Wondering how to learn machine learning without drowning in maths or tools? Start with Python basics, take one structured beginner course, then build small projects on real data. Keep each step short and practical.
Machine learning is one part of a wider field, so if you want the big picture first, read what is artificial intelligence.
| Item | Value |
|---|---|
| Google Machine Learning Crash Course, study time | ~15 hours |
| Google Machine Learning Crash Course, price | Free |
| Andrew Ng’s Machine Learning Specialization, length | Eleven weeks |
| Andrew Ng’s Specialization, certificate cost | $49 per month (free to audit) |
| Dataquest Machine Learning in Python, study time | ~25 hours |
| fast.ai Practical Deep Learning, length | ~7 weeks |
These figures come from Dataquest’s course comparison (dated May 28, 2026) and LearnDataSci’s course guide. Prices, durations and course content can change, so check the source before you rely on them.
What do you need before you start?
You do not need a degree, but two skills help a lot. Dataquest lists Python fundamentals and basic statistics as the prerequisites for its Machine Learning in Python path, as shown in its course comparison.
LearnDataSci makes a similar point about Andrew Ng’s course: because it is not a programming-focused course, you should be somewhat familiar with coding concepts such as loops and functions, or you may have trouble keeping up. It also says some knowledge of linear algebra beforehand would be helpful, according to its guide.
A quick self-check before you pick a course:
- Can you write a loop and a function in Python?
- Do you know what an average and a spread of numbers mean?
- Are you willing to pick up some linear algebra along the way?
If the answer to the first two is no, spend a few weeks on Python and statistics first. It will save you frustration later.
What is a sensible order to learn in?
Google’s Machine Learning Crash Course gives a clear sequence. Its course page recommends that newcomers complete the modules in order, while each module is self-contained so experienced learners can skip ahead.
Following that structure, a beginner path looks like this:
- Basic models: linear regression (linear models, loss, gradient descent, hyperparameter tuning), logistic regression and classification with metrics such as accuracy, precision, recall and AUC.
- Data: working with numerical data, working with categorical data (for example one-hot encoding), and datasets, generalization and overfitting.
- Advanced models: neural networks (perceptrons, hidden layers, activation functions), embeddings and an introduction to large language models, from tokens to Transformers.
- Real-world ML: production ML systems, automated machine learning and fairness.
The crash course has been used by millions of people since 2018, and Google says the refreshed version covers recent advances in AI with more interactive learning.
Which beginner course should you pick?
Dataquest argues that the right course depends on whether you want to understand ML, apply ML, or engineer ML systems in production. Keep in mind that Dataquest is itself a course provider and names its own path as the best hands-on intro, so treat its ranking as one opinion.
Here is how the main options compare, using the figures in the table above:
- Google Machine Learning Crash Course: free, about 15 hours, and a quick way to test whether you enjoy the subject.
- Andrew Ng’s Machine Learning Specialization: built and taught by a Stanford professor as a beginner’s introduction to the maths and programming behind traditional algorithms. It runs over eleven weeks and is free to audit.
- Dataquest Machine Learning in Python: about 25 hours of interactive lessons, with free intro lessons; full path access requires a paid plan.
- fast.ai Practical Deep Learning: free, about seven weeks, and described as top-down and hands-on.
If you are unsure, start with the free crash course. After you finish it, you will know which of the others suits your pace.
How do you practise so it sticks?
Courses only get you so far. Dataquest warns that you can fit a model to a toy dataset and feel you have cracked it, then a real dataset arrives and the model’s accuracy collapses. Its advice is to look for hands-on practice with real datasets, not just toy examples like iris or MNIST.
LearnDataSci also recommends two books alongside any video course:
- Introduction to Statistical Learning, free online, with examples in R that build mathematical intuition.
- Hands-On Machine Learning with Scikit-Learn and TensorFlow, which focuses on applying machine learning in Python.
A simple routine works well. After each module, take a small dataset you care about, split it into training and test data, fit one model, and write down what surprised you.
What pitfalls should you watch for?
The first is trusting a score that looks too good. A model can fit its training data very closely and still fail on new data, which is why the crash course includes a module on datasets, generalization and overfitting. When you reach it, read what is overfitting in machine learning to see how this failure happens and how to spot it.
The second is ignoring fairness. Google’s course includes a module on auditing ML models for fairness, including strategies for identifying and mitigating biases in data. Studying ai bias examples shows why data problems end up as real-world harm.
The third is choosing tools by habit. Dataquest suggests scikit-learn for traditional ML and PyTorch for deep learning, and says courses teaching only older TensorFlow patterns are dating themselves. Google’s crash course is listed with TensorFlow, so pair it with a more current hands-on course if you want to go deeper.
Frequently asked questions
How long does it take to learn machine learning?
Show answer
A first course takes anywhere from about 15 hours (Google’s crash course) to eleven weeks (Andrew Ng’s specialization), according to the sources above. Becoming confident with real projects takes longer, and the sources do not give a fixed time for that.
Do I need to be good at maths?
Show answer
Not at the start. LearnDataSci says Andrew Ng’s course explains the maths in detail, including the calculus and linear algebra behind each concept, though prior linear algebra helps.
Can I learn machine learning for free?
Show answer
Yes, to a large extent. Google’s crash course and fast.ai are listed as free, and Andrew Ng’s specialization is free to audit, with the certificate costing $49 per month as of LearnDataSci’s guide.
Should I learn TensorFlow or PyTorch first?
Show answer
The sources do not settle this. Dataquest recommends scikit-learn for traditional ML and PyTorch for deep learning, so a sensible path is to learn the concepts first and pick a framework when a project needs one.
