Hello,
At the end of the Regularization of linear regression model lecture, after we run RidgeCV to assess the performance of different Ridge models with different values for alpha, this is mentioned:
We can check if the best
alphafound is stable across the cross-validation fold.
What does “stable” mean in this context? And is this “cross-validation” the outer or the inner cross-validation?
stable will refer if the alpha is varying a lot, from non-regularized (e.g. 1e-3) to regularized (1e3). If this is the case, it means that the random change of the train/test dataset induced by the cross-validation will have a huge impact on the model and it is problematic.
It refers to the outer. The outer will change the train/test used and for this specific dataset we set an alpha (using the inner cross-validation).