Stanford CS229 — Machine Learning (Public Materials): Review

I’m Priya, a data scientist at a mid-size SaaS company. I use scikit-learn and gradient boosters daily, and a few years into the job I started feeling the ceiling of knowing how to fit models without fully owning why they work. That gap is what sent me back to CS229’s public materials — and it’s still the best free answer I’ve found, with caveats I’ll date precisely.

What you’re actually getting: Andrew Ng’s Autumn 2018 lecture series, about 27 hours on YouTube, plus the public lecture notes by Ng and Tengyu Ma — a PDF treated as a reference text, last updated June 2023. The on-campus course still runs every quarter, but current-quarter materials (Canvas, forums, new videos) sit behind Stanford login. A tuition-based Stanford Online version exists and was closed to enrollment as of August 2026, no price listed. So “free CS229” means 2018 videos plus living notes, and nothing else.

The syllabus is the theory-first counterweight to code-first courses: supervised learning through GLMs, kernel methods and SVMs, backpropagation derived from first principles, learning theory (bias/variance, generalization, double descent), unsupervised methods including full EM derivations, and reinforcement learning through LQR/LQG and REINFORCE. The notes have added self-supervised learning and foundation models in recent revisions — newer than the videos, which matters. Prerequisites are real: CS106-level Python/NumPy, CS109-level probability, MATH51-level calculus and linear algebra. Third-party self-study guides estimate around 100 hours, and the popular csdiy.wiki rates difficulty 4/5. Both match what I see in learner discussions: the math barrier is the single most recurring theme.

On versions, since that’s my obsession: the videos are 2018, and the deep learning and LLM portions of the live syllabus have moved well past them. The notes are newer but lack matching public lectures consistently. My working method was to treat the 2018 series as the lecture spine and the Ng/Ma PDF as the living document. That split holds up.

What the free track doesn’t give you: graded assignments, an autograder, TAs, a forum, any certificate. Enrolled students get four problem sets, a midterm, and a substantial project; you supply your own structure and error-checking. The paid Stanford Online route also requires a conferred bachelor’s with 3.0+ GPA and, again, was closed when we verified.

The reception is earned. Reddit’s ML communities call it “the real thing” for people who suspect simplified MOOCs are leaving something out, and self-study communities keep the notes as a reference after finishing. Ng’s 2018 classroom teaching is praised for clarity with a consistency I rarely see.

Alternatives: his Machine Learning Specialization is the gentler, certificate-bearing version of the same teacher; fast.ai inverts the order (build first, derive later); CS224n’s public materials are the NLP sibling. CS229’s unique value is rigor per dollar — nothing free teaches why the algorithms work as well.

Bottom line: if you have the math and want foundations that will still be true in ten years, start here. Accept 2018 videos, zero support infrastructure, and self-supplied discipline. Need credit? Check Stanford Online — closed as of our last verification.