AltarDocker
Docker Compose stack for MongoDB, MinIO, Omniboard, and AltarExtractor for local/dev.
Who is Altar for?
Altar is designed for researchers and engineers who have experiment results — containing metadata, time series, images, or any files — organized in folders, and want to:
- Centralize experiment data in a structured database
- Filter and sort results by configuration parameters
- Access and visualize data through a user-friendly interface
- Produce datasets for further analysis or machine learning
Whether you're running scientific experiments, ML training runs, or lab measurements — if your data lives in folders and you need a better way to explore it, Altar can help.
What you can do
What's inside
AltarExtractor
Dash web app to browse Sacred experiments, filter by config, view metrics, and export as CSV or explore in Pygwalker.
AltarSender
GUI to map experiment folders and send runs to Sacred (MongoDB) and artifacts to MinIO.
Quick start
1
Run the local stack
Go to AltarDocker, edit .env, then docker compose up -d. Omniboard connects to your MongoDB.
2
Add the Extractor (optional)
Run docker compose --profile extractor up -d to add AltarExtractor. Open localhost:8050 and connect to MongoDB.
3
Send experiments
Use AltarSender to map folders and push runs to Sacred and files to MinIO.
4
Explore and iterate
Visualize with Omniboard, browse and filter with AltarExtractor, download datasets as CSV.
Why this repo?
Keep experiment data organized, reproducible, and explorable with a small set of focused tools. Use only what you need; each subproject stands on its own.
- Pragmatic: minimal, task‑oriented tools
- Portable: local dev stack in Docker
- Observable: Omniboard and AltarExtractor for runs and metrics
- Adaptable: desktop utilities and web apps