Machine Learning Crash Course is Google's free introduction to how machine learning models work, used by millions of learners since 2018 and recently updated with material on large language models, AutoML, and production systems. Each module combines short readings, interactive visualisations, quizzes, and programming exercises that run in the browser through Google Colab, so nothing has to be installed.
The course does not assume previous knowledge of machine learning, but it is not a gentle programming course either. Google's prerequisites page asks for comfort with variables, linear equations, graphs, histograms, and averages, and recommends solid programming skills, ideally in Python.
Four blocks, from regression to fairness
The modules are grouped into four parts, and the site shows an estimated time for each:
| Block | Modules | What you will practise |
|---|---|---|
| ML models | Linear regression, Logistic regression, Classification | loss, gradient descent, thresholds, precision, recall, ROC and AUC |
| Data | Numerical data, Categorical data, Datasets and overfitting | normalisation, binning, one-hot encoding, feature crosses, generalisation |
| Advanced models | Neural networks, Embeddings, Intro to large language models | hidden layers, activation functions, tokens, Transformers |
| Real-world ML | Production systems, AutoML, Fairness | how ML systems are built and how to audit models for bias |
Individual modules are designed to fit into one sitting: linear regression is listed at around 80 minutes, logistic regression at about 35, and classification at about 70.
Prepare before starting
The prework page suggests three short preparations for newcomers: Google's Introduction to Machine Learning course, and two Colab tutorials on NumPy and pandas that cover all the library knowledge the exercises require. Doing them first saves time later, because the programming exercises use both libraries constantly.
On the maths side, the course lists algebra (functions, logarithms, the sigmoid function), basic linear algebra, and statistics such as mean, median, and standard deviation. Calculus is optional and only needed to fully understand backpropagation.
Who will benefit
The course is a strong fit for developers and data analysts who want to understand the models behind the tools they use, and for students who have finished an introductory Python course. Learners without programming experience can still follow the readings and visualisations, but will miss much of the value of the exercises.
It focuses on concepts and practice rather than on a specific framework, so it pairs well with library-specific tutorials afterwards.
A route through the course
- Complete the NumPy and pandas tutorials if these libraries are new to you.
- Work through the first block in order; later modules reuse its vocabulary of loss, gradients, and metrics.
- Do every interactive exercise and quiz before the Colab notebooks: they reveal misunderstandings quickly.
- Choose the advanced and real-world modules according to your goals; the fairness module is worth reading even if you never train a model yourself.