GNN_Spatial

The repository is the PyTorch implamentation of Graph Convolutional Networks and Graph Attention Networks for spatial information extraction from time-series flow data (PeMS04, METR_LA).

Run the demo

Jamp to /config/ and edit the corresponding .conf as needed.

• nums_of_nodes: number of the nodes on the road network
• len: length of time windows for prediction
• time_interval: time period of each time window
• batch_size: number of time slots per batch
• learning_rate
• epochs
• hidden_dim: the hidden dimensions of the middle layers
• training_days: length of training data(days)
• validation_days: length of validation data(days)
• test_days: length of test data(days)

And then to run the demo

cd module python trainer.py

Models

You can choose between the following models from baseline.py:

• GCN3: Three-layered GCN
• GCN2: Two-layered GCN
• bGCN: Graph convolutional network (Thomas N. Kipf, Max Welling, Semi-Supervised Classification with Graph Convolutional Networks, 2016)
• ChebConv: One-layer Chebyshev polynomial version of graph convolutional network as described in (Michaël Defferrard, Xavier Bresson, Pierre Vandergheynst, Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering, NIPS 2016)
• ChebNet: Two-layered Chebyshev polynomial version of graph convolutional network
• GraphAttentionLayer: One-layer graph attention layer with self-attention mechanism
• GATSubNet: One-layer graph attention layer with multi-attention mechanism (the time-series input is splated into flat one-dimensional array as [B, N, T * C])
• GATNet2: One-layer graph attention layer with multi-attention mechanism (apply attention mechanism on T*[B, N, C] tensor, the input should be [B, N, C])