---

name: ml-glossary description: "Routes self-contained ML Glossary knowledge for machine-learning concepts, cheat-sheet authoring, Sphinx documentation maintenance, and educational code-snippet caveats." disable-model-invocation: true metadata: disco-role: operating license: MIT

ML Glossary Repo Skill

Use this self-contained runtime skill for tasks about ML Glossary / ML Cheatsheet content: explaining machine-learning terms, building beginner-friendly cheat sheets, comparing algorithms, maintaining glossary-style RST documentation, or triaging educational code snippets from the project lineage. The runtime contains distilled domain knowledge from the repository; do not reopen the original docs/, code/, or notebooks/ directories to answer ordinary user questions.

First decision

1. Identify whether the user wants a concept explanation, an algorithm comparison, a neural-network workflow, a resource recommendation, or documentation maintenance.
2. Read the closest route below and only then open the linked bundled references or sub-skill.
3. If the task is about editing a live checkout, use this runtime for style, terminology, and warning triage; only the user's active checkout should be edited, and the original source checkout used to generate this skill is not required.

Route map

User intent	Read next	Why
Term lookup, beginner glossary, feature/label/model/loss/epoch/overfit concepts	sub-skills/basics-and-math/SKILL.md	Owns glossary, notation, linear/logistic regression, calculus, linear algebra, probability/statistics basics, gradient descent, and loss bridges.
Derivations or formulas for MSE, gradient descent, sigmoid/log-loss, matrix dimensions, chain rule	sub-skills/basics-and-math/references/formula-cheatsheet.md	Compact self-contained formula and code-reference map.
Decision trees, ID3/C4.5/CART, KNN, random forests, boosting, SVM, regression variants, clustering/RL placeholder status	sub-skills/classical-algorithms/SKILL.md	Owns non-neural algorithm families and their repo-specific caveats.
Neural network concepts, weighted inputs, layers, activation/loss/optimizers, forward/backpropagation, regularization, CNN/RNN/GAN/VAE/autoencoder/MLP examples	sub-skills/neural-networks/SKILL.md	Owns neural-network conceptual and architecture workflows.
Dataset, library, paper, course, blog, application, or learning-resource selection	references/resources-catalog.md	Distills the repository's large resource catalogs into usable categories and representative picks.
Site map, source-content ownership, or what each original content family represented	references/site-map.md	Provides a self-contained map of the docs/code/notebook evidence without requiring the original files.
Contributing, RST style, Sphinx build setup, docs preview, or glossary entry authoring	references/site-maintenance.md and references/troubleshooting.md	Captures contribution style, Sphinx dependencies, build caveats, and legacy warning patterns.
Staleness or refresh checks	references/repo-provenance.md	Records the source commit, branch, dirty state, and evidence baseline.

Bundled scripts

• scripts/build_docs.sh is a safe Sphinx helper. Run bash scripts/build_docs.sh --self-test to validate that Sphinx can build a tiny bundled RST project. When actively maintaining a user-provided checkout, pass --docs-dir PATH_TO_DOCS and --build-dir PATH_TO_OUTPUT; the script never assumes the generated source checkout still exists.
• scripts/check_runtime_links.py checks that Markdown links inside this runtime resolve within the skill tree or are explicit external URLs. Use it after editing runtime references.
• Sub-skill scripts provide educational, self-contained code examples: sub-skills/basics-and-math/scripts/linear_logistic_demo.py, sub-skills/classical-algorithms/scripts/knn_demo.py, and sub-skills/neural-networks/scripts/activation_loss_demo.py.

Operating rules for future agents

• Treat this as an educational reference and maintainer guide, not an installable Python library. The repository has no package metadata or public Python distribution.
• Prefer plain-language explanations with formulas and toy examples. The original project prioritized concise explanations, citations, visuals, code snippets, and equations.
• Do not claim the legacy code/ examples are production-ready. Several snippets use Python 2 print syntax, missing imports, old APIs, or partial pseudocode. Use bundled scripts when a runnable toy example is needed.
• If asked to edit glossary content, keep entries short, accessible, cited, and supported by visuals/equations/code when practical. Use RST/Sphinx conventions from references/site-maintenance.md.
• If asked to verify a docs build, do not run project tests, lint, or formatters unless the user separately requests them. A Sphinx build is the relevant repository-native check.
• If a user asks for modern best practices beyond the repository content, separate repo-grounded ML Glossary facts from modern additions and label the latter clearly.

Known scope limits

• Many applications, clustering, training, and some reinforcement-learning subsections were placeholders in the source material. This runtime records that status rather than inventing full coverage.
• The original docs used many images; this runtime distills their teaching intent in text so the original image files are not required.
• External URLs in resource references are optional leads. The explanations and routing needed to answer common tasks are included inside this runtime.