MEC-DAG PPO Skeleton

This repository is now trimmed into a PPO-centered training skeleton for the future multi-user, multi-server MEC-DAG offloading task.

Project Rule

Before every Python command, training script, or test script, activate the project environment first:

conda activate dl

What Remains

• onpolicy/algorithms/r_mappo/: PPO updater, policy wrapper, actor, critic
• onpolicy/utils/shared_buffer.py: shared rollout buffer and GAE
• onpolicy/utils/separated_buffer.py: separated rollout buffer and GAE
• onpolicy/runner/shared/base_runner.py: shared-policy runner framework
• onpolicy/runner/separated/base_runner.py: separated-policy runner framework
• onpolicy/envs/env_wrappers.py: vectorized environment wrappers

Removed Legacy Content

• MPE wrappers and training scripts
• SMAC/StarCraft wrappers and training scripts
• Hanabi wrappers and evaluation scripts
• Football task-specific wrappers and scripts
• HAPPO, HATRPO, MAT and their helper scripts

MEC-DAG Replacement Points

The following files are the intended integration points for the new MEC-DAG simulator:

• onpolicy/envs/mec_dag/mec_dag_env.py: implement the MEC-DAG environment class here
• onpolicy/runner/shared/mec_dag_runner.py: implement shared-policy rollout logic here
• onpolicy/runner/separated/mec_dag_runner.py: implement separated-policy rollout logic here
• onpolicy/scripts/train/train_mec_dag.py: wire parser, environment creation, and runner selection here

Installation

conda activate dl
pip install -e .