How rikonenobais/deepq-mobileedgecomputing-modified works
This research code trains deep-Q-learning agents to offload mobile-device tasks in a simulated edge-computing environment. A researcher starts either training script; the script creates per-device agents, runs episodes against the environment, and collects reward, cost, delay, and task-drop metrics. The environment models local computation, wireless transmission, fog computation, and queues; agents select actions and learn from transitions. The README identifies the scripts’ roles, while sampled excerpts do not show the full episode interface or how results are ultimately persisted.
Main entry points: train.py Main training · trainPaperParam.py Paper parameters · fog_env.py Offload environment · RL_brain.py DQN agent
Components 10
Training
- Main training
train.py - Paper parameters
trainPaperParam.py - Episode metrics
train.py
Simulation
- Offload environment
fog_env.py - Task queues
fog_env.py - Offload simulation
fog_env.py
Learning
- DQN agent
RL_brain.py - Replay memory
RL_brain.py - Recurrent Q network
RL_brain.py
Other components
- Researcher
Connections 16
- Researcher → Main training: starts training
- Researcher → Paper parameters: starts experiment
- Main training → Offload environment: runs episodes
- Paper parameters → Offload environment: runs episodes
- Main training → DQN agent: creates and learns
- Paper parameters → DQN agent: creates and learns
- Offload environment → Offload simulation: simulates tasks
- Offload simulation → Task queues: queues tasks
- Task queues → Offload simulation: feeds processing
- Offload environment → Main training: returns episode state
- Offload environment → Paper parameters: returns episode state
- Main training → Episode metrics: aggregates outcomes
- Paper parameters → Episode metrics: aggregates outcomes
- DQN agent → Recurrent Q network: selects and updates
- DQN agent → Replay memory: stores transitions
- Replay memory → DQN agent: supplies batches