The opensource tool Sacred
helps configuring, organizing, logging and reproducing experiments in Deep Learning projects. It's a Python package allowing to log data in a MongoDB
database, among others.
Install Sacred
:
pip install sacred
The documentation is hosted on ReadTheDocs.
Two particularly useful tools for Sacred
:
- The web tool Omniboard allows to visualize logged experiments within a browser.
- The web tool Mongoku allows to browse/edit a
MongoDB
database.
For the sake of simplicity and stability, all these tools may be setup using docker-compose.
# docker-compose.yml
version: '3'
services:
mongo:
image: mongo
ports:
- {IP_ADDRESS}:27017:27017
environment:
MONGO_INITDB_ROOT_USERNAME: user
MONGO_INITDB_ROOT_PASSWORD: password
MONGO_INITDB_DATABASE: sacred
expose:
- 27017
networks:
- omniboard
restart: always
volumes:
- ./mongo:/data/db
omniboard:
image: vivekratnavel/omniboard:latest
command: ["--mu", "mongodb://user:password@mongo:27017/sacred?authSource=admin"]
ports:
- {IP_ADDRESS}:9000:9000
networks:
- omniboard
depends_on:
- mongo
restart: always
mongoku:
image: huggingface/mongoku
networks:
- omniboard
ports:
- {IP_ADDRESS}:3100:3100
expose:
- 3100
environment:
MONGOKU_DEFAULT_HOST: "mongodb://user:password@mongo:27017/?authMechanism=SCRAM-SHA-1"
depends_on:
- mongo
restart: always
networks:
omniboard:
Replace {IP_ADDRESS}
with the host's IP.
The Omniboard's page is accessible on port 9000, Mongoku's page on port 3100.