bpesquet/mlhandbook

(Phased out) My textbook for teaching Machine Learning

Machine Learning Handbook

This project is being phased out and replaced by ainotes.

This repository contains the source files for the Machine Learning Handbook, my textbook for teaching Machine Learning. More info here.

Development notes

Generating the site

The website is generated by Jupyter Book. After installing it, execute the following command in the root folder to generate the HTML output in the _build subdirectory:

jupyter-book build .

A GitHub action is used to publish this output as a website.

Adding or updating katas

The practical challenges, named katas, are generated with nbgrader and contain missing answers. Teacher versions (corrections) live in a separate private repository.

The mlkatas repository is outdated. Its content has been merged into this repository.

Exporting a notebook to a PDF file

Use nbconvert in the same directory as the source file:

jupyter nbconvert --to PDF <chapter_file_name>.ipynb

For now, GIF files must be replaced by their PNG counterparts before launching the conversion process.

mlhandbook's Issues

CNN chapter

• Hide useless cells
• Filters viz
• Bias in conv filters

Style transfer chapter

• Add math formulas for losses

K-NN chapter

• Add definitions
• Improve examples
• Implements TODOs

Logistic regression chapter

• Gradient math proof (source)
• Colors in multiclass decision boundaries

SVM chapter

Intro to ML chapter

• Add definition of learning

Assessing results chapter

• Better ROC/AUROC explanation (https://stats.stackexchange.com/questions/132777/what-does-auc-stand-for-and-what-is-it, https://stats.stackexchange.com/questions/180638/how-to-derive-the-probabilistic-interpretation-of-the-auc)

K-Means chapter

Transformers chapter

Autoencoders chapter

• SELU plot
• Rounded accuracy plot in training history
• Better VAE code explanations

Bayes chapter

Linear regression chapter

• Math output problem with Rise
• Plotly in Colab
• Improve polynomial and regularization code

ANN chapter

• Hide useless cells
• Universal approx theorem
• SELU and leaky RELU
• GD optim viz (source)
• Backprop demo
• Batch norm
• Geometric intuitions (DMxUCL 2020 l2, Ng, Olah)

DT & RF chapter

• Hide useless cells
• Boosting
• Add voting example
• Move ensemble chapter to a dedicated notebook ?

Keras chapter

• RNN layers
• Reshape layer
• GAN layers
• Add examples for all layers
• Functional API