I85PREP
⌘ K
OCI Data Science Pro study materials
OCI Data Science Pro study resources

OCI Data Science Pro practice questions

One question at a time. Build your OCI Data Science Pro understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Data Science Overview and OCI Data Science Concepts

In OCI Data Science, what is the primary role of a project?

  1. A compute shape used only for model deployment
  2. A logical container that organizes related notebook sessions, jobs, and models Correct answer.
  3. A Git repository hosted inside the tenancy
  4. A billing compartment that replaces all IAM policies

Explanation

A project is the top-level organizational unit in OCI Data Science that groups notebooks, jobs, model artifacts, and related resources so access and lifecycle can be managed together. Compartments still govern IAM, so a project does not replace them.

Question 2Data Science Overview and OCI Data Science Concepts

Which OCI Data Science resource provides an interactive JupyterLab environment for exploratory work?

  1. Notebook session Correct answer.
  2. Pipeline run
  3. Model deployment
  4. Data Flow application

Explanation

A notebook session provisions a managed JupyterLab instance on a chosen compute shape, which is the standard environment for interactive data exploration and model development. Jobs and deployments are for batch and serving workloads, not interactive coding.

Question 3Data Science Overview and OCI Data Science Concepts

A data scientist needs to run a long training script nightly without keeping an interactive session open. Which OCI Data Science capability fits best?

  1. A notebook session left idle overnight
  2. A model catalog entry
  3. A job run configured with the training script and a compute shape Correct answer.
  4. A model deployment with autoscaling

Explanation

Jobs are designed for non-interactive, repeatable execution of scripts on managed compute, and they can be scheduled or triggered without a persistent session. This avoids paying for idle interactive compute and gives run history.

Question 4Data Science Overview and OCI Data Science Concepts

What does the OCI Data Science model catalog store?

  1. Only the raw training dataset
  2. Only the IAM policies for the project
  3. Model artifacts together with metadata such as provenance and schema information Correct answer.
  4. Only the notebook source code

Explanation

The model catalog is a repository for versioned model artifacts plus associated metadata, including input/output schema and provenance, so models can be tracked and later deployed. It does not store datasets or policies.

Question 5Data Science Overview and OCI Data Science Concepts

Which statement best describes the relationship between a model deployment and a model artifact in OCI Data Science?

  1. A deployment stores datasets for batch scoring only
  2. A deployment automatically retrains the artifact on new data
  3. A deployment replaces the need for a model catalog
  4. A deployment serves a specific stored model artifact as an HTTP endpoint Correct answer.

Explanation

A model deployment takes a model artifact from the catalog and hosts it behind an HTTP endpoint for online inference. Retraining is a separate job or pipeline activity, not something a deployment performs automatically.

Question 6Data Science Overview and OCI Data Science Concepts

An ML engineer wants to package a scikit-learn model so it can be deployed with OCI Data Science. What is the recommended artifact format?

  1. An OCI model artifact created with the ADS or OCI CLI, containing the serialized model plus a score.py Correct answer.
  2. A raw CSV of predictions
  3. A Docker image pushed to a public registry
  4. A Jupyter notebook containing the training loop

Explanation

Deployable models are packaged as model artifacts that include the serialized model and a score.py defining the predict path, then registered in the catalog. This contract lets the deployment runtime load and invoke the model.

Question 7Data Science Overview and OCI Data Science Concepts

What is the purpose of the OCI Data Science conda environments?

  1. They define compute shapes for deployments
  2. They replace IAM authentication for notebooks
  3. They provide reproducible Python environments with curated or custom package sets Correct answer.
  4. They store model artifacts permanently

Explanation

Conda environments give reproducible dependency sets so notebooks and jobs run with consistent library versions. OCI publishes curated environments and supports custom ones published to Object Storage.

Question 8Data Science Overview and OCI Data Science Concepts

A notebook session needs to read data from an Object Storage bucket without embedding credentials in code. Which approach aligns with OCI Data Science best practice?

  1. Disable IAM policies for the compartment
  2. Use a resource principal or instance principal so the session obtains scoped, temporary credentials Correct answer.
  3. Hard-code a user's API key in the notebook
  4. Store the tenancy admin password in an environment variable

Explanation

Resource principals let OCI Data Science resources authenticate to other OCI services with short-lived credentials scoped by IAM policies, avoiding long-lived secrets in code. This is the recommended pattern for notebooks and jobs.

Question 9Data Science Overview and OCI Data Science Concepts

Which OCI service is commonly used as the durable storage layer for datasets and model artifacts consumed by OCI Data Science?

  1. Object Storage Correct answer.
  2. OCI Streaming
  3. OCI Queues
  4. OCI DNS

Explanation

Object Storage provides durable, scalable buckets that notebooks and jobs read from and write to, and it is the common backing store for artifacts and curated conda environments. Streaming, Queues, and DNS serve different purposes.

Question 10Data Science Overview and OCI Data Science Concepts

What does the OCI Data Science Accelerated Data Science (ADS) library primarily provide?

  1. A managed Spark cluster
  2. A relational database engine
  3. A Python SDK that simplifies accessing OCI resources and common ML tasks Correct answer.
  4. A replacement for the Jupyter kernel

Explanation

ADS is a Python library that wraps common workflows such as dataset loading, model artifact creation, and deployment, and it integrates with OCI services. It runs inside existing environments rather than replacing the kernel or providing infrastructure.

Question 11Data Science Overview and OCI Data Science Concepts

A team wants to automate a multi-step ML workflow with defined dependencies and repeatable runs inside OCI Data Science. Which capability should they use?

  1. Manual CLI commands run locally
  2. Individual notebook sessions
  3. Model deployments
  4. OCI Data Science Pipelines Correct answer.

Explanation

Pipelines let teams define multi-step workflows as directed acyclic graphs with dependencies, run them repeatedly, and track execution. This is the managed way to orchestrate ML steps such as data prep and training.

Question 12Data Science Overview and OCI Data Science Concepts

Which compute shape consideration is most relevant when creating a GPU-backed notebook session for deep learning?

  1. Disabling the conda environment
  2. Selecting a GPU shape and confirming the tenancy has GPU capacity in the chosen availability domain Correct answer.
  3. Choosing a smaller CPU shape to reduce cost
  4. Using a public internet gateway instead of a private subnet

Explanation

GPU workloads require a GPU shape, and capacity depends on availability-domain availability, so the shape and placement must be confirmed. CPU shapes cannot accelerate deep learning training in the same way.

Question 13Data Science Overview and OCI Data Science Concepts

What is the main benefit of using OCI Vault with OCI Data Science workloads?

  1. Automatic feature engineering
  2. Faster model inference latency
  3. Replacement for the model catalog
  4. Centralized storage and controlled access of secrets such as API keys Correct answer.

Explanation

Vault stores secrets and encryption keys centrally so notebooks and jobs can retrieve them through authorized access instead of hard-coding credentials. It addresses secret management, not performance or modeling features.

Question 14Data Science Overview and OCI Data Science Concepts

A data scientist wants to bring a custom Python package into a notebook session. Which approach is supported?

  1. Edit the managed base image directly on the host
  2. Install it in the session or publish a custom conda environment that includes it Correct answer.
  3. Request that OCI modify the shared service image
  4. Convert the package into a model artifact

Explanation

Custom packages can be installed within a session for ad hoc use, or baked into a custom conda environment published to Object Storage for reproducibility across sessions and jobs. Managed service images are not edited directly.

Question 15Data Science Overview and OCI Data Science Concepts

Which OCI Data Science feature helps track experiments and compare model runs?

  1. IAM dynamic groups
  2. VCN security lists
  3. Object Storage bucket versioning alone
  4. ML tracking and metadata logging through ADS and the model catalog Correct answer.

Explanation

ADS supports tracking of parameters, metrics, and artifacts, and the model catalog records provenance so runs can be compared. Bucket versioning, IAM, and network controls do not provide experiment tracking.

Question 16Data Science Overview and OCI Data Science Concepts

Which statement about OCI Data Science jobs is accurate?

  1. A job run executes a script or notebook on managed compute and records logs and status Correct answer.
  2. A job run requires an open notebook session to execute
  3. A job run cannot reference conda environments
  4. A job run can only use CPU shapes

Explanation

Jobs execute specified code on managed compute independent of interactive sessions, capturing logs and status for each run. They support chosen shapes and conda environments.

Question 17Data Science Overview and OCI Data Science Concepts

What is the role of a pipeline in the OCI Data Science ML lifecycle?

  1. It provides a GUI for labeling images
  2. It replaces Object Storage for datasets
  3. It hosts the model endpoint for inference
  4. It orchestrates and automates sequences of ML steps with defined dependencies and parameters Correct answer.

Explanation

Pipelines automate multi-step workflows such as data preparation, training, and evaluation with explicit dependencies and parameterization. Inference hosting is the job of deployments, and data storage remains in Object Storage.

Question 18Data Science Overview and OCI Data Science Concepts

A model deployment must handle variable traffic with occasional spikes. Which deployment configuration supports this?

  1. Set a fixed single instance with no scaling
  2. Configure autoscaling with a minimum and maximum instance count Correct answer.
  3. Store the model in a notebook session
  4. Disable the endpoint and call the model directly from a notebook

Explanation

Model deployments support autoscaling based on load, with bounds on instance count, which absorbs traffic spikes while controlling cost. Fixed single-instance setups cannot scale, and notebooks are not production endpoints.

Question 19Data Science Overview and OCI Data Science Concepts

Which IAM construct allows OCI Data Science resources such as notebook sessions to authenticate to other OCI services without user credentials?

  1. API key
  2. Customer secret key
  3. Auth token
  4. Resource principal Correct answer.

Explanation

Resource principals are the IAM mechanism that lets OCI resources obtain short-lived credentials for service-to-service calls. API keys, auth tokens, and customer secret keys are user-centric long-lived credentials.

Question 20Data Science Overview and OCI Data Science Concepts

A team needs to compare two model versions and roll back quickly if a new deployment underperforms. Which practice best supports this?

  1. Version artifacts in the model catalog and manage deployments that reference specific versions Correct answer.
  2. Delete the previous deployment before testing the new one
  3. Keep models only in notebook memory
  4. Overwrite the single artifact each time training finishes

Explanation

Versioning artifacts in the catalog preserves each model and its metadata, and deployments can reference a chosen version so teams can switch or roll back. Overwriting or deleting removes the ability to compare and revert.

Question 21Data Science Overview and OCI Data Science Concepts

Which statement correctly distinguishes OCI Data Science from OCI Data Flow?

  1. Data Science focuses on ML development and deployment, while Data Flow runs managed Apache Spark jobs Correct answer.
  2. Both are identical services with different names
  3. Data Science runs Spark clusters, while Data Flow hosts JupyterLab
  4. Data Flow stores model artifacts, while Data Science stores streaming data

Explanation

Data Science provides environments, jobs, pipelines, and deployments for ML, whereas Data Flow is a managed Apache Spark service for large-scale data processing. They are complementary rather than interchangeable.

Question 22Data Science Overview and OCI Data Science Concepts

A notebook session cannot reach an Object Storage bucket in a private subnet. Which configuration area should be investigated first?

  1. The model catalog versioning
  2. Network configuration such as the VCN, subnet, and service gateway or private endpoint Correct answer.
  3. The conda environment package list
  4. The number of deployment instances

Explanation

Connectivity for private sessions depends on VCN routing and a service gateway or private endpoint to reach Object Storage. Package lists, catalog versions, and deployment counts do not affect network reachability.

Question 23Data Science Overview and OCI Data Science Concepts

Which approach best supports reproducibility of a training job across environments?

  1. Rely on whatever packages are installed in the latest session
  2. Manually install packages each run without recording versions
  3. Pin dependencies in a conda environment and version the code and model artifacts Correct answer.
  4. Use a different random seed on every run

Explanation

Reproducibility requires pinned dependencies, versioned code, and versioned artifacts so the same inputs yield comparable results. Unrecorded installs and varying seeds undermine repeatability.

Question 24Data Science Overview and OCI Data Science Concepts

What does a model deployment's score.py file define?

  1. The IAM policy for the deployment
  2. The loading and prediction logic invoked when the endpoint receives a request Correct answer.
  3. The Object Storage bucket lifecycle rule
  4. The conda environment package list

Explanation

score.py is the entry point that loads the model and implements the predict function called by the deployment runtime for each request. It is not used for IAM, dependencies, or storage rules.

Question 25Data Science Overview and OCI Data Science Concepts

A data scientist wants to use a GPU for a training job but the job fails with a capacity error. What is the most likely cause?

  1. The model catalog is full
  2. The score.py file is missing
  3. The conda environment is missing NumPy
  4. The requested GPU shape is unavailable in the selected availability domain Correct answer.

Explanation

GPU capacity is availability-domain dependent, so a capacity error usually means the requested GPU shape is unavailable in the selected AD. Missing libraries, catalog limits, and score.py are unrelated to capacity.

Question 26Data Science Overview and OCI Data Science Concepts

In OCI Data Science, which resource is the interactive, browser-based coding environment where a data scientist runs Python code and connects to compute?

  1. Model catalog entry
  2. Model deployment
  3. Data Science job
  4. Notebook session Correct answer.

Explanation

A notebook session is the managed interactive environment that provides a JupyterLab interface backed by a chosen compute shape and block storage. Jobs and deployments are non-interactive execution resources, and the model catalog stores artifacts rather than running code.

Question 27Data Science Overview and OCI Data Science Concepts

Which OCI Data Science construct is used to run a Python script or notebook non-interactively to completion, optionally on a schedule, without keeping an interactive session open?

  1. Job Correct answer.
  2. Conda environment
  3. Model deployment
  4. Notebook session

Explanation

Jobs execute code to completion on managed compute and can be triggered manually or by a schedule, which suits batch training and scoring. A notebook session stays interactive, a deployment serves live inference, and a conda environment only defines installed packages.

Question 28Data Science Overview and OCI Data Science Concepts

What is the primary purpose of the OCI Data Science model catalog?

  1. To cache conda packages for faster environment creation
  2. To provide a hosted JupyterLab interface for exploratory analysis
  3. To schedule recurring training runs on a cron-like trigger
  4. To store versioned model artifacts and their metadata so they can be deployed or shared Correct answer.

Explanation

The model catalog stores trained model artifacts together with provenance metadata such as the training source, and those entries can be deployed directly. It is not an interactive environment, a scheduler, or a package cache.

Question 29Data Science Overview and OCI Data Science Concepts

Which OCI Data Science component provides the HTTP endpoint that applications call to obtain predictions from a trained model?

  1. Project
  2. Pipeline
  3. Notebook session
  4. Model deployment Correct answer.

Explanation

A model deployment provisions the infrastructure and endpoint that serves inference requests for a model artifact stored in the catalog. Projects group resources, pipelines orchestrate steps, and notebook sessions are for interactive work.

Question 30Data Science Overview and OCI Data Science Concepts

In OCI Data Science, what is a project used for?

  1. To define the conda package channels available to the tenancy
  2. To group related Data Science resources such as notebook sessions, jobs, and models for access control and organization Correct answer.
  3. To serve as the billing boundary for all OCI services
  4. To physically host the GPU hardware used by training jobs

Explanation

A project is a logical container that groups Data Science resources so teams can organize work and apply IAM policies at that scope. Hardware is chosen per resource via shapes, conda channels are configured separately, and billing is handled at the tenancy and compartment level.

Question 31Data Science Overview and OCI Data Science Concepts

What does the OCI Data Science Accelerated Data Science (ADS) SDK offer to a data scientist working in a notebook?

  1. A managed feature store with automatic online serving
  2. Helpers for loading data, exploring datasets, training and evaluating models, and saving or deploying models Correct answer.
  3. A replacement for the OCI CLI that manages all tenancy networking
  4. A compiler that converts Python scripts into Spark jobs

Explanation

ADS is a Python library that streamlines common data science tasks such as dataset inspection, model evaluation, and transferring models to the catalog or deployments. It does not manage networking, compile Python into Spark, or act as a feature store.

Question 32Data Science Overview and OCI Data Science Concepts

Which OCI service is the typical durable location for datasets and model artifacts that OCI Data Science reads and writes?

  1. OCI API Gateway
  2. OCI Functions
  3. OCI Vault
  4. Object Storage Correct answer.

Explanation

Object Storage provides durable, scalable buckets that notebooks, jobs, and deployments commonly use for datasets and artifacts. Vault stores secrets, Functions runs event-driven code, and API Gateway fronts APIs.

Question 33Data Science Overview and OCI Data Science Concepts

What does a conda environment published to OCI Data Science define for a notebook session or job?

  1. The IAM policy that grants access to Object Storage
  2. The set of installed packages and their versions used at runtime Correct answer.
  3. The network route table used by the session
  4. The compute shape and number of GPUs allocated

Explanation

A conda environment captures the Python and library versions installed for a session or job, which supports reproducible runtimes. Compute shapes, IAM policies, and networking are configured on the resource itself rather than in the environment.

Question 34Data Science Overview and OCI Data Science Concepts

A data scientist wants a notebook session to access an Object Storage bucket without storing credentials in notebook code. Which OCI mechanism should be used?

  1. A bucket pre-authenticated request embedded in the script
  2. An SSH key pair generated inside the notebook
  3. A resource principal or instance principal authorized through IAM policies Correct answer.
  4. A long-lived user API key pasted into a notebook cell

Explanation

Resource principals let OCI Data Science resources obtain short-lived credentials and act under IAM policies, so no static secrets appear in code. Pasting API keys or embedding pre-authenticated requests exposes credentials in the notebook.

Question 35Data Science Overview and OCI Data Science Concepts

A team runs a training job that must read a large dataset from Object Storage. Which configuration most directly allows the job to reach the bucket in a private subnet?

  1. Place the job in a subnet with a service gateway or appropriate route and allow the traffic in security rules Correct answer.
  2. Attach a public IP address to the job and open all egress ports
  3. Store the dataset in the notebook's block storage instead of Object Storage
  4. Disable the VCN so traffic bypasses network controls

Explanation

Private access to Object Storage is achieved through a service gateway or suitable routing plus security rules that permit the traffic. Public IPs and disabling network controls weaken security, and block storage is not a durable shared dataset store.

Question 36Data Science Overview and OCI Data Science Concepts

Which practice best supports reproducibility when a training job must run in multiple environments?

  1. Pin package versions in a published conda environment and version the code and model artifacts Correct answer.
  2. Use a different compute shape for each run without recording it
  3. Install packages interactively each run with the latest versions
  4. Rely on whatever packages happen to be installed in the base environment

Explanation

Reproducibility requires pinning dependency versions in an environment and tracking code and artifact versions so runs can be recreated. Unpinned installs and unrecorded shape changes introduce uncontrolled variation.

Question 37Data Science Overview and OCI Data Science Concepts

What is the role of the score.py file in an OCI Data Science model deployment?

  1. It configures the load balancer bandwidth for the endpoint
  2. It defines how the deployment loads the model and produces predictions from input data Correct answer.
  3. It lists the conda packages installed in the notebook session
  4. It stores the IAM policy that authorizes endpoint callers

Explanation

score.py implements the artifact loading and the prediction logic invoked when the deployment receives a request. IAM, bandwidth, and notebook packages are configured elsewhere and are not defined by score.py.

Question 38Data Science Overview and OCI Data Science Concepts

A data scientist needs a GPU for a training job but the job fails immediately with a capacity error. Which cause is most likely?

  1. The conda environment did not include the ADS library
  2. No GPU shape was available in the selected availability domain at request time Correct answer.
  3. The model catalog entry was created in a different compartment
  4. The score.py file was missing from the artifact

Explanation

Capacity errors indicate the requested compute shape, such as a GPU shape, could not be provisioned in the chosen availability domain at that moment. Missing libraries, compartment differences, and scoring scripts produce different failure modes.

Question 39Data Science Overview and OCI Data Science Concepts

A deployment must absorb sudden traffic spikes without manual intervention. Which configuration addresses this requirement?

  1. Store the model in multiple compartments
  2. Enable autoscaling on the model deployment with a suitable minimum and maximum instance count Correct answer.
  3. Increase the notebook session's block storage size
  4. Set the deployment to a single fixed instance with the largest available shape

Explanation

Autoscaling adjusts the number of deployment instances in response to load, which handles variable and spiky traffic. A fixed single instance cannot scale, and compartment or block storage changes do not affect serving capacity.

Question 40Data Science Overview and OCI Data Science Concepts

How does a pipeline in OCI Data Science help manage a multi-step machine learning workflow?

  1. It replaces the need for any storage of intermediate artifacts
  2. It converts the workflow into a Data Flow Spark application
  3. It defines ordered steps with dependencies so the workflow runs repeatably and each step's outputs feed later steps Correct answer.
  4. It automatically writes IAM policies for each step

Explanation

Pipelines express steps and their dependencies, enabling repeatable orchestration where outputs of one step become inputs to another. They still rely on storage for artifacts and do not author IAM policies or convert workflows into Spark.

Question 41Data Science Overview and OCI Data Science Concepts

A team wants to compare several training runs and identify which hyperparameters produced the best model. Which OCI Data Science capability supports this?

  1. Creating a service gateway in the VCN
  2. Tracking experiments and their runs with recorded metrics and parameters Correct answer.
  3. Enabling autoscaling on a model deployment
  4. Publishing a new conda environment for each run

Explanation

Experiment tracking records parameters, metrics, and artifacts per run so results can be compared and the best configuration selected. Autoscaling, networking, and environment publishing do not provide run comparison.

Question 42Data Science Overview and OCI Data Science Concepts

Which IAM construct lets a notebook session call other OCI services under a defined identity without embedding user credentials?

  1. An API signing key stored in the notebook
  2. A user account shared by the whole team
  3. A dynamic group matching the Data Science resource plus a policy granting the required access Correct answer.
  4. A bucket lifecycle policy

Explanation

Dynamic groups match resources such as notebook sessions, and policies grant those members access, enabling credential-free authorization. Shared users, stored keys, and lifecycle policies do not provide scoped resource identity.

Question 43Data Science Overview and OCI Data Science Concepts

A team must roll back quickly if a newly deployed model version underperforms. Which practice best supports this?

  1. Rely on notebook session history to restore the model
  2. Keep prior versions in the model catalog and deploy versioned artifacts so the previous version can be reactivated Correct answer.
  3. Delete the previous deployment before validating the new one
  4. Overwrite the existing artifact in place on every release

Explanation

Retaining versioned artifacts in the catalog allows a fast switch back to a known-good model. Overwriting artifacts, deleting the prior deployment, and depending on notebook history all remove or weaken the rollback path.

Question 44Data Science Overview and OCI Data Science Concepts

A data scientist needs to use a Python library that is not in the provided conda environments. Which approach is supported in OCI Data Science?

  1. Create or publish a custom conda environment that includes the library and use it for the session or job Correct answer.
  2. Modify the read-only base image on the host directly
  3. Convert the library into an Object Storage bucket
  4. Request that OCI support install the package into the shared image

Explanation

Custom conda environments can be built and published so sessions and jobs install the needed packages at startup. Host images are managed, support does not install arbitrary packages, and a bucket cannot serve as a library.

Question 45Data Science Overview and OCI Data Science Concepts

Which OCI service should be used to store API keys and other secrets that a Data Science job retrieves at runtime?

  1. OCI API Gateway
  2. OCI Vault Correct answer.
  3. OCI Data Flow
  4. OCI Object Storage

Explanation

Vault securely stores secrets and keys that workloads can retrieve at runtime under IAM control. Data Flow runs Spark, Object Storage holds data objects, and API Gateway fronts APIs, so none of them provides secret management.

Question 46Data Science Overview and OCI Data Science Concepts

An ML engineer wants to package a model so it can be deployed from the OCI Data Science model catalog. Which combination is required?

  1. A notebook file with no scoring script
  2. Only a requirements.txt with no model file
  3. A conda environment with no model artifact
  4. The serialized model artifact plus a score.py that defines loading and prediction logic Correct answer.

Explanation

Deployment needs the saved model artifact and a scoring script that loads it and implements prediction. Requirements or environments alone lack the model, and a notebook without a scoring script cannot serve requests.

Question 47Data Science Overview and OCI Data Science Concepts

A pipeline step must consume the output produced by an earlier step in the same run. How should the engineer design this?

  1. Hard-code a local path that exists only on one machine
  2. Rely on the notebook session staying open for the entire pipeline run
  3. Have the earlier step write its output to a durable location and pass that location as an input to the later step Correct answer.
  4. Assume the later step can read the earlier step's in-memory variables

Explanation

Pipeline steps run as separate executions, so outputs must be persisted to durable storage and referenced as inputs by later steps. In-memory state, an open session, and machine-local paths do not survive across steps.

Question 48Data Science Overview and OCI Data Science Concepts

A team wants to minimize inference latency for a model deployment that receives steady, predictable traffic. Which choice is most appropriate?

  1. A fixed instance count sized to the observed load with a shape matching the model's compute needs Correct answer.
  2. Storing the model in several compartments
  3. A single smallest shape with autoscaling disabled
  4. Autoscaling from zero instances with no minimum

Explanation

For steady, predictable load, a fixed count sized to demand avoids cold-start delays and keeps latency stable. Scaling from zero introduces startup latency, an undersized shape throttles inference, and compartment duplication adds no capacity.

Question 49Data Science Overview and OCI Data Science Concepts

A data scientist must choose between running preprocessing as a Data Science job or as a Data Flow application. The input is a very large dataset requiring distributed Spark processing. Which choice fits best?

  1. A model deployment, because it processes data in parallel at inference time
  2. A notebook session, because interactive execution distributes work across nodes
  3. A Data Science job on a single small shape, because jobs always parallelize automatically
  4. OCI Data Flow, because it provides managed Apache Spark for distributed processing at scale Correct answer.

Explanation

Data Flow is the managed Spark service intended for large-scale distributed processing. A single-shape job does not distribute automatically, notebook sessions are interactive, and deployments serve inference rather than batch preprocessing.

Question 50Data Science Overview and OCI Data Science Concepts

A model deployment intermittently returns errors because the scoring script cannot locate the model artifact at startup. Which area should be investigated first?

  1. The block storage size of the notebook session used for training
  2. How score.py resolves the artifact path relative to the deployed model artifact location Correct answer.
  3. The IAM policy that grants callers permission to invoke the endpoint
  4. The autoscaling maximum instance count

Explanation

Startup failures that reference a missing artifact usually stem from incorrect path handling in the scoring script relative to where the artifact is mounted. Caller permissions, autoscaling limits, and notebook storage do not affect artifact resolution at startup.