---
license: mit
task_categories:
  - text-generation
  - text-classification
language:
  - en
tags:
  - git
  - commits
  - code
  - software-engineering
size_categories:
  - 10K<n<100K
---

# GitGraph Training Data

This dataset contains commit data extracted from various Git repositories for training the GitGraph model.

## Contents

- `gitgraph.duckdb` - DuckDB database containing:
  - Commit metadata (hash, message, author, timestamp)
  - File changes with diffs
  - Repository information

## Usage

```python
from huggingface_hub import hf_hub_download
import duckdb

# Download the database
db_path = hf_hub_download(
    repo_id="blankenshipdanielj/gitgraph-data",
    repo_type="dataset",
    filename="gitgraph.duckdb"
)

# Query the data
conn = duckdb.connect(db_path, read_only=True)
commits = conn.execute("SELECT * FROM commits LIMIT 10").fetchdf()
print(commits)
```

## Schema

### commits table
- `commit_hash` (VARCHAR): Unique commit identifier
- `repository_id` (VARCHAR): Repository identifier
- `message` (VARCHAR): Full commit message
- `message_subject` (VARCHAR): First line of commit message
- `author_name` (VARCHAR): Author name
- `author_email` (VARCHAR): Author email
- `authored_at` (TIMESTAMP): Authoring timestamp

### file_changes table
- `id` (VARCHAR): Unique file change ID
- `commit_hash` (VARCHAR): Associated commit hash
- `file_path` (VARCHAR): Path to changed file
- `change_type` (VARCHAR): Type of change (added, modified, deleted)
- `diff_content` (VARCHAR): Unified diff content

## License

MIT License