explaingit

oreilly-japan/deep-learning-from-scratch

★ 4,753Jupyter NotebookAudience · researcherComplexity · 1/5LicenseSetup · easy

TLDR

Source code for a Japanese book that teaches deep learning by building neural networks from scratch in Python, without using any machine learning frameworks, chapter by chapter, from fundamentals up.

Mindmap

mindmap
  root((repo))
    What it does
      Chapter-by-chapter code
      Neural nets from scratch
      No ML frameworks
    Tech Stack
      Python
      NumPy
      Matplotlib
      Jupyter Notebook
    Use Cases
      Learn backpropagation
      Study convolutions
      Run in SageMaker free
    Audience
      ML beginners
      Japanese readers
      Students
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Things people build with this

USE CASE 1

Work through each chapter alongside the book to understand how neural networks learn at a foundational level.

USE CASE 2

Run the notebooks in a browser for free using Amazon SageMaker Studio Lab, with no local installation required.

USE CASE 3

Study the from-scratch implementations to understand backpropagation, convolutions, and optimization without library abstractions.

Tech stack

PythonNumPyMatplotlibJupyter NotebookAWS SageMaker

Getting it running

Difficulty · easy Time to first run · 30min
MIT license, use freely for any purpose including commercial, just keep the copyright notice.

In plain English

This repository holds the companion source code for the Japanese-language book "Deep Learning from Scratch" (ゼロから作る Deep Learning), published by O'Reilly Japan in 2016. The book teaches how neural networks and deep learning work by building them from the ground up in Python, without relying on high-level machine learning frameworks. The code is organized into folders matching the book's chapters, from chapter 1 through chapter 8, plus shared utility code and dataset handling. Readers work through each chapter's folder alongside the text. Running the examples requires Python 3, NumPy (a library for numerical calculations), and Matplotlib (a library for drawing charts). The README is written in Japanese, as the book is aimed at a Japanese-speaking audience. For readers who do not want to set up a local Python environment, the repository provides direct links to run each chapter's notebooks on Amazon SageMaker Studio Lab, a free cloud computing environment from AWS. You sign up with an email address and can run the code in a browser without installing anything locally. The code is released under the MIT license, meaning it can be freely used for personal or commercial purposes. An errata page on the repository's wiki tracks corrections to the book's printed text, and errors not listed there can be reported to O'Reilly Japan directly by email. This is a reading companion, not a standalone tool. Its value is in following the book chapter by chapter to understand how deep learning algorithms work at a foundational level.

Copy-paste prompts

Prompt 1
I am reading Deep Learning from Scratch by O'Reilly Japan, chapter 4. Show me how the gradient descent update rule works in Python using only NumPy, matching the book's from-scratch style.
Prompt 2
Walk me through the backpropagation implementation from the deep-learning-from-scratch repo and explain what each calculation step computes.
Prompt 3
How do I set up the deep-learning-from-scratch environment locally with Python 3, NumPy, and Matplotlib on macOS?
Prompt 4
Explain the difference between the neural network implementations in chapter 3 and chapter 5 of this repo and what new concept each one adds.
Open on GitHub → Explain another repo

← oreilly-japan on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.