OCI Data Science Pro flashcards
OCI Data Science Pro · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Project
- A compartment-scoped container in OCI Data Science that groups notebook sessions, jobs, models, deployments, and pipelines. It is the prerequisite resource for all data science work and the unit for IAM policy scoping.
- Notebook session
- An interactive JupyterLab environment backed by a chosen compute shape and block volume. It persists data on its block storage across stop and start, and must be Active before you can connect.
- Job
- A non-interactive, repeatable execution of a script or notebook on managed compute. Jobs are suited to scheduled or long-running training and write logs to OCI Logging.
- Model catalog
- A registry of model artifacts stored in Object Storage along with metadata such as provenance, training job, and framework. Catalog entries are the source for creating model deployments.
- Model deployment
- A managed HTTP endpoint hosting one or more models with a chosen shape, instance count, and bandwidth. It provides load balancing, logging, and monitoring for online inference.
- score.py
- The custom inference script a deployment runs. It must implement load_model and predict functions so the container can load the artifact and serve predictions through the endpoint.
- ADS library
- Oracle's Accelerated Data Science Python library. It wraps datasets, model training, evaluation, explainability, and deployment with higher-level APIs on top of the OCI SDK.
- Resource principal
- An IAM identity assumed by OCI resources such as jobs and deployments. It lets them call services like Object Storage and Vault without embedding user credentials, using dynamic group policies.
- Dynamic group
- An IAM construct whose membership is defined by matching rules on resource attributes. It is paired with policies to grant resource principals access to buckets, models, and secrets.
- Compartment
- The logical IAM boundary that contains projects and their resources. Policies are written per compartment, and resources are billed and tagged within it.
- Idle timeout
- A session setting that automatically stops a notebook session after a period of inactivity. Stopping preserves the block volume, while deleting the session destroys it permanently.
- Conda environment
- An isolated Python environment inside a notebook session. Custom environments install oracle-ads and other libraries, and can be captured for reproducible jobs and deployments.
- Block volume
- Persistent storage attached to a notebook session or job. It holds notebooks, data, and installed environments, and survives stop and start of a session until deletion.
- Job run
- A single execution instance of a job definition. Runs move through Accepted, InProgress, Succeeded, Failed, or Canceled states, and the run OCID is what pipelines and monitoring reference.
- Pipeline
- A directed acyclic graph of steps in OCI Data Science, such as custom script, job, and deployment steps. Each step runs in its own container and passes outputs downstream.
- Pipeline step
- A single node in a pipeline DAG. It runs a defined container command, writes artifacts to Object Storage, and its failure halts downstream steps in the run.
- Model group
- A deployment configuration hosting multiple model versions behind one endpoint. Traffic split percentages enable A/B testing and canary rollout between the member models.
- Vault secret
- A securely stored credential or configuration value in OCI Vault. Jobs and deployments retrieve secrets at runtime by OCID rather than hardcoding them in code.
- OCI Logging
- The service that captures job run logs, deployment access logs, and prediction logs. Logs are queried and retained per log group for troubleshooting and audit.
- OCI Monitoring
- The metrics and alarm service. Data Science deployments emit invocation, latency, and error metrics that can trigger alarms and notifications for operational response.
- Data Science shape
- The compute configuration (CPU or GPU, memory) selected for a session, job, or deployment. Shape choice determines cost and must satisfy model memory and concurrency needs.
- Subnet selection
- The network placement chosen when creating a session, job, or deployment. It determines whether traffic is public or private and which endpoints and databases are reachable.
- ADSData
- An ADS dataset object that loads data from Object Storage, databases, or local files. It offers profiling, splitting, and visualization helpers before model training begins.
- Model provenance
- Metadata recorded with a catalog entry describing the training job, notebook, code version, and input data. It supports reproducibility and audit of how a model was produced.
- Traffic splitting
- A deployment feature that routes a configured percentage of inference requests to each member model in a model group, enabling gradual rollout and controlled experiments.
- IAM policy
- A statement granting a group or dynamic group verbs on resource types within a compartment. Least-privilege policies scope access to data-science-family, buckets, and vaults separately.
- Predict endpoint
- The REST URL exposed by a model deployment for inference. Callers authenticate with an IAM token or resource principal and post input data to receive predictions.
- Session lifecycle states
- The states a notebook session passes through: Creating, Active, Stopping, Stopped, Updating, Deleting, and Failed. JupyterLab is reachable only in the Active state.
- Job artifact
- The script, notebook, or container image stored in Object Storage that a job executes. The job definition references it by location and runs it on the selected shape.
- OCI Python SDK
- The official Python client for OCI REST APIs. It exposes the same operations as the CLI, letting you create projects, sessions, jobs, models, and deployments programmatically.
- Budgets and alarms
- Cost and health controls in OCI. Budgets track spend per compartment and alarms watch monitoring metrics, together supporting operational governance of data science resources.
- Model deployment shape
- The instance type and count chosen for a deployment. It sets memory, CPU or GPU capacity, and replica count, directly affecting throughput, latency, and cost of inference.