Predict Bike Dock Demand
Quick Overview
This question evaluates a data scientist's competency in time-series demand forecasting, covering temporal target definition, temporal feature engineering, model family selection, evaluation metric alignment with business priorities, and handling data-quality issues like missing data, cold-starts, and distribution shifts.
Two Sigma
Mar 18, 2026, 12:00 AM
Data Scientist
Technical Screen
Machine Learning
2
0
Loading...
You are working on a docked bike-sharing system. Build a model that predicts how many bikes will be checked out from a specific dock in the next hour.
Assume you have access to:
-
trips(trip_id, start_time, start_station_id, end_station_id, user_type) -
station_status(station_id, ts, bikes_available, docks_available, capacity) -
weather(ts, temperature, precipitation, wind_speed) -
calendar(date, is_holiday, is_weekend, special_event)
Discuss:
- How you would define the prediction target and unit of analysis.
- What features you would engineer without leaking future information.
- What model family you would start with and why.
- Which evaluation metrics you would use, and how your choice changes if the business cares more about stock-outs than raw count error.
- How you would split train and validation data for a time-series problem.
- How you would prevent overfitting and handle cold-start stations, missing data, and distribution shifts such as severe weather or holidays.
Solution
Show