Cross-City Extension Framework

Goal

Design a framework for extending the benchmark to new cities.

Current Status

Validated: NYC only. All experiments use NYC TLC Yellow Taxi data (2022-2025).

City Configuration

Each city is defined by a YAML config file (see configs/city_template.yaml):

city_name: "nyc"
zones: 263
data_source: "NYC TLC Yellow Taxi"

Required Data for New Cities

  • Trip records with pickup time and location

  • Zone definitions (lat/lon boundaries or zone IDs)

  • (Optional) Weather, calendar, event data for external features

Extension Steps

  1. Create city config in configs/cities/{city_name}.yaml

  2. Implement data adapter for city-specific format

  3. Run feature engineering pipeline

  4. Train forecasting models

  5. Calibrate simulator if applicable

  6. Run benchmark

Limitations

  • Feature engineering assumes NYC-style grid/time features

  • Simulator calibration requires trip-level data with fare and travel time

  • External features (airport, events) are NYC-specific

  • Cross-city transfer learning is NOT yet implemented

Future Work

  • Chicago taxi data integration

  • Ride-hail platform data (Uber/Lyft)

  • International city formats