Take-Home Projects
0 of 1 free take-home accessed
Two-part take-home from Goodwater Capital. Part 1: Write a SQL query computing YoY growth of the trailing 28-day average of product-level app downloads across iOS and Android, joining a Products table and a Downloads table. Part 2: Analyze sales and cohort data for several successful e-commerce companies to identify common 'winner' patterns, then train a predictive model in Python to forecast Brandless's monthly sales over 12 months (through September 2019) and assess Brandless's likelihood of becoming a winner. Deliverable is a presentation for data-savvy investors plus the modeling code.
Goodwater capitalBuild a binary classifier to predict whether a building permit's 'type' field is 'ELECTRICAL' using a training set of about 100k labeled permits, then apply it to about 25k unlabeled test permits.
BuildzoomBuild a Python ML model (using customer and product features + last month's purchase history) to predict which books next-month customers will keep. Use those predictions to calculate projected revenue, compare it against the original loan amount and next purchase order cost, and print 'Yes' or 'No' to whether the company will be financially solvent. The assignment also asks for a brief write-up on model choice, validation strategy, and how to run the script.
Candidates are given daily orders, visits, and revenue data across four geographies (CA, IL, NYC, IA) for the first half of 2014 from Grubhub, and are asked to recommend which geography should receive the company's sole available product-development team, with reasoning and stated assumptions where data is insufficient.
Grubhub4-part take-home for Hopper's Senior DS-Air role (about 1 hour total). Exercise 1: write a nearest-airport function using lat/long data. Exercise 2: analyze a Price Freeze A/B test (Champion vs. Challenger) with device-type and discount confounders. Exercise 3: design a numeric trip representation from search-session data and define a similarity metric. Exercise 4: define success and improvement metrics for Hopper's Price Drop feature.
HopperData Analyst take-home from MasterClass. The candidate is given one week of clickstream data (11/1–11/7/2017) for users who visited the Gordon Ramsay course marketing page, captured via Segment's analytics.js and stored in Redshift/Amplitude. Five tables are provided covering pageviews, homepage clicks, course marketing page clicks, purchase intent clicks, and completed purchases. The task is open-ended: explore the data and derive insights about user behavior, then submit findings in any format along with any code used.
MasterclassA data analytics take-home asking the candidate to: (1) define and label driver churn (binary 0/1), (2) calculate churn rate from the provided data, (3) cluster drivers into churn-based segments, (4) propose interventions to reduce churn and assess business impact, and (5) develop a hypothesis to size the opportunity of churn reduction. Three relational CSV datasets are provided: driver onboarding dates, ride-level metrics, and ride event timestamps spanning April–June 2016.
LyftRepair a preprocessing pipeline so training, validation, and inference agree. Includes Python starter code, CSV data, and tests. A focused practice adaptation of a reported Capital One Data Scientist assessment.
Capital One