Machine Learning Specialization — Andrew Ng

Andrew Ng's original Stanford machine learning course is the reason a whole generation of engineers know what gradient descent is. This is its modern replacement, rebuilt by DeepLearning.AI and Stanford Online as a three-course specialization, and the biggest change is that it's in Python with NumPy, scikit-learn and TensorFlow rather than the Octave everybody used to complain about.

Three courses, roughly 33, 34 and 28 hours by Coursera's own estimates, with a stated pace of about two months at ten hours a week.

What it covers

• Supervised Machine Learning — linear regression, logistic regression, cost functions and gradient descent, built with NumPy and scikit-learn
• Advanced Learning Algorithms — neural networks in TensorFlow, decision trees and tree ensembles, plus practical advice on debugging a model that isn't working
• Unsupervised Learning, Recommenders, Reinforcement Learning — clustering, anomaly detection, recommender systems and an introduction to RL

The pricing problem

This one is not free, and it should not sit in a free-course directory without a very loud caveat. Coursera's page states you cannot take the specialization for free — access runs through a subscription at around $49 a month, with financial aid available on application, and certificates only issued on paid completion. If you're subscribed already the cost per hour is fine; if you came here looking for something free, this isn't it.

What it assumes you know

Coursera labels it Beginner, and Ng's teaching is patient enough to justify that, but the recommended experience is basic Python and high school mathematics. In practice, if you're rusty on what a derivative is, the gradient descent material will be uncomfortable — not impossible, since Ng explains the intuition carefully, but uncomfortable.

Who it's for

People who want the actual mechanics rather than an API tour. The optional labs are the best part of the specialization: you implement things by hand before being shown the library call, which is why graduates of this course can debug models that others can only restart. Skip it if you need free, skip it if you want deep learning specifically — Ng's separate Deep Learning Specialization is that course — and skip it if you're looking for MLOps, deployment or data engineering, none of which appear.

Our take

Still the best-taught introduction to machine learning fundamentals available anywhere, and the price is the honest objection. The other weakness is currency of scope: it's a fundamentals course in a field that has moved heavily toward large pretrained models, so nothing here prepares you for working with transformers or LLMs. Take it for the foundations, and be clear-eyed that it's a paid course in a free directory.