50 Oracle Cloud Infrastructure 2025 Data Science Professional Practice Questions: Question Bank 2025

A data scientist is onboarding to OCI Data Science and needs to train models using notebooks while ensuring network traffic never traverses the public internet. Which configuration best meets this requirement?

You trained a classification model and want a quick, reliable estimate of generalization performance during development. Which approach is generally recommended when you have enough data?

Your model is deployed as an OCI Data Science model deployment. You need to reduce client-side failures during brief periods when new model instances are starting. What should the client application do as a best practice?

A team wants to store training datasets, intermediate features, and model artifacts in a durable, cost-effective repository accessible from notebooks and jobs. Which OCI service is the most appropriate primary storage?

You are building a reproducible training workflow in OCI Data Science Jobs. The job runs fine interactively in a notebook but fails as a job because some Python packages are missing. What is the recommended way to ensure consistent dependencies across runs?

A regression model shows very low error on training data but significantly higher error on validation data. Which issue is most likely, and what is an appropriate first mitigation?

Your OCI Data Science model deployment intermittently returns 502 errors under increased load. Metrics show CPU saturation and elevated latency. What is the best next step to improve availability and performance?

A pipeline must retrain a model whenever new data lands in Object Storage, run feature engineering, train, evaluate, and register the model. The team wants automation with clear separation of steps and auditable runs. Which design best fits OCI-native MLOps?

A financial services company must ensure that model deployments can only be invoked by specific applications and that the model artifacts are protected with least-privilege access. Which approach best satisfies this requirement on OCI?

You trained a model using a time-ordered dataset (for example, demand forecasting). The model performs well in offline validation but performs poorly in production after deployment. Investigation shows validation used random train/test splits. What is the most likely root cause and best corrective action?

A data scientist wants to run exploratory analysis in an OCI Data Science Notebook Session and access Object Storage. Which approach follows OCI best practices for authentication from the notebook?

You have an imbalanced binary classification dataset where the positive class is ~1% of observations. Which evaluation metric is generally the most informative for model selection in this scenario?

A model is deployed as an OCI Data Science Model Deployment. Users report that some prediction requests intermittently return HTTP 5xx errors during traffic spikes. What is the most appropriate first action to troubleshoot?

Your team trains models in OCI Data Science and wants every run to be reproducible and auditable. Which combination best supports this goal?

A deployed model needs to call an OCI Vault secret (for example, an API key to a third-party enrichment service) at runtime. What is the recommended way to grant this access?

You are designing a feature pipeline where new data lands in Object Storage each hour. The training dataset must be built consistently from raw files and should be re-creatable for any past date. Which design is most appropriate?

A classification model shows high training accuracy but significantly lower validation accuracy. Which action most directly addresses the likely issue?

A model deployment works in a test compartment but fails in production with authorization errors when reading from an Object Storage bucket. Both deployments use the same code and model artifact. What is the most likely cause?

You need to deploy a model that requires a custom system library not present in the default Data Science serving environment. The deployment must be repeatable across compartments and regions. Which approach is most appropriate?

A regulated enterprise requires that only approved model versions can be promoted to production, and each promotion must be traceable to training data, code, and evaluation results. Which MLOps control best satisfies this requirement?

You need to ensure that only data scientists in a specific OCI group can create and run Data Science jobs, while other users can only view job runs and logs. Which approach best meets this requirement?

During model development, you discover that 2% of rows have missing values in a critical numeric feature, and the model will be deployed to production where missingness is expected to continue. Which approach is the most appropriate best practice?

A deployed model is returning valid predictions but the team cannot troubleshoot failures because they do not see detailed request/response logs. What is the most likely missing configuration for operational visibility?

Your training dataset is stored in Object Storage and contains sensitive PII. The security team requires that training jobs and notebook sessions must not traverse the public internet when accessing the data. Which design best satisfies this?

A model performs well offline but shows degraded accuracy in production. Investigation suggests that the live feature distribution differs from training. Which monitoring/validation approach most directly detects this type of issue?

You are building a reproducible ML workflow using OCI. You want training to be triggered automatically when new labeled data lands in Object Storage, with clear separation between data prep, training, and evaluation steps. Which approach is most appropriate?

Your team wants to standardize environments for training jobs so that the same dependencies are used across all projects, and security wants curated base images with vulnerability scanning. What is the best practice approach?

A real-time model deployment intermittently returns HTTP 504 timeouts during traffic spikes. Metrics show CPU is saturated and request latency increases sharply. Which remediation is most appropriate?

You need to train a model on data stored in Autonomous Data Warehouse (ADW). The security team mandates that database credentials must not be embedded in notebooks, job scripts, or environment variables. What is the best solution?

A regulated workload requires end-to-end lineage: the team must prove which exact data snapshot, code version, and environment were used to produce a deployed model. Which combination best satisfies this requirement in OCI?

You need to run an exploratory notebook in OCI Data Science and securely access an Object Storage bucket in the same tenancy without storing long-lived credentials in the notebook. What is the recommended approach?

A data scientist wants to run a training job that needs a specific Python library not included in the default OCI Data Science job runtime. What is the best way to ensure reproducibility and consistent dependencies across runs?

You deployed a model as an OCI Data Science Model Deployment. The endpoint returns HTTP 401 errors when called from an application running on a Compute instance. The model itself works in local tests. What is the most likely cause?

A team wants to track datasets, metrics, and artifacts for multiple experiments and compare model performance across runs. Which capability best supports this requirement in OCI Data Science workflows?

A production batch scoring pipeline must run nightly: extract data from Autonomous Database, transform it, score with a trained model, and write predictions back. The team wants orchestration with retries, scheduling, and dependency management. Which OCI approach best fits?

You need to allow a model deployment to call an external OCI service (for example, Object Storage to fetch reference data) without embedding credentials in the inference code. What is the best practice?

A training dataset is highly imbalanced (1% positive class). The model achieves 99% accuracy but performs poorly in identifying the positive class. Which evaluation metric is most appropriate to focus on?

A model deployment is experiencing intermittent timeouts during peak traffic. You suspect request bursts are exceeding current serving capacity. Which action is most appropriate to improve reliability while keeping the deployment managed?

Your organization requires that only approved containers can be used for OCI Data Science jobs and model deployments, and that images are scanned and controlled centrally. What architecture best meets this requirement?

A regulated workload requires end-to-end network isolation for training and inference: no public internet egress, private access to Object Storage and other OCI services, and restricted inbound access. Which design best satisfies this in OCI for Data Science notebooks/jobs and model deployments?

A data scientist is using an OCI Data Science notebook session and needs to pull training data from Object Storage in the same tenancy without storing long-lived credentials in code. What is the recommended approach?

You trained a model in an OCI Data Science notebook and want to package it for reproducible deployment so that the inference environment matches training dependencies. Which artifact best supports this requirement?

A team is building an ETL pipeline in OCI Data Flow that writes curated features to Object Storage. They want downstream model training jobs to trigger automatically when new feature data arrives. Which OCI service is most appropriate to react to Object Storage events and initiate the next step?

A model deployed as an OCI Data Science model deployment intermittently returns HTTP 500 errors under load. The container logs show timeouts while downloading the model artifact from Object Storage during startup. What is the best mitigation?

A regulated enterprise requires that only approved container images can be used for model deployments. Images must be scanned and pulled from a controlled registry. Which approach aligns best with OCI services and best practices?

A data scientist needs to run a training job that uses a private PyPI mirror and must also access a private endpoint in a VCN. Which configuration is required for the training job to reach private network resources?

Your classification model has a high AUC but performs poorly for the minority class, which is the business-critical outcome. The dataset is significantly imbalanced. What is the most appropriate improvement to prioritize?

A feature engineering pipeline accidentally includes a field derived from post-outcome information (a leakage feature). In offline evaluation the model looks excellent, but in production performance collapses. Which practice best prevents this issue?

A team must deploy a model to an OCI Data Science model deployment that can only egress through a corporate inspection layer. The model deployment is placed in a private subnet and must reach OCI services (for example, Object Storage) without using public IPs. Which network design best meets this requirement?

You need a repeatable MLOps workflow that: (1) runs feature generation, (2) trains a model, (3) evaluates it against a baseline, and (4) conditionally promotes it for deployment if it meets thresholds. The workflow must be auditable and parameterized per environment (dev/test/prod). Which approach best fits these needs on OCI?