Skip to content

linkpark/metarl-offloading

Repository files navigation

System environment

Unbuntu 16.04

Preinstall Package

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev

Create Conda Environment

conda create --name tf-1.15 anaconda python=3.6
conda activate tf-1.15

Install Tensorflow-1.15 GPU or CPU

pip install tensorflow-gpu==1.15

or

pip install tensorflow==1.15

Install Third-party Python Pakage

pip install gym
pip install graphviz
pip install pydotplus
pip install pyprind
pip install mpi4py

Start Meta Training:

python meta_trainer.py

All the hyperparameters are defined in meta_trainer.py including the log file and save path of the trained model.

Start Meta Evaluation:

After training, you will get the meta model. In order to fast adapt the meta model for new learning tasks in MEC, we need to conduct fine-tuning steps for the trained meta moodel.

python meta_evaluator.py

The training might take long time because of the large training set. All the training results and evaluation results can be found in the log file.

Related paper: Fast Adaptive Task Offloading in Edge Computing based on Meta Reinforcement Learning

If you like this research, please cite this paper:

@article{wang2020fast,
  title={Fast Adaptive Task Offloading in Edge Computing Based on Meta Reinforcement Learning},
  author={Wang, Jin and Hu, Jia and Min, Geyong and Zomaya, Albert Y and Georgalas, Nektarios},
  journal={IEEE Transactions on Parallel and Distributed Systems},
  volume={32},
  number={1},
  pages={242--253},
  year={2020},
  publisher={IEEE}
}

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages