OCI Data Science Pro: Data Science Projects, Notebook Sessions, and Environments

Domain 2 - Projects, Notebook Sessions, and Environments

This domain tests the working environment: how projects scope resources, how notebook sessions are shaped and networked, and how conda environments make work reproducible across notebooks and jobs.

Core rules

• A notebook session is a managed compute instance with block storage, a shape, a subnet, and an optional private endpoint. Stopping preserves block storage; terminating can delete it.
• Shapes can be fixed or flexible. Pick CPU for data prep and small models; GPU shapes for deep learning. Block storage is separate from the shape and must be sized for datasets and caches.
• Service-managed conda environments are versioned and maintained by OCI. For extra libraries, clone/publish a custom environment to Object Storage, then import it as a project environment.
• Environment selection must match between notebook and job. A package installed only in an interactive session will not appear in a job run.
• Use Vault for secrets and environment variables for non-secret configuration; do not bake credentials into notebooks.

Worked example

A notebook needs xgboost at a pinned version and the same code must run as a job. Create a custom conda environment from a service-managed base, install and pin the package, publish the environment archive to an Object Storage bucket, import it into the project, and select it in both the notebook session and the job. Put the bucket URI and model output URI in job arguments, not hardcoded paths.

Traps

• Terminating a notebook session before copying work off its block storage.
• Using a public subnet when the security policy requires a private subnet and private endpoint.
• Assuming pip install in a running session is reproducible for jobs.
• Forgetting that a notebook session and job need network paths to Object Storage.