Machine Learning Glossary - https://machinelearning.wtf/

About

This is a website to build an online glossary for machine learning terms--with an emphasis on brief, concise descriptions.

Development

Most of the website's content is in Pandoc Markdown files. LaTeX mathematics is rendered with MathJax.

This site is built by Netlify and deployed to https://machinelearning.wtf/

If you want to preview your changes, there are three ways (listed in order of easiest to hardest):

• Open a pull request against this repo. Netlify will build the pull request and host it on a staging subdomain.
• Use the included Docker image. Install Docker, run ./build.sh from this repo's root directory to build the container, and then run ./serve.sh to to build and deploy the site on your computer at http://localhost:4000.
• Install Jekyll on your computer and run jekyll serve in this repo's root directory.

Writing

Terms should be written in Markdown, as that is what the templates are assuming they're dealing with.

A typical YAML front matter for a given term might look like:

---
title: Term Title Here
related_terms:
 - some-term-filename-without-extension
 - other-term
references:
 - link_title: A description of a website
   link_url: https://google.com
 - link_title: "Use quotes when titles have characters: like colons"
   link_url: https://arxiv.org/abs/1506.01497
---

Titles

The title refers to the title of the term.

Terms with acronyms should have the acronym following the term:

Generative Adversarial Network (GAN)
Harmonic Precision-Recall Mean (F1 Score)
Kullback-Leibler (KL) Divergence

Acronyms are then found and collected at /meta/acronyms.

The filenames always include the acronym at the same place as the title:

generative-adversarial-network.md
harmonic-precision-recall-mean-f1-score.md
kullback-leibler-kl-divergence.md

• Filenames are always lowercase.
• Words are separated by the dash (-) symbol.
• Files are always written in Markdown with the .md extension.

Metadata

The related_terms are the filenames (without directory path or extension) of related terms within the glossary. Each term's page will list the related terms in its Markdown file and all references to the term from other terms' related_terms lists.

Links external to the glossary can be placed in the references with a link_title and a link_url.

For terms that have been filled out, but need further review for accuracy and cleanup, please add needs_review: true to the YAML front-matter.

Redirects

For when two terms are synonyms, both terms should be added, but one of them as a redirect to another. The entire content of the Redirect term should have a title with the term title, layout: redirect to use the redirect template, and destination to mark the filename of the term (without extension) to redirect to. An example is below:

---
title: Latent semantic analysis (LSA)
layout: redirect
destination: latent-semantic-indexing-lsi
---

Philosophy

1. Write about terminology, not people.
2. Keep it simple and short. Link elsewhere for the most advanced material, and avoid excessive expository writing.
3. Cover common terms, but focus on obscure terms that appear in a handful of papers, but won't make into Wikipedia.
4. Use mathematics when it makes definitions more concise, but not when it makes definitions more confusing than images or examples.
5. MLG should be easy to read, and easy to write.

Todo

1. Build a linter that checks for:

• broken internal links
• broken external links
• trailing spaces before newlines
• forgetting to put a newline before bullets or other Markdown syntax
• common misspellings
• uncompilable LaTeX code

2. Build a better user interface.

3. Integrate Algolia search.