DeepLearning.AI & Stanford Online (Coursera)

Machine Learning Specialization Review — Honest Analysis of Andrew Ng's Course

Andrew Ng's updated Machine Learning Specialization is the strongest modern on-ramp into machine learning for learners with basic Python and high-school math. The move from Octave to Python, the expanded neural-network unit and the auto-graded notebooks address the loudest complaints about the 2012 version. It is still a foundation course — not a path to a job on its own — and prior learners gain little by retaking it.

Final score

from 38 analysed opinions

Published AI-researched, editor-audited

Distribution of opinions

25 positive7 neutral6 negative/ 38 total

Per-criterion scores

Content quality4.2 / 5

Praised for intuitive explanations and the expanded neural networks unit, but reviewers note the new version trades depth for accessibility — backprop is brushed past, RL feels like a preview.

Instructor4.6 / 5

Andrew Ng's pedagogy gets near-universal praise across HN and blogs. Multiple commenters describe him as the best instructor they ever had; complaints are essentially absent.

Value for money4.1 / 5

Content is strong relative to cost, and auditing remains possible. The friction comes from Coursera's subscription gating around grading and certificates — a recurring HN gripe.

Support3.9 / 5

Browser-hosted Jupyter notebooks with auto-grading remove a major friction point from the original. The community forum is active but not deeply mentioned in reviews.

Real-world use3.9 / 5

Builds a real foundation in ideas and Python tooling, but datasets are clean and deployment is out of scope. Reviewers flag the need to supplement with Kaggle or a portfolio project.

What learners said

What people loved

5
  • Andrew Ng's pedagogy is exceptional — clear, intuitive, and famously good at gradient descent×18
  • The shift from Octave to Python with Jupyter and TensorFlow modernises the original course×14
  • Auto-graded browser notebooks remove install friction that blocked beginners in the old version×9
  • Neural networks coverage is expanded and explained with stronger visual intuition×8
  • Beginner-friendly enough for self-taught learners without a heavy maths background×11

What frustrated learners

5
  • Less mathematically rigorous than the 2012 version — backpropagation is barely touched×7
  • Datasets are clean and toy-like; deployment and real-world data wrangling are out of scope×8
  • Coursera's subscription and paywall patterns around grading frustrate some learners×5
  • Reinforcement learning unit in Course 3 feels like a preview rather than a course×4
  • Limited added value for anyone who already completed the original Stanford/Coursera course×4

Real quotes from real users

I took this course and Dan Boneh's cryptography course and both were truly excellent.
criddellHacker News
Andrew Ng's MOOC is among the best game in town. Ng is among the best teachers I have ever seen.
rg111Hacker News
They employ a ton of grey patterns to get you to pay for it. I haven't been able to find out where to audit it yet.
dwallinHacker News
I've taken part 1 of 3 in Andrew Ng's machine learning specialization which covers the math you need to get started.
jonnycoderHacker News
Andrew Ng was the best instructor I ever had in grad school.
levocardiaHacker News
The main ideas of neural networks are explained better in the new specialization, with more intuitive examples and better flow. The notebooks contain amazing visualizations and auto grading.
Usama AhmedBlog
I wish there were more real-world tasks like Kaggle competitions or industry use cases. The reinforcement learning section feels a bit shallow — more of a preview than a full course.
Valentyn DruzhininBlog
Octave and MATLAB will not get you very far in the current field of ML and AI. The new specialization is built on Python, which is the language of modern machine learning.
TMoshForum

Frequently asked questions

Ready to enrol?

You read the score, the pros, the cons and the quotes. If it's still a fit, here's the link.

How we evaluated this

This review synthesizes 38 opinions collected across the public web. Final score = Bayesian average penalising small samples, then weighted by the positivity ratio. No paid placements, no hidden agenda.

  • 25 from Hacker News
  • 10 from Blogs
  • 3 from Forums
Read full methodology