Group stratified k-fold cross-validation

Module 7. Evaluating model performance Choice of cross-validation
#1

Hi,
We have learned separately about stratification and grouping in cross-validation. Could you specify what is the meaning of “group stratified k-fold cross-validation” (quiz M7.02) ? It seems it could not be compatible at least in some cases…
Thanks in advance.
Camille

#2

Group stratified K-Fold would be something like: sklearn.model_selection.StratifiedGroupKFold — scikit-learn 1.0.dev0 documentation

You will stratify in each group.

#3

Many thanks for the indication. I note the important sentence in the description : “attempts to create folds which preserve the percentage of samples for each class as much as possible given the constraint of non-overlapping groups”, which answers to my doubts and indicates what is the priority.
Best wishes,
Camille

#4

Yes true, sometimes we cannot be strict.