway=4 * 3600,       # keep stops served only once a day
    max_speed_kmph={'default': 130, 2: 400},  # per route type, so high speed rail survives
    min_segment_travel_time=30, # repair segments whose timetable rounds to zero
    max_speed_repair_factor=2,  # slow over-fast segments down instead of dropping their trip
)

Take transfer times from the timetable instead of assuming half a headway, which matters for regional and long-distance services whose connections are deliberately scheduled:

G = gx.transit_graph(
    gtfs_paths='path/to/GTFS-feed.zip',
    transfer_model='timetable',   # look up the next departure rather than guessing
    transfer_wait_quantile=0.1,   # a traveller picks a departure whose connections line up
    min_transfer_time=180,        # platform changes are not instantaneous
    max_transfer_wait=3600,       # safety net for routes that never connect
)

See the API docs for more details and the getting-started notebook for a small demo.

Exemplary NetworkX analysis

import networkx as nx

# access precomputed stop characteristics
frequency = nx.get_node_attributes(G, 'frequency')

# routing
route = nx.shortest_path(G, from_stop, to_stop, weight='weight')

# travel time to other stops
travel_times = nx.single_source_dijkstra_path_length(G, source=from_stop, weight='weight')

# centrality analysis
centrality = nx.closeness_centrality(G, distance='weight')

Further use cases

• Using graph neural networks to classify missing GTFS route types
• Validating NetworkX transit graph: Comparing routes and travel times to Google Maps

Development

Build from source using poetry:

poetry build
pip install dist/gtfs2nx-*.whl

Configure post-checkout hook for branch specific .gitignore files:

git config --local core.hooksPath .githooks/