scaredboard-deploy provides a dockerized reference deployment setup of the sacredboard dashboard for the sacred ML experiment management framework.

Docker and Docker Compose are required. Although the setup only has been explicitly tested against Docker version 18.06.1-ce, most other recent Docker versions should be supported too.

A custom Docker image for sacredboard is defined in Dockerfile.sacredboard. This image may also be used independently of the remaining setup, e.g. with a custom MongoDB instance. For this purpose, the MongoDB url and database name with which sacredboard is started can be configured using the MONGODB_URL and MONGODB_DATABASE environment variables.

docker-compose.yml defines two services. A sacredboard container based on the provided sacredboard image, publishing on the 5000 port and a MongoDB instance with persistent data storage in a volume, publishing on the host port 27017.

To clone the project and start the sacredboard stack in detached mode, the following commands must be executed:

git clone https://github.com/kklemon/sacredboard-deploy.git
cd sacredboard-deploy/
docker-compose up -d

When the containers are spun up, sacredboard should be available on localhost:5000.

To save results of sacred experiments into the provided MongoDB instance and make them visible in sacredboard, a MongoObserver has to be added to an experiment, configured with localhost:27017 as url and sacred as database name. An exemplary experiment configuration may look like as follows:

from sacred import Experiment
from sacred.observers import MongoObserver

ex = Experiment('my-experiment')
ex.observers.append(MongoObserver.create(url='localhost:27017',
                                         db_name='sacred'))

Alternatively, an MongoObserver can be configured through the command line by running the experiment with the following option:

./my_experiment.py -m localhost:27017:sacred