Cross validation design

Hi @LeoGrin, Regarding the splits, I understand that the data is split depending on the size.

Is there a reference that guided the choice of the folds depending on size ?
As it is not mentionned, I understand the splits are random. Why not use some stratification ?

LeoGrin / tabular-benchmark