Snapshot experiment data and reproduce experiments

Windocks provides the ability for reproducing experiment data used in an experiment run. This functionality is available for SQL Server and Oracle in the Enterprise Plus edition of Windocks. A data scientist conducting experiment runs can instantly reproduce tha data used in a successful run by snapshotting the data into a new image in seconds (no matter how big the data).

Start by creating a container from an image that has experimental data (See this to build the image and model. Use the web app Images from top menu or REST API to create a container with data or run the command below:

docker run -d yourImageName

Make any needed changes to the data in the container for your experiment and run the experiment (See the tutorial of creating a Python model with Windocks machine learning datasets on demand here). When data changes are complete and after the experiment run, commit the container with the data. Use the web app (Go to Containers and clones, find your container, type in a new image name and click Snapshot). Or use the docker commit REST API. Or run the command line below.

docker commit YourContainerId FreshImageName

Create containers from the fresh image for other team members so they can reproduce the experiment easily. Use the web app or REST API to create a container with data or run the command below:

docker run -d FreshImageName

This container will now have the exact same data as was used in the experiment run.