Free
Google's rebuilt ML Crash Course — regression, classification, neural networks, embeddings and LLMs, with interactive visualisations and per-module badges.
Intermediate10+ hours
Models learn patterns from data instead of following rules you wrote. That's the whole idea, and the distance between understanding that sentence and training something useful is shorter than it was five years ago — the tooling has absorbed most of the difficulty.
Five courses here, ranging from a three-hour notebook introduction to a full deep learning curriculum. One of them is paid.
Kaggle Learn: Intro to Machine Learning is six lessons in hosted notebooks — decision trees, validation, overfitting, random forests, with pandas and scikit-learn — and it runs about three hours. No install, no environment problems, and by the end you'll know whether this holds your interest. The lessons build on one dataset the whole way through, so nothing stays abstract for long, and there's a completion certificate at the end that costs nothing. Nothing else here is that cheap to try.
The two serious courses take opposite approaches, and it's worth choosing deliberately. Practical Deep Learning for Coders from fast.ai puts a trained, working model in front of you in lesson one and explains the theory later, with PyTorch and fastai underneath. Andrew Ng's Machine Learning Specialization builds upward instead — regression, neural networks, trees, clustering, recommenders — reasoning before result. Ng's is subscription-only; fast.ai is free, and its part one was recorded in 2022.
Everyone hits it eventually: linear algebra, probability, a bit of calculus. Our only answer on this page is StatQuest with Josh Starmer, whose drawn, deliberately silly explanations run from p-values through to transformers. It's excellent, and it's a reference library rather than a course — you arrive with a question, not for a syllabus. Google's Machine Learning Crash Course partly fills the gap with interactive visualisations across regression, classification, neural networks, embeddings and LLMs.
Three hours on Kaggle, then commit to fast.ai or Ng and finish it. What we don't list is a proper maths course, anything on NLP or LLMs beyond one Google module, and anything at all on deploying a model — which is most of the job once someone is paying you to do this.
Google's rebuilt ML Crash Course — regression, classification, neural networks, embeddings and LLMs, with interactive visualisations and per-module badges.
Six lessons, about three hours, in hosted notebooks — decision trees, validation, overfitting and random forests with pandas and scikit-learn.
DeepLearning.AI
Andrew Ng's rebuilt three-course ML specialization in Python — regression, neural networks, trees, clustering and recommenders. Subscription-only, not free.
Jeremy Howard's top-down deep learning course — train a working model in lesson one, PyTorch and fastai underneath. Part 1 was recorded in 2022.
Josh Starmer
Josh Starmer's drawn, deliberately goofy explanations of statistics and ML — p-values through transformers. A reference library, not a course.