Altar — Experiment Data Management

A monorepo of small tools and deployment configs to manage experiment data end‑to‑end:

• MongoDB (Sacred) for experiment metadata
• MinIO (S3-compatible) for raw/heavy files
• Omniboard for visualization
• AltarExtractor for browsing and filtering Sacred experiments
• Desktop utility for sending experiments to Sacred/MinIO

Repository structure

Altar/
├─ AltarDocker/      # Docker Compose: MongoDB, MinIO, Omniboard, Extractor + docs
├─ AltarExtractor/   # Dash app to browse, filter, and export Sacred experiments
├─ AltarSender/      # GUI to send experiments to Sacred (+ MinIO artifacts/raw data)
├─ assets/           # Website assets
├─ index.html        # Landing page
└─ README.md

Subprojects

• AltarDocker: Local stack for MongoDB + MinIO + Omniboard + AltarExtractor

  • What it provides: a reproducible local/dev environment to store and visualize experiments.
  • Start here if you need a database and storage running locally.
  • Docs:
    • Deployment: AltarDocker/DEPLOY.md
    • User management: AltarDocker/MANAGE_USERS.md
    • Compose file: AltarDocker/docker-compose.yml

• AltarExtractor: Dash web app for browsing Sacred experiments

  • What it does: connect to MongoDB, browse Sacred runs, filter by config keys, view metrics, and export data as CSV or explore in Pygwalker.
  • Can run standalone or as part of AltarDocker (via --profile extractor).
  • Get started: pip install -r requirements.txt, run python app.py, or use Docker.
  • Docs: AltarExtractor/README.md

• AltarSender: GUI to send experiments to Sacred (and MinIO)

  • What it does: select per-experiment folders, map files (config/results/metrics/artifacts/raw data), then send to MongoDB Sacred and optionally MinIO or a filesystem path.
  • Get started: create/activate a virtual env, pip install -r requirements.txt, run python app.py.
  • Docs: AltarSender/README.md

Quick start

1. Run the local stack (MongoDB + Omniboard):

   cd AltarDocker
   # Edit .env with your credentials
   docker compose up -d

2. Add the Extractor (optional):

   docker compose --profile extractor up -d

   Then open http://localhost:8050 and connect to MongoDB (host: mongo, port: 27017).

3. Add MinIO (optional):

   docker compose --profile minio up -d

4. Send experiments: Use AltarSender → python app.py and configure sources (config/results/metrics/artifacts/raw data).

URLs (default)

Service	URL
MongoDB	mongodb://localhost:27017
Omniboard	http://localhost:9004
AltarExtractor	http://localhost:8050
MinIO Console	http://localhost:9001
MinIO S3 API	http://localhost:9000

Requirements

• Python 3.x (recommended: use a virtual environment)
• pip for Python dependencies
• Docker Desktop (Windows/macOS) or Docker Engine + Compose (Linux) to run the local data stack

Useful links

• Deployment: AltarDocker/DEPLOY.md
• Manage users and access: AltarDocker/MANAGE_USERS.md
• Experiment sender docs: AltarSender/README.md
• Extractor docs: AltarExtractor/README.md

Notes

• Credentials and endpoints for MongoDB/MinIO are configured in the Docker .env file.
• Omniboard connects to your MongoDB to visualize runs and artifacts; ensure its connection string matches your database/auth setup.
• AltarExtractor can save MongoDB credentials in browser local storage for convenience.

Related Projects

• AltarDocker — Docker Compose stack for MongoDB, MinIO, Omniboard, and AltarExtractor
• AltarExtractor — Dash web app to browse and filter Sacred experiments
• AltarSender — GUI to send experiments to Sacred and MinIO

License

GNU General Public License v3.0. See LICENSE at the repository root.