Normalize features and rank logistic coefficients
Company: Roblox
Role: Data Scientist
Category: Machine Learning
Difficulty: hard
Interview Round: Online Assessment
You are given a binary classification training dataset:
- `X`: a 2D array of shape (n_samples, n_features) containing numeric features.
- `feature_names`: a list of length n_features with the feature names.
- `y`: a 1D binary array (0/1) of length n_samples.
Task:
1) **Normalize** each feature column of `X` using z-score standardization based on the training set:
\[
X'_{:,j} = \frac{X_{:,j} - \mu_j}{\sigma_j}
\]
(where \(\mu_j\) and \(\sigma_j\) are the mean and standard deviation of feature \(j\) over the training data.)
2) Fit a (linear) **logistic regression** model on the normalized features.
3) Extract the fitted coefficients and **rank features by coefficient value** (largest to smallest). Return the **top 3 feature names**.
Clarify any modeling choices needed to make this well-defined (e.g., intercept, regularization, handling zero-variance features).
Overview: This question evaluates understanding of feature preprocessing (z‑score normalization), linear logistic regression fitting and coefficient-based feature ranking, along with practical considerations such as handling zero-variance features, intercept treatment, and regularization; it is commonly asked to test applied skills in producing comparable feature scales and interpretable model weights. Category/domain: Machine Learning; abstraction level: applied, implementation-focused data science task appropriate for Data Scientist interviews.
Read the full Roblox Data Scientist interview experience this question came from