ML
ML Interview Roadmap
← Study plan·Traditional ML

Day 35 of 133

Trad-ML consolidation + week-5 wrap

Wrap classical ML; record yourself answering the 5 hardest questions.

DSA · NeetCode Heap / Priority Queue

  • Design TwitterDSA · Heap / Priority Queue

    Interview questions to prep

    1. Why is a heap the right structure? Could a balanced BST or sorted list work — why is heap better?
    2. Explain the heap-of-k pattern: keep size k, push new, pop if over k. What's the resulting complexity?
    3. What does the comparator look like, and how would you tweak it to flip min/max behaviour?

ML · Regularization (L1, L2, ElasticNet)

  • L1 vs L2 vs ElasticNet — geometry and selectionTraditional MLStatQuest

    Interview questions to prep

    1. Why does L1 produce sparse solutions while L2 doesn't? Show the geometric picture.
    2. When do you use ElasticNet over L1 or L2 alone?
  • Early stopping & dropout as regularizersTraditional MLRead

    Interview questions to prep

    1. Why is early stopping equivalent to L2 regularization in some cases?
    2. How would you choose the early-stopping patience and what happens when it's too small or too large?
  • Regularization as a Bayesian priorTraditional MLRead

    Interview questions to prep

    1. Show that L2 regularization corresponds to a Gaussian prior on weights and L1 to a Laplace prior.
    2. When does the Bayesian framing actually change a modeling decision in practice?

ML · Feature engineering

  • Scaling, log/Box-Cox, binningTraditional MLscikit-learn

    Interview questions to prep

    1. When do tree models NOT need feature scaling, and when do they (gradient boosting libraries with regularization)?
    2. When would you apply a log transform vs Box-Cox?
  • One-hot, target, frequency, hashing encodersTraditional MLcategory-encoders

    Interview questions to prep

    1. Compare one-hot, target, frequency, and hashing encoders — trade-offs in cardinality and leakage.
    2. Why is target encoding leak-prone and how does k-fold target encoding fix it?
  • Missing data: deletion, imputation, indicatorsTraditional MLscikit-learn

    Interview questions to prep

    1. When is mean/median imputation harmful?
    2. Why do tree models often handle missing values natively while linear models cannot?
  • Outliers, multicollinearity, and distribution shiftsTraditional MLscikit-learn

    Interview questions to prep

    1. How do you detect and handle outliers using box plots, robust scaling, winsorization, or model choice?
    2. Why does multicollinearity destabilize linear models, and why are tree models less sensitive?
    3. How would you distinguish true data drift from a one-off outlier spike in production?

ML · Semi-supervised and proxy labels

  • Semi-supervised learning: pseudo-labels, weak labels, proxy labelsTraditional MLLilian Weng

    Interview questions to prep

    1. When would you use pseudo-labeling, weak supervision, or proxy labels instead of waiting for fully supervised labels?
    2. How do proxy labels introduce bias, and how would you validate that they predict the true product target?
    3. What safeguards prevent a semi-supervised training loop from reinforcing its own mistakes?

References & further reading