How to specify exogenous variables for TCN or Torch about darts

[hrzn]

There is also another way, which consists of shifting your time series of known future values, and bundle it with your other features as a multi-variate time series. For instance, say that you know the weather temperature 7 days in advance, you can create (up to) 7 new time series, which for time step t would represent the temperature at t+1, t+2, ... t+7.
Such a new time series at time step t should be interpreted as "the estimate known at time t of the temperature at time t+1".

[pennfranc]

Unfortunately this functionality is not supported at the moment. What you can do is use the StandardRegressionModel class to try to predict the target time series train_target in combination with multiple model forecasts and exogenous variables as train_features.
Adding support for exogenous variables for forecasting is definitely on our radar, but not available right now.