How girayonur/composite-gradient-learning works
This repository contains the paper’s composite-gradient learning experiments for traffic-network control. The main workflow trains an agent against a simulated traffic environment, where learned ramp-meter actions interact with an MPC controller and a METANET traffic model; saved agents are then evaluated and results recorded. The two network directories implement parallel versions for different network sizes. README capabilities also include demand scenarios, gradient diagnostics, and saved evaluation results. Relationships reflect the excerpts, SOURCE INDEX references, and README; unsampled details are not assumed.
Main entry points: src/traffic_env_dmpc.py Network 1 environment · src/traffic_env_dmpc.py Network 2 environment · src/train_ddpg_dmpc.py Network 1 learner · src/train_ddpg_dmpc.py Network 2 learner
Components 26
Experiment workflows
- Network 1 training
network_1/train.py - Network 2 training
network_2/train.py - Network 1 evaluation
network_1/evaluate.py - Network 2 evaluation
network_2/evaluate.py - Network 1 evaluator
network_1/src/evaluate_ddpg_dmpc_noise_seeds.py - Network 2 evaluator
network_2/src/evaluate_ddpg_dmpc_noise_seeds.py
Traffic control
- Network 1 environment
network_1/src/traffic_env_dmpc.py - Network 2 environment
network_2/src/traffic_env_dmpc.py - Network 1 demands
network_1/src/demands.py - Network 2 demands
network_2/src/demands.py - Network 1 MPC
network_1/src/dmpc.py - Network 2 MPC
network_2/src/dmpc.py - Network 1 traffic model
network_1/src/metanet_model.py - Network 2 traffic model
network_2/src/metanet_model.py - Network 1 PI solver
network_1/src/mpc_solve_pi_alinea.py - Network 2 PI solver
network_2/src/mpc_solve_pi_alinea.py - Network 1 parameters
network_1/src/params.py - Network 2 parameters
network_2/src/params.py
Learning and diagnostics
- Network 1 learner
network_1/src/train_ddpg_dmpc.py - Network 2 learner
network_2/src/train_ddpg_dmpc.py - Network 1 gradient metrics
network_1/src/gradient_metrics.py - Network 2 gradient metrics
network_2/src/gradient_metrics.py
Experiment artifacts
- Trained agents
network_1/agents - Evaluation results
network_1/results/paper - Gradient logs
network_1/training/gradients
Other components
- Researcher
Connections 34
- Researcher → Network 1 training: starts training
README.md - Researcher → Network 2 training: starts training
README.md - Researcher → Network 1 evaluation: starts evaluation
README.md - Researcher → Network 2 evaluation: starts evaluation
README.md - Network 1 training → Network 1 learner: runs learner
README.md - Network 2 training → Network 2 learner: runs learner
README.md - Network 1 evaluation → Network 1 evaluator: runs evaluation
README.md - Network 2 evaluation → Network 2 evaluator: runs evaluation
README.md - Network 1 learner → Network 1 environment: steps environment
network_1/src/train_ddpg_dmpc.py - Network 2 learner → Network 2 environment: steps environment
network_2/src/train_ddpg_dmpc.py - Network 1 learner → Network 1 MPC: uses controller
network_1/src/train_ddpg_dmpc.py - Network 2 learner → Network 2 MPC: uses controller
network_2/src/train_ddpg_dmpc.py - Network 1 learner → Network 1 gradient metrics: computes diagnostics
network_1/src/train_ddpg_dmpc.py - Network 2 learner → Network 2 gradient metrics: computes diagnostics
network_2/src/train_ddpg_dmpc.py - Network 1 environment → Network 1 demands: generates demand
network_1/src/traffic_env_dmpc.py - Network 2 environment → Network 2 demands: generates demand
network_2/src/traffic_env_dmpc.py - Network 1 environment → Network 1 MPC: requests control
network_1/src/traffic_env_dmpc.py - Network 2 environment → Network 2 MPC: requests control
network_2/src/traffic_env_dmpc.py - Network 1 environment → Network 1 traffic model: simulates traffic
network_1/src/traffic_env_dmpc.py - Network 2 environment → Network 2 traffic model: simulates traffic
network_2/src/traffic_env_dmpc.py - Network 1 MPC → Network 1 traffic model: optimizes rollout
network_1/src/dmpc.py - Network 2 MPC → Network 2 traffic model: optimizes rollout
network_2/src/dmpc.py - Network 1 MPC → Network 1 PI solver: uses solver
network_1/src/dmpc.py - Network 2 MPC → Network 2 PI solver: uses solver
network_2/src/dmpc.py - Network 1 environment → Network 1 parameters: configures model
network_1/src/traffic_env_dmpc.py - Network 2 environment → Network 2 parameters: configures model
network_2/src/traffic_env_dmpc.py - Network 1 evaluator → Network 1 environment: evaluates policy
network_1/src/evaluate_ddpg_dmpc_noise_seeds.py - Network 2 evaluator → Network 2 environment: evaluates policy
network_2/src/evaluate_ddpg_dmpc_noise_seeds.py - Network 1 learner → Trained agents: saves models
README.md - Network 2 learner → Trained agents: saves models
README.md - Network 1 evaluator → Evaluation results: records metrics
README.md - Network 2 evaluator → Evaluation results: records metrics
README.md - Network 1 gradient metrics → Gradient logs: logs gradients
README.md - Network 2 gradient metrics → Gradient logs: logs gradients
README.md