Significance of choosing ShuffleSplit here?

M4. Linear Models M4. Regularization of linear regression model
#1

I feel I am missing why we switch from the Kfold cross-validation to the ShuffleSplit method here when doing the nested cross validation to find an ideal alpha range.

Wouldn’t the switch to ShuffleSplit have made more sense if we wanted to increase the amount of folds? Here we reduced them from 10 in the previous CV runs to 5.

My apologies if I am missing something obvious here.

#2

I think this is something really discussed because this is tricky.
If I recall well, there is a kind of structure in the order of data. Running a KFold without shuffling induce some bad score for some of the fold. I assume that we switch to ShuffleSplit to break this structure. We could have also used shuffle=True in the KFold constructor.