Trending data/ML repos (weekly)

Weekly velocity ranking of trending data/ML GitHub repositories. Five topic-qualified searches against the official keyless GitHub REST API (https://docs.github.com/en/rest/search/search#search-repositories) collect repositories created in the trailing 7 days (topic: machine-learning, llm, deep-learning, computer-vision, data-science; top 3 star-sorted pages each), merged and deduped by repository id. Each repository is scored for star velocity — stars divided by days since creation (age floored at 1 day); for repos created inside the window this is exactly the trailing-7-day star gain — and ranked fastest first. Forks are excluded (original projects only), as are archived, disabled, and zero-star repositories; every kept repository is real and active by construction (created < 7 days ago, pushed recently, publicly listed by the GitHub search index). Columns: ISO week, fetch timestamp, repository id / full name / URL, a normalized one-line summary (authors' public description, whitespace-collapsed and truncated to 220 chars; emails stripped), owner and owner type, primary language, comma-joined topics, per-repo SPDX license id, a deterministic rule-based ML-subtopic tag (agents, llm, rag, finetuning, quantization, vision, audio, nlp, multimodal, evals, mlops, data-engineering, robotics, education, other), stars / forks / watchers / open issues, age in days, stars per day, velocity rank, creation and last-push timestamps. Primary key: (week, repo_id). Cadence: weekly; each snapshot reflects the trailing-7-day creation window at fetch time. Nullability: summary, language, license_spdx and topics may be empty when the author provided none; stars, velocity rank and stars_per_day are never null. Caveats: the topic filter is author-applied, so untagged ML repositories are missed; star counts are a popularity proxy, not a quality measure; subtopic tags are keyword rules, not a classifier. Sample use: order by velocity_rank for the week's fastest-rising ML projects, or filter ml_subtopic = 'agents'.

Les titres et les descriptions proviennent des sources de données, en anglais.