Question regarding the use of projected exogenous variable values

[azylbertal]

I would like to provide some exogenous variable values as part of the target sequence (a known 'forced' future sequence that modulate the other variables). Does the model support it? Could it be achieved by simply editing dec_y = torch.zeros_like(y_t).to(self.device) (in forward_model_pass) such that these variables will remain nonzero?

Thanks!