Cheat Sheet: Machine Learning

The essentials of all 27 lessons, F1 to E5, on eight landscape A4 pages (four sheets, printed on both sides):

• Pages 1 to 3, the supervised workflow step by step: both workflows at the top of page 1 (supervised, and unlabeled with text), then, for each supervised step, what to decide, the formulas, the code that recurs in the labs, the traps and a quick "symptom → cause → what to do" diagnosis. It runs from framing the problem and splitting the data to explaining the model and putting it in production.
• Page 4, unlabeled data and text: dimensionality reduction, clustering (k-means, DBSCAN, HDBSCAN and hierarchical), text representation (bag-of-words, TF-IDF and embeddings) and topic modeling (LDA, NMF and BERTopic).
• Pages 5 and 6, the models one by one: from the baselines to gradient boosting and neural networks (page 5), and from the unsupervised methods to the topic models (page 6). How each one predicts or groups, what it learns, its knobs (with the scikit-learn name), where it shines and where it misleads.
• Pages 7 and 8, lesson by lesson: each lesson's question, what to take away from it and its most common trap.

The colors are the same as in the workbook, the concept map and the study cards: gray for the problem and delivery, blue for data and splitting, purple for the model, orange for evaluation and red for the traps.

Keep it at hand while you work through the workbook or the project. To review, the concept map shows how the ideas connect and the study cards train your memory.

Download the cheat sheet (PDF) Concept map Study cards

If the PDF does not show above (some phones do not display PDFs inside a page), use the Download button.