Go a little deeper.
Build a little more.

Start with the idea

Explore lessons 1–4 here. Use Google’s course for a second explanation of a line, a probability or an error. You can leave the coding exercises for later.

Make the experiment fair

Explore overfitting and model comparison. Then try the Inria course with basic Python: keep the split fixed, add a baseline and explain the result.

Choose one small project

Pick a clearly documented dataset, write down who a prediction helps, and decide what an error costs. A modest, well-explained result is a strong first project.

Read + watch Foundations

Google ML Crash Course

A practical companion covering regression, classification, data and overfitting, with visual explanations and exercises.

Best for: lessons 1–5. Start with our lab, then read the matching module.

Bring: comfort with algebra and charts; Python, NumPy and pandas for the coding exercises.

Build Python route

Inria’s scikit-learn MOOC

A structured coding route through preprocessing, validation, linear models, trees and ensembles. The static course can be browsed freely.

Best for: lessons 2–6 and the planned ensemble lesson. Rebuild an experiment with a pipeline.

Bring: basic Python and a little familiarity with tables.

Read Reference

scikit-learn: common pitfalls

A useful reference for data leakage and consistent preprocessing, including why a pipeline belongs inside your validation process.

Best for: lessons 2 and 6. Check a project’s split and preprocessing before comparing its scores.

Bring: the train/validation idea; Python for the code examples.

Read + build Go deeper

An Introduction to Statistical Learning

The authors provide a free Python-edition book, with chapter labs. Read selectively when you want more detail behind a concept.

Best for: chapters 3–5 after regression, classification and comparison; chapters 8 and 12 for trees, PCA and clustering.

Bring: algebra, basic statistics and Python for labs.

Watch Next horizon

3Blue1Brown: neural networks

A visual introduction to a network’s layers, weights and activations. Treat this as a bridge to neural networks after the core ML path.

Best for: connecting the features and fitting ideas you have already explored to a new kind of model.

Bring: algebra and curiosity; no coding needed to watch.

Watch + build Next horizon

fast.ai: practical deep learning

A free, project-oriented course for people with coding experience, covering applications including text, images and tabular data.

Best for: an extension after model comparison and your first small project; revisit evaluation as the models become larger.

Bring: coding experience and time for a longer course.

Build Practice data

UCI Machine Learning Repository

Start with a small, documented dataset such as Iris for classification. Keep the original creator, citation and dataset page with your project.

Best for: a Python extension to classification, clustering or the planned capstone.

Bring: basic table skills. Check the individual dataset’s license and context before using its files.

Build Further practice

OpenML

Explore datasets and evaluation tasks when you are ready to make an experiment repeatable. Record the dataset ID and version you choose.

Best for: a follow-on project after the comparison lesson.

Bring: Python and validation basics. Read the selected record’s license, original source and citation.

Read + code Statistics

Practical Statistics for Data Scientists

Companion repository to the second edition (2020) by Peter Bruce, Andrew Bruce and Peter Gedeck, with examples in Python and R.

Best for: exploratory data analysis, sampling and uncertainty, then a statistical perspective on validation.

Bring: some Python or R familiarity and comfort with tables and charts.

The authors mark this repository as no longer maintained and point to the third-edition repository. Our review refers to the second-edition repository linked below.

Read + build Python

Hands-On Machine Learning with Scikit-Learn and PyTorch

Companion notebooks by Aurélien Géron for the first PyTorch edition of his hands-on machine-learning book.

Best for: chapter 2’s end-to-end project after our data and comparison lessons; chapter 3 after classification; chapter 4 after regression.

Bring: Python familiarity and the ability to run notebooks. Work through one relevant chapter at a time.

Predict tomorrow’s demand

Use our original 180-day synthetic kitchen project. Inspect the data, fit training-only preprocessing, compare with a median baseline and evaluate on later days. The notebook has been tested end to end; its setup guide lists the expected results.

Sort a small collection

Use flower measurements from a properly attributed dataset. Compare a simple baseline with a classifier, then inspect which classes it confuses.

Write a useful model note

Explain the task, data, baseline, validation result and limits on one page. Include an example where a person should check the model’s suggestion.