One typo, opposite answer
October 27, 2026
DataFest wrapped on Sunday. Thank you to everyone who learned, built, and climbed the leaderboard this past week.
This issue starts with a plain-English guide to data science, where a single typo in ten days of sales notes flips the answer completely. AI can calculate an answer in seconds. Knowing whether the data and conclusion hold up is the skill Dataquest helps you build.
Here's what's inside:
- Top Read: What data science is, walked through one small project from question to answer. Read the guide
- DataFest Winners: Meet the three learners who finished at the top of the leaderboard.
- Live Webinar: Join us on Thursday, October 29 at 12:30 PM ET. Save your spot
- Project of the Week: Predict heart disease risk and learn why accuracy isn't enough. Start the project
- From the Community: A construction dashboard, a lesson on comparing variability, and a backprop question you can help answer. See the dashboard
- Data & Resources: DuckDB 2.0, a global forest-type dataset, Python 3.15, and more to build with. Explore the resources
- What We're Reading: 200,000 similes, and what they show about treating text as data. Start reading
- Refer a Friend: Learning sticks better with someone alongside you. Refer a friend
Top Read
What Is Data Science?
One coffee cart, ten days of notes, and the six steps from question to answer.
Data science is using data to answer questions and make better decisions. This guide shows what that looks like in practice with one small project: a friend's coffee cart, ten days of high temperatures and iced drink sales, and one question. Do hotter days go with more iced drinks sold?
The notes hide two problems: a blank Sunday and a sales figure of 240 on a cool Tuesday. Fix the typo (it was 24) and drop the blank day, and the correlation is 0.99. Leave the typo in, and it comes out at −0.29, pointing the wrong way entirely.
Why it matters: AI tools can write the code for every step. Catching the 240 and knowing that nine days can't prove cause and effect is still on the person reading the result. The guide also covers how data science differs from AI, analytics, and machine learning, and where to start learning.
DataFest 2026
The DataFest Code Challenge is a wrap.
Thank you to every learner who joined in, whether you finished one lesson or a whole course. Congratulations to the three learners at the top of the leaderboard:
- Winner: Mohamed Lamine Tekli, 1,895 pts
- 1st runner-up: Halyna Merkotan, 1,600 pts
- 2nd runner-up: Joseph A. Swierczek, 1,410 pts
Live Webinar
Storytelling Data Visualization on Exchange Rates
Build a narrative graph in Python, live.
Thursday, October 29, 12:30 PM ET
A chart can be technically correct and still say nothing. In this live session, we'll take Euro exchange rate data, clean it, smooth the noise with a rolling mean, then find a story angle and sketch the graph before writing a line of Matplotlib.
The design decisions are the point: what to highlight, what to cut, and how information design and pre-attentive attributes guide the reader's eye. You'll leave with a portfolio-worthy visualization.
Guided project · Python · Matplotlib · Intermediate
Project of the Week
Predicting Heart Disease · Guided project · KNN · UCI Heart Disease dataset · 5–8 hrs total
Classify patients as high or low heart disease risk using anonymized medical data. In healthcare, a false negative has real consequences, so the trade-off between precision and recall stops being abstract fast.
Skills and tools you'll practice: KNN, scikit-learn, pandas, confusion matrices, precision and recall.
Take it further: compare KNN against logistic regression and random forest. A model comparison section makes a much stronger portfolio piece than a single model in isolation.
Why employers care: choosing precision and recall as your primary metrics, and explaining why, shows you can evaluate a model beyond accuracy.
From the Community
See how other learners are putting the fundamentals to work, and where one of them could use a hand.
- Construction Agency Dashboard: Mohammad's Power BI project combines a program summary, project-level reporting, a client dashboard, financial analysis, and geographic insights in one report, with strong domain knowledge behind every view.
- Comparing Statistical Dispersion Metrics: Mohammad explains why standard deviation can mislead when variables sit on different scales, how the coefficient of variation helps, and why proportions work better for binary variables.
- Understanding Backpropagation Through Neural Network Layers: Hooi is working through the Zero to GPT skill path and would appreciate help with the calculations behind backpropagation. Know your gradients? Jump in.
Data & Resources
- DuckDB 2.0.0 LTS (DuckDB Foundation): Upgrade a SQL-on-files project in a copy, test the new VARIANT type and async I/O, and document what changes.
- Forest Typology (ForTy) 2020 and 2017–2025 time series (Google DeepMind, Google Research, WRI, IIASA): Map natural forests, plantations, and tree crops at 10-meter resolution, then measure land-use change over time.
- MultiLangBench (OpenAI): Run the open SQLite-based test suite to compare coding agents across languages and see how reproducible, behavior-based evaluations are designed.
- AI Economic Indicators (Stanford Digital Economy Lab): Use the monthly dashboards on AI exposure, hiring, and adoption to write an evidence-based career or labor-market brief.
- What's New in Python 3.15 (Python Software Foundation): Port a small data project and test lazy imports, UTF-8 defaults, frozendict, the new profiler, and the typing and free-threading updates.
What We're Reading
- 200,000 similes from popular fiction: The Pudding pulled every "as ___ as ___" comparison from popular fiction to see how original they are. The answer: not very. Beyond the fun, it's a tidy case study in treating text as data: define a pattern, extract it at scale, and visualize what you find.
- The changing landscape of skills in the age of AI: A joint report from the ILO, ETF, Cedefop, Eurofound, the European Commission, and UNESCO finds that AI adoption raises demand for higher-order thinking, digital, and data science skills, with safe and ethical AI use becoming a foundation rather than an extra.
- The most in-demand skills in data and AI jobs: An October 1 index of 33,406 data and AI job postings found SQL in 53.5%, Python in 43.2%, and data analysis in 30.4%, with machine learning at 18.9%. It's one index, but the pattern is clear: the foundations still lead.
Refer a Friend
Learning sticks better with company.
Know someone who'd like to build data skills alongside you? Refer a friend to Dataquest and they'll get 20% off while you get $20.