README.md

metadata

dataset_info:
  features:
    - name: repo_name
      dtype: string
    - name: language
      dtype: string
    - name: created_at
      dtype: string
    - name: description
      dtype: string
    - name: description_language
      dtype: string
    - name: description_language_score
      dtype: float64
    - name: license_key
      dtype: string
    - name: forks_count
      dtype: float64
    - name: watchers_count
      dtype: float64
    - name: size
      dtype: float64
    - name: last_pr_id
      dtype: int64
  splits:
    - name: full
      num_bytes: 5567563250
      num_examples: 40058194
    - name: sample
      num_bytes: 138717361
      num_examples: 1000000
  download_size: 2298358914
  dataset_size: 5706280611
configs:
  - config_name: default
    data_files:
      - split: full
        path: data/full-*
      - split: sample
        path: data/sample-*
license: mit
tags:
  - code
size_categories:
  - 10M<n<100M

📊 Metadata for 40 million GitHub repositories

A cleaned, analysis-ready dataset with per-repository statistics aggregated from GH Archive events: stars, forks, pull requests, open issues, visibility, language signals, and more. Column names mirror the GH Archive / GitHub API semantics where possible. GitHub repo: https://github.com/ibragim-bad/github-repos-metadata-40M

Source: GH Archive (public GitHub event stream).

🚀 Quickstart

from datasets import load_dataset

ds = load_dataset("ibragim-bad/github-repos-metadata-40M", split="full")  # or 'sample' for random sample of 1 million repositories.

📈 EDA Notebook

• data_visualisation.ipynb: Simple exploratory data analysis of the dataset (column overview, basic distributions, missing-value checks, sample visualizations).

💡 Use cases

• Trend analysis of stars/forks/PRs
• Language mix & repo growth signals
• Ranking by activity / stability
• Input features for ML models (quality or popularity prediction)

📦 What’s inside

Each row aggregates repository-level stats derived from GH Archive events and GitHub metadata snapshots.

Schema (columns & types)

Column	Type	Description
repo_name	string	owner/name
language	string	Primary language (human-readable)
created_at	timestamp	Repo creation time (UTC) if available
description	string	Repo description
description_language	string	Detected natural-language code of the repository description (ISO 639-1/2)
description_language_score	float32	Confidence score for description_language (0–1)
license_key	string	SPDX-like license key
forks_count	int64	Current forks count
watchers_count	int64	Watchers/stars count (note: GitHub’s legacy “watchers” vs “stargazers”)
size	int64	Repo size (KB, as reported by GitHub)
last_pr_id	int64	Latest available PR id

Values may be null when unavailable from the source.

🏗️ How it’s built (ETL)

• Data source: GH Archive public event stream; window through 2025-07-23 (UTC).

• Event filter:

  • CreateEvent (ref_type=repository) to capture initial repo creation and timestamps.
  • PullRequestEvent to obtain rich repo snapshots embedded in PR payloads.

• Extraction:

  • Parse event.repo and payload.*.repo snapshots for: repo_name (owner/name), description, language, license_key, size, visibility, created_at, open_issues_count, forks_count, watchers_count.
  • Track latest observed PR identifier as last_pr_id.

• Aggregation:

  • Group by repo_name; keep the most recent snapshot within the window.

Notes

• Coverage primarily comes from PullRequestEvent snapshots; repositories without any PRs are generally missing.
• Stars/watchers and forks are taken from event-time snapshots and may be slightly stale versus live GitHub.
• Values can be null if not present in the underlying event payloads.

⚠️ Limitations

• GH Archive reflects public events only; private repos are out of scope.
• Some GitHub fields (e.g., watchers vs stargazers) have historical quirks.
• Missing values where source is unavailable.

📚 Citation

If you use this dataset, please cite:

@misc{github_repos_activity_stats_2025,
  title        = {GitHub Repos Activity Stats (from GH Archive)},
  author       = {Ibragim Badertdinov},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/ibraigm-bad/github-repos-metadata-40M}}
}

📜 License

Distributed under the MIT License. See LICENSE for more information.