Overall, there were three parts: SQL, Python, and experimentation.
SQL: the main focus
They provide several table schemas. The first question starts with basic aggregation. You just need to use count(distinct ...) flexibly together with sensible filtering, so the key is to understand the table schemas carefully first. You can ask more questions.
The second question generally tests a simple metric definition. The interviewer gives you a fairly vague metric, and you need to define it according to your own understanding. Then there are joins, where using LEFT JOIN and RIGHT JOIN accurately is important. A friendly reminder: when a prompt asks you to rank something, do not rush to use a ranking function. Sometimes LIMIT is enough.
The third question continues from the second and adds a new condition. It basically comes down to whether you can write multiple CTEs and join them sensibly. It is very simple.
Python: the second main focus
They test how you write the simplest function. For example: create a function that takes two integers and reports the larger one without using the built-in max function.
Follow-up: use a list of integers instead of two integers.
Follow-up: instead of returning the maximum integer, return the list in ranked order.
Experimentation
Power analysis is always tested. What do you need in order to perform a power analysis, and after getting the result, how do you determine how long the experiment needs to run?
What is statistical significance? What would you recommend if your experiment is not statistically significant, but you believe its results are still valuable?
Discussion
Loading comments…