Adaptive Routing with Hierarchical Reinforcement Learning on Dragonfly Networks

• hierarchical.ipynb is for Q-hierarchical routing test

• shortest.ipynb is for Minimal and Global routing test

• VAL.ipynb is for VAL routing test

• qroute.ipynb is for Q-routing test

• qadapt.ipynb is for Q-adaptive routing test

• The routing algorithms is stored in the models folder

• Trained models is stored in the dump_dragonfly folder

• dragonfly_generator.py provides useful functions for generating and reading different dragonfly topologies

Model training and testing

To train Q-routing, Q-adaptive, and Q-hierarchical, initialize agent with static=False, i.e.

agent = Qroute(network, static=False)

To test, initialize agent with static=True