PyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neural Machine Learning ModelsPyTorch Geometric Temporal: Spatiotemporal Signal Processing with Neural Machine Learning Models - 2104.07788.pdf

Tags

Top Highlights

• Dynamic graph with temporal signal

• Dynamic graph with static signal

• Static graph with temporal signal

• Spatiotemporal Signal Iterators

• These iterators store spatiotemporal dataset

• By iterating over a Spatiotemporal Signal Iterator at each step a graph snapshot is returned which describes the graph of interest at a given point in time. Graph snapshots are returned in temporal order by the iterators.

• Graph Snapshots

• The time period specific snapshots which consist of labels, features, edge indices and weights are stored as NumPy arrays [58] in memory, but returned as a PyTorch Geometric Data object instance [15 ] by the Spatiotemporal Signal Iterators when these are iterated on.

• Train-Test Splitting

• tem- poral train-test splitting function which creates train and test snap- shot iterators from a Spatiotemporal Signal Iterator given a test dataset ratio

• cost = cost / (time+1)

• We normalize the cost

• Chickenpox Hungary

• A spatiotemporal dataset about the officially reported cases of chickenpox in Hungary. The nodes are counties and edges describe direct neighbourhood relationships. The dataset covers the weeks between 2005 and 2015 without missingness.

Toughts & Comments

Laurent B

Paper for publication of torch_geometric_temporal package