Examples¶
A set of worked, end-to-end scripts. The four core stages below are the path you would actually meet them in: audit an unknown feed, collect snapshots into a data lake, analyse the resulting history, then measure network equity. Further pages build on that base (robustness and equity, free-floating and ghost fleets, service reliability, transit and weather context, macro-scale comparison). Each has its own page with a step-by-step walkthrough, the command to run it, and the full source.
Core workflow¶
flowchart LR
A["1. Audit a feed"] -->|trustworthy stations| B["2. Collect a snapshot"]
B -->|repeat every few minutes| B
B --> C["3. Analyse history"]
A --> D["4. Equity and coverage"]
Auditing comes first, because every later step should run on the trustworthy subset of stations. Collection is repeated on a schedule to build a panel; analysis and the equity report then read from what was collected. The remaining how-to pages extend this core path.
| Example | Goal | Extras |
|---|---|---|
| Audit a feed | Inspect an unknown feed and keep the trustworthy stations | [fetch] |
| Collect a snapshot | One cron-driven collection run into a Parquet lake | [fetch], [parquet] |
| Analyse history | Coverage, daily typologies and turnover from a built-up lake | [parquet], [cluster] |
| Equity and coverage | Capacity concentration and spatial equity of a network | [fetch], [geo] |
| Rigorous & comparative analysis | Threshold robustness, bootstrap CIs, FDR hotspots, Theil/Palma, E2SFCA, rebalancing tension | none (bundled data) |
| Free-floating & ghost fleets | Reconcile docked + free-floating, ghost vehicles, spatial-entropy collapse | none (synthetic) |
| Service reliability | Stockout episodes, Kaplan–Meier outage survival, sampling vulnerability | none (synthetic) |
| Contextualize with transit & weather | Transit feeders, leak-free exogenous join, autocorrelation | none (synthetic) |
| Macro-scale comparative audit | Filter the world catalogue and audit N feeds in one call | [fetch] (live) |
The scenarios marked none use the bundled load_example() dataset and small
synthetic frames, so they run with no network and are executed in CI.
Install what a script needs, for example:
The runnable .py files live in the
examples/ directory of the
repository. For instant, network-free reproduction, a
quickstart notebook runs the audit, occupancy and a descriptive
profile on the bundled sample.
Finding a system id¶
Examples 02 and 03 resolve a system by its MobilityData catalogue id (velib,
bike_share_toronto, and so on). Examples 01 and 04 take a gbfs.json URL directly. To find
an id:
import gbfs_toolkit as gb
cat = gb.systems_catalog()
gb.filter_catalog(cat, country_code="FR")[["system_id", "name"]]
Feeds change without notice
GBFS feeds are operator-run. A URL that worked last month may move or break, and the exact fields a feed publishes vary by operator and GBFS version. The audit is designed to make those differences visible rather than to assume them away.