What are the main parameters we usually set for DAG tasks, and why should the graph be structured this way (top-level DAG logic)?
Machine Learning / AI
Have you worked with Spark? What is your general understanding of what it is?
When we talk about metrics in class imbalance — which are appropriate and which are not?
Given a string containing letters and digits. Find the maximum number that appears in the string (considering consecutive digits as a single number).
What happens in Python during string concatenation (e.g., current += char)? Why might this approach be bad in this task, and how to optimize code to avoid frequent string concatenation?
How would you solve the same last click search problem without using JOIN?
Have you encountered Random Forest in work or study? What are the main differences between Random Forest and Boosting?
Have you worked with Airflow or Spark? Can you explain what a DAG is in Airflow, and can you give an example from your experience? Have you written DAGs yourself or only used/runned them?
Why does accuracy show an inadequate quality measure when there is class imbalance?
Have you worked with time series? What metrics are used to evaluate the quality of time series forecasting models?
Given a logs table with user events (user_id, item_id, action_type, timestamp). Write an SQL query to find the ID of the last item each user clicked on.
Есть 100 монеток, одна из них фальшивая — с двумя орлами с обеих сторон. Мы выбираем случайную монетку, подбрасываем её, и выпадает орёл. Какова вероятность, что монетка фальшивая? Распишите решение (по формуле Байеса).
If the optimal number of trees for a model is 500, but we accidentally trained both a random forest and boosting with 1000 trees, which model is more likely to overfit and why?
Why can a task get stuck and run very long in Spark?
Have you ever worked with binary classification? Have you encountered class imbalance in such tasks, and what can be done about it?