[Exogenous tansformation] Lagged exogenous - target features

[ncooder]

Description

Is there a way to create a feature from the exogenous and the target value with a lag and without a data leakage? Also in the AutoML can one add a list of exogenous transformations that can be passed for optuna to optimize which one gives better accuracy (also the original exogenous count)?

Use case

This would be very appreciated in AutoML models where the exogenous transformation is not simple to add.

[jmoralez]

Hey. The target features should be created with the arguments of the MLForecast constructor, those won't have any leakage. For the exogenous I'm not sure what you mean, if they're all known then their lags will be known and there won't be leakage. To tune the exogenous you could generate all the features that you want to try and just select a subset in the optimization.