Control vs Conditional
Hi!
Could someone please help to clarify the difference between control columns and conditional columns?
There are 2 examples where the same set of columns used in both ways
Channel-Mix Finetuning [Example] - as conditional;
Finetuning and Forecasting with Exogenous/Control Variables [Example] - as control;
Are conditional columns used only for channel mixing?
Can I use both types at the same time (and in which case if yes)?
Just check the docstring in the tsfm library files.
Conditional columns with #mix_channel = use an additional channel in the past (only in the past tense) for the predicted purpose
Control columns with #exog = use an additional channel in the past+future() for the predicted purpose, it is necessary to provide input future values (without this, accuracy will suffer after further training)