This is the code accompanying the submission to the Federated Traffic Prediction for 5G and Beyond Challenge of the Euclid team and the corresponding paper entitled "Federated Learning for 5G Base Station Traffic Forecasting" by Vasileios Perifanis, Nikolaos Pavlidis, Remous-Aris Koutsiamanis, Pavlos S. Efraimidis, 2022.
An extension of this work with an in-depth analysis of the energy consumption of the corresponding machine learning models was presented in 2023 Eighth International Conference on Fog and Mobile Edge Computing (FMEC) with the paper entitled "Towards Energy-Aware Federated Traffic Prediction for Cellular Networks " by Vasileios Perifanis, Nikolaos Pavlidis, Selim F. Yilmaz, Francesc Wilhelmi, Elia Guerra, Marco Miozzo, Pavlos S. Efraimidis, Paolo Dini, Remous-Aris Koutsiamanis.
This code can serve as benchmark for federated time-series forecasting. We focus on raw LTE data and train a global federated model using the measurements of three different base stations on different time intervals. Specifically, we implement 6 different model architectures (MLP, RNN, LSTM, GRU, CNN, Dual-Attention LSTM Autoencoder) and 9 different federated aggregation algorithms (SimpleAvg, MedianAvg, FedAvg, FedProx, FedAvgM, FedNova, FedAdagrad, FedYogi, FedAdam) on a non-iid setting with distribution, quantity and temporal skew.
We recommend using a conda environment with Python 3.8
- First install PyTorch
$ conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
- Install additional dependencies
$ pip install pandas scikit_learn matplotlib seaborn colorcet scipy h5py carbontracker notebook
You can also use the requirements' specification:
$ pip install -r requirements.txt
.
├── dataset # .csv files
├──── ...
├── ml # Machine learning-specific scipts
├──── fl # Federated learning utilities
├────── client # Client representation
├─────── ...
├────── history # Keeps track of local and global training history
├─────── ...
├────── server # Server Implementation
├─────── client_manager.py # Client manager abstract representation and implementation
├─────── client_proxy.py # Client abstract representation on the server side
├─────── server.py # Implements the training logic of federated learning
├─────── aggregation # Implements the aggregation function
├───────── ...
├─────── defaults.py # Default methods for client creation and weighted metrics
├─────── client_proxy.py # PyTorch client proxy implementation
├─────── torch_client.py # PyTorch client implementation
├──── models # PyTorch models
├───── ...
├──── utils # Utilities which are common in Centralized and FL settings
├────── data_utils.py # Data pre-processing
├────── helpers.py # Training helper functions
├────── train_utils.py # Training pipeline
├── notebooks # Tools and utilities
└── README.md
Refer to notebooks for usage examples.
For an extensive overview of the data collection and processing procedure please refer to datataset.