You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have a general question about how to handle training where some of the covariates are themselves forecasts. I am building a TFT model where the decoder part of each training context window for some covariates would be sourced from a forecast. At each time index that forecast would be different and thus cannot be easily represented by a single time series.
For example:
Target variable is y(t),
Covariate time series x(t). is a forecast in the decoder but a known value in the encoder.
At each time (t) that forecast will be potentially different.
How would I set up a dataset/dataloader to handle this case, and is this even possible in the current implementation?
Thanks in advance for any pointers!
p.s. thank you for all the work everyone does on this package, it's an incredible tool!
The text was updated successfully, but these errors were encountered:
I have a general question about how to handle training where some of the covariates are themselves forecasts. I am building a TFT model where the decoder part of each training context window for some covariates would be sourced from a forecast. At each time index that forecast would be different and thus cannot be easily represented by a single time series.
For example:
Target variable is y(t),
Covariate time series x(t). is a forecast in the decoder but a known value in the encoder.
At each time (t) that forecast will be potentially different.
How would I set up a dataset/dataloader to handle this case, and is this even possible in the current implementation?
Thanks in advance for any pointers!
p.s. thank you for all the work everyone does on this package, it's an incredible tool!
The text was updated successfully, but these errors were encountered: