I’m trying to use the TemportalFusionTransformer model in
TemporalFusionTransformer — pytorch-forecasting documentation
Looking at the original paper architecture there is the parameters $\tau_{max}$, which specifies the number of timesteps forward to look at known future inputs, I can’t find this paramter in the pytorch formulation, my best guess is that it just sets it equal to max_prediction_length. Having run some toy examples it looks to default to 0, any direction much appreciated?
