Who cancels in the first 90 days?
Pennant loses too many subscribers in their first three months. Build a churn table without peeking at the future, find who leaves, fit a first model and plan what the retention team should try.
Level 1: Newcomer
0 points
Sign in to save your progress across devices.
Learn Python and pandas, query your own data in SQL, then build models you can evaluate honestly and explain. Every exercise runs and is checked, and the pathway ends in a churn project, a timed assessment across the tools and the interview.
About 44 hours of lessons, from the estimate on each lesson.
Nothing done yet. Progress saves in this browser as you finish lessons, and in your account when you sign in.
Take the steps in this order. Each step is a set of units from one course, not the whole course: the units this role uses. Open a unit to start its first lesson.
Data science is done in code. You learn values, collections, control flow and functions, then how to read and write the files your data arrives in.
Most of a data scientist's time goes into preparing data. You load and inspect tables, filter, clean and merge them, group and reshape, work with time series and chart what you find.
The data you model usually lives in a database, and many interviews include a SQL round. You query, summarize and join tables, then use window functions and the analytics patterns behind cohorts and funnels.
Here you build models the way the job does: frame the problem, split the data so the evaluation is honest, fit regression and classification baselines, prepare features in a pipeline, choose between models, handle rare events and read the probabilities they give you.
Build a local document index, check source scope and citations, then keep development and held-out evaluation cases separate. The notebook uses deterministic fixture vectors and scripted replies rather than learned embeddings or live generation.
Each project starts from a fictional company's brief and moves between tools the way the job does: what you build in one stage is where the next one starts. Every brief is free to read.
Pennant loses too many subscribers in their first three months. Build a churn table without peeking at the future, find who leaves, fit a first model and plan what the retention team should try.
A checkout test offered half the visitors an annual plan. The dashboard says it worked; count the test fairly and decide whether the difference is real before it ships.
When you finish the units, test yourself across the tools of the job at once: 12 practical tasks in 50 minutes, each in the tool it is about and checked automatically. The clock keeps running if you close the page.
| Tool | Tasks | Minutes |
|---|---|---|
| Python | 3 | 13 |
| pandas | 3 | 13 |
| SQL | 3 | 11 |
| Machine learning | 3 | 13 |
The assessment is free. The result is practice feedback in the Interview Hub's report format, not a hiring decision. Retakes draw different tasks where the question pools allow.
The Interview Hub runs the data scientist interview round by round, at a fictional employer: you answer out loud or in writing, the interviewer follows up, and the report quotes your answers.
A data scientist uses data to predict or explain something a business cares about: which customers are about to leave, whether a new checkout really converts better, how much stock a store will need next week. The job is less about clever algorithms than about getting the question right, preparing data that doesn’t leak the answer, evaluating honestly and explaining the result to the people who will act on it.
Python comes first because everything else is written in it. pandas follows, because preparing data is most of the work. SQL gets you the data in the first place. Machine learning follows, when you can prepare the inputs yourself and judge whether a model’s score means anything. The final step applies source boundaries and held-out evaluation to a language-model application.
At Pennant, a subscription company loses too many members in their first three months. You build a churn table without peeking at the future in SQL and pandas, find who leaves, fit a first model and plan what the retention team should try. In the annual offer case you decide whether a checkout test result is real before it ships. Both are the kind of work data science interviews ask you to walk through. The assessment then shows which tool to practice before you apply.
Many data scientist jobs ask for a quantitative degree, and some ask for a master's. Read the ads you plan to apply to. This pathway practices the work those jobs test: preparing data, modeling, evaluating and explaining. It is practice, not a qualification.
Analysts answer business questions with data; data scientists also build predictive models and run experiments. If you are starting out, the data analyst pathway shares most of the first steps and leads to more entry-level jobs.
The lessons on this pathway add up to about 44 hours. The machine learning units take the longest. We don't promise how long getting hired takes.
The first unit of every course is free, with no account. One Course opens one course in full; All Access opens the whole pathway. The pricing page shows every plan and its price.