Description

It's simple, the TransformedTargetRegressor forces you to fit. I'm in a scenario where i want to multiply the outputted value by some number, but i don't want to check my test scores with the transformed targets, so when deploying my model I wrap it in TransformedTargetRegressor.

Yet, that forces me to fit, my transformer and my estimator (which they already are). Does it make sense to allow this use case? It seems weird that the implementation doesn't check if the regressor and transformer themselves are fitted but only checks if they are set

Steps/Code to Reproduce

    target = TransformedTargetRegressor(transformer=multiplier, regressor=fit_regressor)
    # target.predict(..) now fails

    # my workaround
    target.regressor_ = fit_regressor  # otherwise check_is_fitted fails (╯°□°)╯︵ ┻━┻
    target.transformer_ = multiplier  # same as above
    target._training_dim = 2 # column vector for regression

Thanks for your awesome project!

EDIT: had to add the training dim as well