学机器学习最常见的尴尬是：教程要么只讲概念，听完还是一头雾水；要么直接甩代码，跑得动却说不清为什么。

我今天在 GitHub 挖到一个很值得收藏的开源项目：Machine Learning Visualized。它把算法“讲明白”的方式很直接——用可视化把每一步训练过程摊开给你看，让原理不再停留在文字和公式里。

复杂算法被拆成直观的动态图流程，配上完整的数学推导和交互式界面，你能清晰看到训练中权重如何更新、如何一步步收敛到更优解。

GitHub：https://github.com/gavinkhung/machine-learning-visualized…

你能得到什么：

• 神经网络、逻辑回归、感知器等核心算法的完整实现
• 从第一性原理出发的详细推导，公式不跳步、思路讲清楚
• 交互式 Marimo 笔记本，参数随调随看效果
• 训练过程可视化，收敛路径与权重变化一目了然
• 覆盖 PCA、K-means、梯度下降等常用方法

项目基于 Jupyter Book 构建：既可在线浏览学习，也支持 Docker 本地一键部署，拉起就能用。

gavinkhung/machine-learning-visualized

Source: https://github.com/gavinkhung/machine-learning-visualized

Machine Learning Visualized

URL: https://ml-visualized.com/

Machine Learning Visualized is a Jupyter Book containing Jupyter Notebooks that implement and mathematically derive machine learning algorithms from first-principles.

There are also Interactive Notebooks built with Marimo that allow you to see how the weights influence the loss functions.

The output of each notebook is a visualization of the machine learning algorithm throughout its training phase, ultimately converging at its optimal weights.

There is a separate Github Repository for each machine learning algorithm. Thus, this repository is simply the code to configure and build the Jupyter Book. At a very high level, Jupyter Books allow you to build a website with Markdown files and Jupyter Notebooks. Notice that none of the Jupyter Notebooks are in this repository. There is a SH script to download the relevant Jupyter Notebooks from other Github Repos. Once that is complete, the Jupyter Book can be built. The website is updated using the GitHub Action at .github/workflows/ci.yml after every commit or pull request. To build the website locally, see the Usage section below.

Jupyter Notebooks

• Neural Networks Repo
• Logistic Regression Repo
• Perceptron Repo
• Principal Component Analysis Repo
• K Means Repo
• Gradient Descent Repo

Jupyter Book Info

Table of Contents and structure of the book is specified at _toc.yml.

Configuration is specified at _config.yml.

For more information, check out the Jupyter Book Docs.

Usage

Step 1: Download the Jupyter Notebooks

chmod +x ./download_notebooks.sh
./download_notebooks.sh

Step 2: Building the Jupyter Book

Option 1: jupyter-book CLI

pip install -U jupyter-book
jupyter-book build .

Option 2: Docker Compose

docker compose run --rm jupyter-book
docker compose down --remove-orphans --volumes --rmi local

Option 3: Docker

docker build -f Dockerfile.book -t jupyter-book .
docker run --rm -v "$(pwd)":/usr/src/app jupyter-book

docker stop jupyter-book
docker rm jupyter-book
docker rmi jupyter-book

Step 3: Open the Jupyter Book

Navigate to _build/html/index.html

Build EPUB (NEW)

brew install --cask mactex
nbmerge $(ls chapter1/*.ipynb chapter2/*.ipynb chapter3/*.ipynb chapter4/*.ipynb | sort) -o book/combined.ipynb
jupyter nbconvert --to latex book/combined.ipynb

docker build -f Dockerfile.pandoc -t my-pandoc .
docker run --rm -v $(pwd):/data my-pandoc pandoc book/combined.tex -o book/combined.epub --mathml --embed-resources --standalone

Output

Marimo Interactive Notebooks

Mathematically Explained