How umnilab/rl_pricing works
RL_pricing is a research simulator for reinforcement-learning-based ride-hailing prices. The documented workflow trains a model and then tests it; in the sampled code, main.py configures an Environment and Platform, while runner.py performs simulation steps and updates or evaluates the pricing controller. The main runtime stages are demand response, vehicle/passenger simulation, pricing control, and policy learning. The repository also includes static and dynamic small-grid input datasets. Some main.py loading and dispatch sections are omitted, so dataset wiring and several runtime calls are not fully evidenced here.
Main entry points: center.py Pricing platform · main.py Experiment setup · runner.py Train and test · sim.py Simulation environment
Components 13
Experiment entry
- Experiment setup
main.py - Train and test
runner.py
Ride-hailing simulation
- Simulation environment
sim.py - Vehicle behavior
sim.py
Pricing and control
- Pricing platform
center.py - Vehicle matching
center.py - Baseline pricing
center.py
Reinforcement learning
- Experience memory
center.py - RL pricing algorithms
pricer.py - Policy networks
model.py
Scenario data
- Dynamic grid inputs
grid_small_dynamic - Static grid inputs
grid_small_static
Other components
- Experiment operator
Connections 13
- Experiment operator → Experiment setup: starts experiment
README.md - Experiment setup → Simulation environment: initializes environment
main.py - Experiment setup → Pricing platform: creates controller
main.py - Experiment setup → Train and test: dispatches training
main.py - Train and test → Simulation environment: steps simulation
runner.py - Train and test → Pricing platform: updates prices
runner.py - Train and test → Pricing platform: trains policy
runner.py - Simulation environment → Vehicle behavior: initializes vehicles
sim.py - Pricing platform → Vehicle matching: matches vehicles
center.py - Pricing platform → RL pricing algorithms: selects prices
center.py - Pricing platform → Baseline pricing: provides alternatives
center.py - Pricing platform → Experience memory: stores experience
center.py - RL pricing algorithms → Policy networks: uses networks
pricer.py