Skip to content

API Setup

Setup Instructions

Prerequisites

  • Docker & Docker Compose
  • Or locally:
  • Python 3.9+
  • Java 17+
  • Maven 3.9+

Steps to Deploy

  1. Copy your trained model: 

    # Copy trained model to the models directory
# Example: Copy from C:\Users\gemer\Sumo\my-network\Results\sweeps\<sweep_dir>\<seed>\<variant>\model.zip
mkdir -p rl-inference-service/app/trained_models
copy "C:\Users\gemer\Sumo\my-network\Results\sweeps\pressure\seed_42\A\model.zip" "rl-inference-service\app\trained_models\model.zip"

  2. Set environment variables (optional): 

    # Create .env file for RL Inference Service
echo MODEL_PATH=/app/trained_models/model.zip > rl-inference-service/.env
echo OBSERVATION_SHAPE_DIM=10 >> rl-inference-service/.env
echo NUM_AGENTS=1 >> rl-inference-service/.env

  3. Build and start services: 

    docker-compose up --build

  4. Test the API: 

    # Get traffic action (auto-generated observations)
curl http://localhost:8080/api/traffic/action

# Predict action with custom observations
curl -X POST http://localhost:8080/api/traffic/action \
  -H "Content-Type: application/json" \
  -d '{"observations": [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]}'

# Check health
curl http://localhost:8080/api/traffic/health

Option 2: Local Development

Python Service: 

cd rl-inference-service

# Create virtual environment
python -m venv venv
venv\Scripts\activate  # Windows
source venv/bin/activate  # Linux/Mac

# Install dependencies
pip install -r requirements.txt

# Set environment variables
set MODEL_PATH=path\to\your\model.zip  # Windows
export MODEL_PATH=path/to/your/model.zip  # Linux/Mac

# Run service
python -m uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Java Gateway: 

cd java-api-gateway

# Set environment variable
set RL_INFERENCE_SERVICE_URL=http://localhost:8000/predict_action  # Windows
export RL_INFERENCE_SERVICE_URL=http://localhost:8000/predict_action  # Linux/Mac

# Build and run
mvn clean install
mvn spring-boot:run

Includes Swagger/OpenAPI documentation for all traffic control endpoints.