No. 73 of 500 No. 5 in Hands-On AI & Programming

Deep Learning with Python

By François Chollet, Tomasz Kalinowski and J. J. Allaire

Print length
648 pages
Publisher
Manning
Reading level
Intermediate
ISBN-13
978-1633436589
Published
November 18, 2025
Written for
Python programmers and data scientists who want to build and train modern deep learning models.

Synopsis

First published in 2017 and now in a third edition (2025) coauthored with Google engineer Matthew Watson, Deep Learning with Python teaches neural networks by building them. François Chollet begins with what deep learning is and the tensor operations and gradient descent beneath it, introduces TensorFlow, PyTorch, JAX and Keras side by side, and lays out a universal workflow for framing a problem, choosing a metric, beating a baseline and controlling overfitting. Later chapters apply that workflow to image classification, segmentation and object detection, timeseries forecasting, text classification, and a chapter on language models and the Transformer that leads into building a small GPT-style text generator and diffusion-based image generation.

The method avoids mathematical notation, explaining each principle in prose and pairing it with runnable Keras 3 code that works on TensorFlow, PyTorch or JAX backends. Chollet adds chapters on best practices for real-world training and on the future of AI, where he contrasts the local generalization of deep networks with the abstraction and reasoning humans use, a theme he has pursued through the ARC benchmark. Readers who finish the book can train, debug and scale modern models and understand why the techniques work. The full third edition is free to read online at the authors' site, with print and ebook editions sold by Manning.

Scorecard

Editorial score

How it scored

Editorial score: 86.3 / 1005 weighted criteria

  • Influence30%

    25.5/30
  • Authority25%

    22.5/25
  • Practical value20%

    17.0/20
  • Currency15%

    12.8/15
  • Readability10%

    8.5/10
  • Editorial score

    86.3/100

How the rank was computed

Rank No. 73 of 500 comes from a composite of 65.3: observable signals 26.5 of 55 points plus editorial review 38.8 of 45 points.

All 11 inputs
InputPoints
Observable signals26.5/55
Reach (Amazon ratings volume)6.2/16.5
Consensus across best-of lists8.3/13.8
Awards and bestseller evidence3.3/8.3
Reader rating, sample-adjusted4.8/8.3
Amazon sales rank1.1/5.5
Recency2.8/2.8
Editorial review38.8/45
Author authority10.1/11.3
Influence11.5/13.5
Practical value7.7/9.0
Readability3.8/4.5
Currency5.7/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

Among the clearest code-first introductions to deep learning, written by Keras's creator and rebuilt in 2025 for transformers and generative models.

Strengths

  • Written by the creator of Keras
  • Runnable code for every concept
  • Free to read online in full

Keep in mind

Light on mathematics by design; readers who want derivations, proofs or research depth will need a companion textbook such as Bishop's or Goodfellow's.

Key takeaways

  1. Deep learning reduces to chains of tensor operations whose weights are tuned by gradient descent, a framing that makes every later architecture easier to understand.
  2. A universal workflow, from defining the task and success metric to beating a baseline and then regularizing, structures every project in the book.
  3. Keras 3 runs on TensorFlow, PyTorch or JAX, so the same model code can move between backends as tooling choices change.
  4. Deep networks generalize locally from their training data; Chollet argues that humanlike intelligence requires abstraction and reasoning that current models still lack.

Notable facts

About the author

François Chollet, Tomasz Kalinowski and J. J. Allaire

François Chollet created the Keras deep learning library, which Google has reported at more than two million users, and spent nearly a decade at Google before leaving in November 2024 to co-found the AGI research lab Ndea. He also created the ARC-AGI benchmark and co-launched the ARC Prize. Third-edition coauthor Matthew Watson is a Google software engineer working on Gemini and a core Keras maintainer.

Questions

Is Deep Learning with Python worth reading in 2026?
Yes. The 2025 third edition was rewritten around Keras 3 and adds chapters on transformers, building a GPT-style language model and diffusion-based image generation, alongside coverage of PyTorch and JAX. That keeps it current for practitioners, and the full text can be read free at deeplearningwithpython.io before deciding whether to buy a print or ebook copy from Manning.
How much math do I need for Deep Learning with Python?
Very little. The authors avoid mathematical notation and explain ideas through runnable code and plain prose; the book's site says high-school mathematics is sufficient. Reasonable Python proficiency is the real prerequisite. Readers who later want formal derivations can pair it with a theory text such as Pattern Recognition and Machine Learning or Deep Learning by Goodfellow, Bengio and Courville.
Which edition of Deep Learning with Python should I read?
Read the third edition, published by Manning in 2025 and coauthored with Matthew Watson. Earlier editions, the first of which appeared in 2017, predate Keras 3 and lack the newer chapters on language models, text generation and image generation. The book's official site describes the third edition as a comprehensive overhaul rather than a light refresh.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.