This repo is a benchmark for spatio-temporal graph data under federated (distributed) scenarios. We collect 12 spatio-temporal datasets among different Real-world scenarios.

• Requirements
• Quick Start
• Datasets
• Partition Settings
• Architecture

numpy>=1.17.2
pytorch>=1.3.1
torchvision>=0.4.2
matplotlib>=3.1.1
prettytable>=2.1.0
ujson>=4.0.2

1. To get a federated partition of METR-LA dataset

python generate_fedtask.py --benchmark metr-la_traffic_forecasting

2. To train METR-LA in federated system

python main.py --config config/metr-la_config.yml

3. To get a visualization of the result

python result_analysis_forecasting.py

For METR-LA & PEMS-BAY dataset, we partition dataset to each client hold one node(sensor) data.

TODO 🔨

P.S. This is basically refer to the older version of easyFL: A Lightning Framework for Federated Learning.