Build a model using only pandas/numpy
Company: Nextdoor
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
You are given a tabular dataset as a pandas DataFrame `df` with:
- Feature columns (numeric and/or categorical)
- A target column `y` (either binary classification or continuous regression)
You may use **pandas** and **numpy** (and standard Python), and you may use Google for documentation, but you may **not** use AI assistants or high-level ML libraries (e.g., scikit-learn).
Tasks:
1. **Data preparation**
- Handle missing values.
- Encode categorical variables.
- Split into train/validation (or implement cross-validation).
- Standardize/normalize features when appropriate.
2. **Modeling (from scratch)**
- Choose a reasonable baseline model (e.g., linear regression for regression; logistic regression for binary classification).
- Implement training using numpy (e.g., gradient descent).
- Implement prediction.
3. **Evaluation**
- Pick suitable metrics (e.g., MSE/RMSE for regression; accuracy/precision/recall/F1/AUC for classification).
- Explain how you would detect overfitting and what you would do about it.
4. **Concept questions** (be prepared to explain)
- Bias–variance tradeoff
- Regularization (L1 vs L2) and how it changes the objective
- Class imbalance handling
- Feature scaling: when it matters and why
- Train/validation/test leakage and how to avoid it
Overview: This question evaluates proficiency in applied machine learning competencies including data preprocessing, categorical encoding, feature scaling, implementing baseline models and optimization using numpy/pandas, and selecting appropriate evaluation metrics.