Back to index
Coursera3 courses, roughly 2 months at a few hours a week·Free to audit, or about $49/mo on a Coursera subscription for the certificate

Machine Learning Specialization (Andrew Ng)

4.8

Still the best place to actually understand machine learning rather than just run a library. The 2022 rebuild swapped Octave for Python and modernised the examples, and the teaching is as clear as it has ever been. If you only do one foundational ML course, I would point you here.

What We Liked

  • Andrew Ng explains the intuition behind the maths better than almost anyone teaching today
  • The rebuild uses Python, NumPy and scikit-learn, so the skills carry straight into real work
  • Genuinely beginner-friendly, you only need high-school maths and a little Python
  • Free to audit if you skip the certificate, which makes the value almost unbeatable

What Could Be Better

  • It is foundations, not a job-ready portfolio, you will still need projects of your own afterwards
  • The labs hold your hand a lot, so you can pass without writing much code from scratch
  • Three courses can feel slow if you already know the basics and just want depth

Detailed review

I have recommended this course in one form or another for years, and the 2022 rebuild made it easier to recommend, not harder. The original taught machine learning in Octave, which was always the one thing people grumbled about, and the new version moves everything into Python with NumPy and scikit-learn, which is what people actually use. The three courses take you from linear and logistic regression through neural networks and on to unsupervised learning, recommenders and reinforcement learning, and the sequencing is so well judged that each idea lands before the next one needs it. What still sets Andrew Ng apart is that he teaches the intuition first.

You do not just learn that gradient descent works, you come away with a mental picture of why it works, and that picture is what lets you debug a model later when something goes wrong. That is rare. Most courses hand you a recipe and move on. The trade-off is that the guided labs do a lot of the lifting for you.

You fill in the interesting lines, but a fair amount of scaffolding is already there, so it is entirely possible to finish a course feeling confident and then freeze the first time you open a blank notebook. My advice is to fight that by rebuilding at least one lab from an empty file, and by starting a small project of your own partway through. The other honest caveat is scope. This is a foundations specialization, not a bootcamp, and it will not by itself land you a job or fill a portfolio.

What it will do is give you the conceptual base that makes everything afterwards faster to learn, from the Deep Learning Specialization to fast.ai to just reading papers without drowning. For most people that base is exactly what is missing, and the fact that you can audit the whole thing for free and only pay if you want the certificate makes it one of the best-value entries in the entire field.

[ final ]

The verdict.

If you want to truly understand what is happening under the hood before you start stacking libraries, start here. Pair it with the Deep Learning Specialization once you are through, and build something of your own on the side so the knowledge sticks.