The Deep Learning bible, rewritten in raw NumPy

What it does

This repository is a chapter-by-chapter companion to Goodfellow, Bengio, and Courville’s Deep Learning (the “花书”). For each concept, the author re-derives the mathematical principles and implements them using only Python and NumPy—no TensorFlow, PyTorch, or even scikit-learn. The result is a set of annotated PDFs and a code folder meant to be read side-by-side with the theory.

The interesting bit

The stubborn minimalism is the point. Where most educational repos lean on frameworks, this one builds everything from linear algebra up, treating NumPy as the only dependency. The author also goes beyond the book’s scope, adding sections on XGBoost, Gaussian process regression, and quasi-Newton methods that the original text doesn’t cover in the same depth.

Key highlights

• Implements concepts from linear algebra through CNNs and optimization using only raw NumPy
• Pairs mathematical derivations with source-level code in PDF chapters for side-by-side reading
• Explicitly bans deep-learning frameworks; scikit-learn appears only as a correctness check
• Adds supplementary material not in the original book: ensemble methods (Random Forest, AdaBoost, GBDT, XGBoost) and Bayesian optimization
• Released as a continuously updated Chinese-language PDF alongside the repository

Caveats

• Several chapters listed in the table of contents (including 10, 12, and 13–20) have not yet been uploaded
• The author openly notes their own skill limitations and actively requests error reports via Issues or email
• All explanatory material and PDFs are in Chinese, which may limit accessibility for non-Chinese readers

Verdict Worth bookmarking if you are a Chinese-speaking student who wants to see the Deep Learning textbook brought to life in raw NumPy. Skip it if you need a quick framework-based tutorial or if you don’t read Chinese.