// the find
justmarkham/scikit-learn-tips
:robot::zap: 50 scikit-learn tips
A collection of 50 bite-sized scikit-learn tips from Kevin Markham (Data School), each paired with a Jupyter notebook and YouTube video. It's squarely aimed at practitioners who know the basics but keep hitting the same friction points — wrong imputer, leaking the test set, fumbling Pipeline introspection.
The tips target real mistakes people actually make, not toy examples — tip 4 (fit_transform vs transform) and tip 35 (don't use .values) are exactly the kinds of things that silently corrupt results. Each tip is self-contained, so you can drop into any notebook without reading from the start. The multi-modal format (notebook + video + LinkedIn post) means you can skim the code or watch someone walk through it depending on how you learn. Binder support means zero setup to run any notebook.
No updates since September 2022 — scikit-learn has shipped set_output(), metadata routing, and significant Pipeline improvements since then, none of which are covered. About 10 of the 50 tips have no notebook at all, just a LinkedIn post, which is a dead end if you want to run the code. The collection stops at 50 and was never extended, so gaps like custom transformers, calibration, and newer model inspection APIs go unaddressed. Star count (1.7k) suggests this didn't build much of a community beyond the original social media audience.