Can Convolutional Layers Deal With Batch Trafffic Data?

[ZLXJC]

I use the PEMS-04 Dataset in traffic prediction mission,and the dataset is divided into train and test. The dimension of train is [Batch,Node_number,Time_step,Feature].So,my question is if the cnvolutional layers in PyG are able to accept this kind of data and to predic the future time_step.

[rusty1s]

In general, the format of [batch_size, #nodes, #time, #features] is not understandable by PyG. PyG relies on two-dimensional matrices as node features. As such, it is best to merge both the batch dimension and the time dimension into the node feature dimension ([batch_size * #nodes * #time, #features]). This way, model weights are shared across all nodes in all examples and timestamps. After GNN processing, you can reshape the output node embeddings to your desired size.