Machine learning from first principles
8 steps · about 4 h 10 min · Beginner
- Frame what machine learning does (guide, Google for Developers, 20 min)
Start with the kinds of problems models can and cannot solve.
- Learn from a simple predictive model (tutorial, Google ML Crash Course, 30 min)
Linear regression exposes features, labels, loss and fitting without much machinery.
- Turn scores into classifications (tutorial, Google ML Crash Course, 30 min)
Logistic regression introduces probabilities and decision boundaries.
- Measure classification quality (tutorial, Google ML Crash Course, 35 min)
Precision, recall and thresholds prevent accuracy from hiding costly mistakes.
- Recognise overfitting (tutorial, Google ML Crash Course, 25 min)
Train and validation behaviour reveal whether a model learned patterns that generalise.
- Train a model with scikit-learn (tutorial, scikit-learn, 35 min)
A small end-to-end example turns the concepts into a repeatable Python workflow.
- Check fairness and hidden harms (guide, Google ML Crash Course, 30 min)
Useful evaluation includes who is represented and who bears each kind of error.
- Continue with practical deep learning (tutorial, fast.ai, 45 min)
The free course provides a project-led next step once the basic workflow is clear.