Katkat
Back

Build an end-to-end ML classification pipeline

NextdoorMachine Learning Engineer · Technical Screen
MediumMachine Learning EngineerTechnical ScreenML System Design

Last updated

This question evaluates a candidate's competency in building end-to-end tabular classification pipelines, including data loading and splitting, missing-value handling, categorical encoding, feature scaling, model training and comparison, hyperparameter tuning, metric-based evaluation, model persistence, and batch inference.

Given a tabular dataset in a CSV file, implement an end-to-end pipeline to perform a classification task. Requirements: (

  1. load the data; (
  2. create stratified train/validation/test splits; (
  3. handle missing values and encode categorical features; (
  4. standardize numeric features; (
  5. train a simple baseline (e.g., logistic regression) and at least one stronger model (e.g., gradient boosting or a small neural network); (
  6. tune key hyperparameters with cross-validation; (
  7. report accuracy, precision, recall, and ROC-AUC on validation and test; (
  8. persist the trained model and preprocessing steps; (
  9. implement batch inference via a predict(input_csv_path, output_csv_path) function or CLI. If using a neural network, write a correct training loop with optimizer initialization, forward pass, loss computation, backward pass, and an explicit optimizer step. Briefly explain design choices and how you would productionize this pipeline.