The main goal of the library is to make temporal geometric deep learning available for researchers and machine learning practitioners in a unified easy-to-use framework. PyTorch Geometric Temporal was created with foundations on existing libraries in the PyTorch eco-system, streamlined neural network layer definitions, temporal snapshot generators for batching, and integrated benchmark datasets. These features are illustrated with a tutorial-like case study.

Repo:

https://github.com/benedekrozemberczki/pytorch_geometric_temporal

Paper:

https://arxiv.org/abs/2104.07788

Data handling functionalities:

1. Temporal splitters
2. Iterators for discrete time spatiotemporal snapshots

a, Dynamic graph - static signal.

b, Static graph - temporal signal.

c, Dynamic graph - temporal signal.

Models covered from (AAAI, IJCAI, KDD, NeurIPS):

Recurrent temporal aggregation:

MPNN LSTM, GCONVGRU, GCONVLSTM, DCRNN, TGCN, GCLSTM LRGCN etc.

Attention based temporal aggregation:

GMAN, STGCN, ASTGCN, MSTGCN etc.

New datasets:

- Windmill output prediction

- Chickenpox forecasting

- Wikipedia traffic management

- Bicycle deliveries by PedalMe

Existing datasets:

- METR-LA and PEMS-BAY

- Twitter Tennis

- COVID 19 England