OCI Data Science Pro study plan
OCI Data Science Pro · Give each week a clear focus.
8 weeks · About 6 hours per week
Data Science Overview and OCI Data Science Concepts
Map each OCI Data Science component (projects, notebook sessions, conda environments, jobs, model catalog, model deployments, pipelines) to the lifecycle stage it serves
Distinguish when to use a notebook session versus a job versus a pipeline run, and justify the choice for three sample workloads
Explain the shared responsibility boundary between the data scientist and the OCI tenancy administrator for networking, policies, and identity
Describe how the model catalog, model deployment, and pipeline resources relate to one another in a governed workflow
Your checkpoint
A one-page architecture diagram plus a 300-word written justification of component choices for three workloads
Data Science Overview and OCI Data Science Concepts & Data Access, Pipelines, and MLOps Operations
Identify the IAM policy statements and dynamic group patterns required for data scientists, jobs, and pipelines to reach Object Storage and other services
Trace how VCN, private endpoints, and subnet choices affect access from notebook sessions and jobs to data sources
Compare Object Storage, Autonomous Database, and external sources for reading training data inside OCI Data Science
List the logging and audit surfaces (logging, audit, metrics) that support operational troubleshooting
Your checkpoint
A checklist of IAM policies, dynamic groups, and network prerequisites that you can reuse when provisioning a new project
Data Science Projects, Notebook Sessions, and Environments
Provision a project and notebook session, then configure its shape, block storage, and subnet settings deliberately
Create and version a custom conda environment, install a package, and publish it so it can be reused by jobs and pipelines
Use the OCI Python SDK or CLI from inside a notebook session to list and manage resources programmatically
Demonstrate environment reproducibility by exporting the environment specification and recreating it
Your checkpoint
A reproducible environment specification file plus a short README documenting how to recreate the session and environment
Data Science Projects, Notebook Sessions, and Environments & Model Training, Jobs, and Model Catalog/Deployment
Package a training script so it runs unchanged in a notebook session and as a job, with arguments and artifact output paths
Configure and run a job with the right shape, environment, and output location, then retrieve its artifacts
Diagnose a failed job run using its logs and status, and document the fix
Compare ADS (accelerated data science) helpers with direct SDK calls for running jobs and explain the trade-offs
Your checkpoint
A working job definition and a run log showing at least one successful run and one diagnosed failure with its root cause
Model Training, Jobs, and Model Catalog/Deployment
Register a trained model in the model catalog with provenance metadata, schema, and versioning
Deploy a model to a model deployment endpoint and invoke it with a sample payload
Explain how deployment shapes, scaling, and artifact location influence latency and cost
Update or roll back a model version and observe the effect on the live endpoint
Your checkpoint
A deployed model endpoint with a sample request/response capture and a short note on versioning and rollback behavior
Data Access, Pipelines, and MLOps Operations
Author a pipeline (YAML or SDK) with at least three steps that pass artifacts between them
Run the pipeline, inspect per-step outputs and logs, and identify where a failure would surface
Parameterize the pipeline so the same definition can run against different data inputs
Document how pipeline runs, jobs, and model catalog entries link together for traceability
Your checkpoint
A parameterized multi-step pipeline definition and a run report showing step-level outputs
Data Access, Pipelines, and MLOps Operations & Model Training, Jobs, and Model Catalog/Deployment
Automate a retrain-and-redeploy flow triggered on a schedule or by a new data artifact
Define monitoring checks for a deployed model (invocation metrics, error rates, drift considerations)
Implement a rollback path and verify it under a simulated bad model version
Record lineage from data source through job, model catalog entry, and deployment
Your checkpoint
An end-to-end MLOps runbook covering trigger, retrain, register, deploy, monitor, and rollback steps
Data Science Overview and OCI Data Science Concepts & Data Science Projects, Notebook Sessions, and Environments & Model Training, Jobs, and Model Catalog/Deployment & Data Access, Pipelines, and MLOps Operations
Rehearse scenario-style questions on component selection, environment reproducibility, job troubleshooting, and policy design
Close gaps by re-reading notes on the two weakest domains and re-running one hands-on task from each
Time yourself on a mixed set of practice scenarios and review every wrong answer to its root concept
Write a condensed one-page reference of commands, policy patterns, and decision rules for final review
Your checkpoint
A scored practice run with a written gap analysis and a one-page final reference sheet
Milestones
Week 2: You can state from memory the IAM policies, dynamic group, and network prerequisites needed to let a job read training data from Object Storage
Week 4: You have run a job successfully and diagnosed at least one failure from logs alone, without outside help
Week 6: You have executed a multi-step pipeline end to end and can explain how artifacts pass between steps
Week 7: You have completed a retrain, register, deploy, and rollback cycle and can describe each traceability link
Week 8: You scored at least 80 percent on a mixed timed practice set and have a written plan for your two weakest areas