Classification task with strong class imbalance: which metrics are good, which are bad?
Machine Learning / AI
In what order are SQL operators executed in a query?
What metrics did you use to evaluate system quality and how did you achieve a 40% improvement?
What is the advantage of ROC-AUC compared to F1 and Precision?
What is the difference between the apply and map methods in Pandas?
Is a list a mutable data type in Python?
Pre-Layer Norm vs Post-Layer Norm — what is the difference and why is it important for training stability?
How was validation conducted?
Where and what needs to be stored to ensure fault tolerance in auto-call rejection tasks?
How to extract text from scans?
Tell us about Batch Normalization: what it is and why it is needed?
How can you achieve a strictly structured response model, for example in JSON?
What metrics were used to measure retrievability (which dataset and tool)?
What did you most enjoy: building models or working with data?
Rate your English separately for reading/writing and speaking according to CEFR levels.
How to train a model? Seq-to-seq is expensive, and there are many hyperparameters. How to do it?
Write a short paragraph for self-presentation, highlighting that recommendation systems are very similar to RAG - retrieval, reranking, architectural similarities, and so on, so I can remember and explain.
How do you consider the situation when a funny answer has many likes but is not correct?
Is the total experience of 3 years 2 months documented?
Did I understand correctly: we have a ranker with an average score from the content ranker based on content filling, with user and block features, and we took positive — what exactly?