As I understand,
sklearn works by doing random splits instead of minimizing a metric like gini for classification or mae for regression.
I don't understand why there's a
criterion parameter, as the criterion for the splits should be random.
Is it just for code compatibility, or am I missing something?