Interview Practice Questions

Which ROC-AUC variation is better: 0.1, 0.7, or 0.85?

For heavily imbalanced classes, which to choose: ROC-AUC or PR-AUC?

We have regression task, target from 0 to 100. What will be the output distribution for each algorithm?

What categorical variable encoding methods exist? Pros and cons?

What are the disadvantages of target encoding?

How does target encoding work in CatBoost?

How to build a tree in general, what criteria to use?

What are the advantages and disadvantages of trees?

In terms of bias-variance, how does a tree behave?

What kind of trees are built in boosting and random forest?

Best practice for boosting tuning - what to tune first?

What happens to variance in random forest?

Why does model variance plateau with increasing complexity?

How is gradient used in gradient boosting?

What does the next tree learn in boosting?

How to evaluate feature importance - feature importance vs SHAP?

Why doesn't boosting extrapolate beyond training data?

What do you understand by hyperparameter optimization?

How to prevent data leakage in medical data?

What is model governance and why is it important in production?