How neptune-glitch/air-quality-openenv works
Air Quality Env is an OpenEnv-compatible reinforcement-learning simulation for industrial emissions control. A user can run episodes through the CLI or web service: an action policy reads the current AQI state, chooses an intervention, and the environment applies it, returns an observation and reward, and checks whether the target was reached. Difficulty levels alter starting AQI and target thresholds. The web runner can use an OpenAI-compatible model, with a heuristic fallback. The graph reflects the sampled runtime files and README; no separate storage subsystem is evident.
Main entry points: env.py AQI environment · inference.py CLI episode runner · app.py Web API
Components 10
Agent interaction
- Action policy
inference.py - Web action policy
server/app.py
Environment simulation
- AQI environment
env.py - Difficulty tasks
env.py - State and reward
env.py - Observation and score
env.py
Runtime interfaces
- CLI episode runner
inference.py - Web API
server/app.py
Other components
- User
- OpenAI-compatible API
Connections 11
- User → CLI episode runner: runs episodes
README.md - User → Web API: requests endpoint
README.md - CLI episode runner → Action policy: requests action
inference.py - Action policy → OpenAI-compatible API: requests completion
inference.py - CLI episode runner → AQI environment: resets and steps
inference.py - Web API → Web action policy: requests action
server/app.py - Web action policy → OpenAI-compatible API: requests completion
server/app.py - Web API → AQI environment: resets and steps
server/app.py - AQI environment → Difficulty tasks: selects task
env.py - AQI environment → State and reward: applies action
env.py - State and reward → Observation and score: builds observation
env.py