Hi, I'm trying to understand how to implement a model that uses TGCN to predict n future timesteps utilizing an input of n timesteps.
From what I read in the source code, TGCN is a cell that can only process a single timestep. Is this true?
If I want to process n input timesteps, do I have to create a for loop to process the inputs sequentially by the TGCN cell?
As an analogy to a Seq2Seq traditional LSTM model, if I would like to build a seq2seq TGCN model, must I first create a TGCN layer that processes the inputs sequentially with a TGCN cell inside?
In the ChickenPox TGCN example, this is the output of next(iter(dataset)):
Does this mean that x=[20,4] is a tensor with four sequential weekly count values for each of the 20 counties? The sequences' timesteps are encoded as node-features?
Hi, I'm trying to understand how to implement a model that uses TGCN to predict n future timesteps utilizing an input of n timesteps.
From what I read in the source code, TGCN is a cell that can only process a single timestep. Is this true?
If I want to process n input timesteps, do I have to create a for loop to process the inputs sequentially by the TGCN cell?
As an analogy to a Seq2Seq traditional LSTM model, if I would like to build a seq2seq TGCN model, must I first create a TGCN layer that processes the inputs sequentially with a TGCN cell inside?
In the ChickenPox TGCN example, this is the output of next(iter(dataset)):
Data(x=[20, 4], edge_index=[2, 102], edge_attr=[102], y=[20])
Does this mean that x=[20,4] is a tensor with four sequential weekly count values for each of the 20 counties? The sequences' timesteps are encoded as node-features?
I would appreciate your help. Many thanks.