Cursarium logoCursarium
intermediateFree

Machine Learning

by Andrew Ng · Stanford Online

4.9
(5,200 reviews)
500K+ enrolled11 weeksUpdated 2024-09

Our Verdict

Worth it — with caveats

Stanford CS229 (Stanford Online / Stanford Engineering Everywhere), the graduate-level Machine Learning course created by Andrew Ng, is the gold-standard free option for learners who want the mathematics behind ML rather than just an API tour. The most-cited free version is the Autumn 2018 lecture series taught by Andrew Ng (20 lectures on YouTube), backed by the canonical CS229 lecture notes and four problem sets covering supervised learning, learning theory, unsupervised learning, and reinforcement learning. It is rigorous and unforgiving: it assumes comfort with linear algebra, multivariable calculus, probability, and Python/NumPy, and offers no certificate, grading, or hand-holding when self-studied. Take it if you want first-principles depth (derivations, GLMs, EM, SVM duality); skip it if you want a gentle, applied, project-first introduction.

Outstanding, free, mathematically rigorous ML foundation from Stanford and Andrew Ng, but only worthwhile if you already have solid linear algebra, calculus, probability, and Python; absolute beginners will struggle and there is no official certificate.

Best for: CS/engineering students and working developers who already know linear algebra, multivariable calculus, and probability, want to understand the math behind ML algorithms (not just call libraries), and are comfortable self-pacing through derivations and problem sets. Ideal as preparation for research, deep-learning courses (CS231n/CS224n), or interviews that probe ML fundamentals.

Skip if: Complete beginners, non-technical professionals, or anyone wanting a gentle, applied, project-first path. People who need a certificate, graded feedback, deadlines, or hand-holding, or who are rusty on linear algebra/probability, should start elsewhere (Coursera ML Specialization, fast.ai, or Google ML Crash Course) first.

About This Course

Stanford's foundational machine learning course covering supervised learning, unsupervised learning, and best practices in ML.

What You'll Learn

Supervised learning: linear regression, weighted least squares, logistic regression, Newton's method, perceptron
Generalized Linear Models and the exponential family, plus generative models (Gaussian Discriminant Analysis, Naive Bayes)
Support Vector Machines, kernels, and the kernel trick
Learning theory: bias-variance tradeoff, regularization, and model/feature selection
Neural networks (basics and training) and tree ensembles / decision trees
Unsupervised learning: K-means, mixture of Gaussians, Expectation-Maximization, factor analysis, PCA, and ICA
Reinforcement learning: MDPs, Bellman equations, value/policy iteration, LQR/LQG, Q-learning, and policy search

Curriculum

Introduction & Linear Regression

Course motivation, supervised-learning setup, and linear regression with least squares.

Logistic Regression, GLMs & the Exponential Family

Weighted least squares, logistic regression, Newton's method, perceptron, and generalized linear models.

Generative Learning (GDA & Naive Bayes)

Gaussian Discriminant Analysis, Naive Bayes, and Laplace smoothing.

Support Vector Machines & Kernels

SVM formulation, the kernel trick, and margins.

Learning Theory & Model Selection

Bias-variance tradeoff, regularization, and model/feature selection.

Tree Ensembles & Neural Networks

Decision trees, ensembles, and neural-network basics plus training.

Practical Advice for ML Projects

Debugging learning algorithms and applying ML to real projects.

Unsupervised Learning

K-means, mixture of Gaussians and EM, factor analysis, PCA, and ICA.

Reinforcement Learning

MDPs, Bellman equations, value/policy iteration, LQR/LQG, Q-learning, and policy search (REINFORCE, POMDPs).

Problem Sets 0-4

Four graded-style problem sets (plus PS0) with data files and solutions, including derivations like the exponential-family mean/variance and Hessians.

Prerequisites

  • Linear algebra (vectors, matrices, eigenvalues) — MATH51 / CS205L level
  • Multivariable calculus (gradients, Hessians, partial derivatives)
  • Probability and statistics — CS109 / MATH151 level
  • Programming able to write non-trivial Python/NumPy code
  • Computer-science fundamentals at CS106A/B level

Instructor

Andrew Ng

Instructor · Stanford Online

Pros & Cons

Pros

  • Genuinely rigorous: teaches the mathematical derivations behind ML (GLMs, EM, SVM duality, RL), not just library usage — the depth its Coursera counterpart lacks.
  • Completely free: all 20 Autumn-2018 lectures stream on YouTube, with lecture notes, problem sets, and solutions plus linear-algebra/probability refreshers via Stanford Engineering Everywhere.
  • Authoritative source: created and taught by Andrew Ng (co-founder of Coursera and Google Brain); CS229 is Stanford's flagship ML course and a long-standing field reference.
  • Broad, coherent coverage of supervised, unsupervised, and reinforcement learning plus a dedicated lecture on debugging/applying ML in practice.
  • Self-paced with no deadlines or fees when studied via the public materials.

Cons

  • Steep prerequisites and difficulty: official problem sets demand things like deriving the exponential family's mean/variance and computing Hessians; historically hard exams (one cited median ~46/100). Rusty math means a hard wall.
  • No certificate, grading, or instructor feedback for self-learners — you must self-assess against provided solutions.
  • Theory-first and light on modern hands-on tooling/large projects; it is not an applied, portfolio-building course.
  • Fragmented free experience: the public version is YouTube videos + PDF notes; the live cs229.stanford.edu site restricts current materials to Stanford-login affiliates, and the canonical free run is Andrew Ng's 2018 edition (current quarters are taught by other faculty).

Alternatives To Consider

Frequently Asked Questions

Is Machine Learning free?

Yes — Machine Learning is free to access. Free. All 20 Autumn-2018 lectures are on YouTube; Stanford Engineering Everywhere provides the videos, canonical lecture notes, four problem sets with solutions, and math refreshers at no cost. No official certificate is offered for the free materials. The current cs229.stanford.edu site shares live syllabus/materials only with Stanford-affiliated logins; a paid graduate-credit/professional version (CS229 / XCS229) exists separately through Stanford.

Who is Machine Learning for?

CS/engineering students and working developers who already know linear algebra, multivariable calculus, and probability, want to understand the math behind ML algorithms (not just call libraries), and are comfortable self-pacing through derivations and problem sets. Ideal as preparation for research, deep-learning courses (CS231n/CS224n), or interviews that probe ML fundamentals.

What will you learn in Machine Learning?

Supervised learning: linear regression, weighted least squares, logistic regression, Newton's method, perceptron; Generalized Linear Models and the exponential family, plus generative models (Gaussian Discriminant Analysis, Naive Bayes); Support Vector Machines, kernels, and the kernel trick; Learning theory: bias-variance tradeoff, regularization, and model/feature selection.

What are the prerequisites for Machine Learning?

Linear algebra (vectors, matrices, eigenvalues) — MATH51 / CS205L level; Multivariable calculus (gradients, Hessians, partial derivatives); Probability and statistics — CS109 / MATH151 level; Programming able to write non-trivial Python/NumPy code; Computer-science fundamentals at CS106A/B level.

Is Machine Learning worth it?

Outstanding, free, mathematically rigorous ML foundation from Stanford and Andrew Ng, but only worthwhile if you already have solid linear algebra, calculus, probability, and Python; absolute beginners will struggle and there is no official certificate.