Roblox·Data Scientist·Technical Phone Screen·Senior

Senior

Jun 2026

Summary

Roblox data scientist interview with a coding/technical problem that looked like a stats exercise but turned into a full ML pipeline question. Pretty involved for a single prompt.

Questions Asked (1)

You're given a labeled binary classification dataset with numeric features. Walk through how you'd z-score normalize the features, fit a logistic regression, then rank and return the top 3 features by coefficient magnitude. Also explain any modeling choices you'd need to make explicit, like how you handle regularization, the intercept, or zero-variance columns.

Algorithms & Data StructuresTechnical Trade-offsData Modeling

Author's notes

More involved than I expected.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Walk through the pipeline step-by-step: first handle data quality issues like zero-variance columns, then z-score normalize features, fit logistic regression with explicit choices about regularization and intercept, and finally rank features by absolute coefficient magnitude. Emphasize that normalization is crucial for comparing coefficients and that regularization affects feature ranking.

Pro tip: Mention that you would use a Pipeline to avoid data leakage and that you'd consider standardizing after splitting data, not before. Also, note that if features are on different scales, coefficients are not comparable without normalization.

1. Data Preprocessing

Identify and handle zero-variance columns (drop them) and check for missing values. Split data into train and test sets before normalization to prevent leakage.

2. Z-score Normalization

Compute mean and standard deviation for each feature on the training set, then transform both train and test sets using these statistics. This ensures features are on the same scale.

3. Fit Logistic Regression

Choose regularization (L1 or L2) and its strength (C), and decide whether to fit an intercept. Fit the model on the normalized training data.

4. Rank Features

Extract model coefficients, take absolute values, and sort in descending order. Return the top 3 feature names.

Key Points to Mention

Zero-variance columns provide no information and can cause issues in normalization (division by zero), so they should be removed.
Z-score normalization (standardization) is essential for comparing coefficients when features have different units or scales.
Regularization (L1/L2) shrinks coefficients and can affect feature ranking; L1 can zero out coefficients, performing feature selection.
The intercept term is not penalized in most implementations, but its inclusion affects the model; it should be explicitly decided based on whether features are centered.
Use a Pipeline to chain preprocessing and modeling, ensuring proper cross-validation and avoiding data leakage.
Coefficient magnitude indicates feature importance only if features are on the same scale; after normalization, larger absolute coefficients mean stronger impact on the log-odds.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.