Oracle Cloud Infrastructure 2025 Data Science Professional Practice Exam 2025: Latest Questions

You need to give a data scientist the ability to create and manage Data Science projects and notebook sessions in a compartment, but you must prevent them from creating or managing any network resources (VCNs, subnets, security lists). What is the best approach?

A data scientist is developing a model in an OCI Data Science notebook session and wants to persist datasets and feature files so they remain available after the notebook session is stopped or replaced. Which storage option is the best fit?

You are training a model and want to reduce overfitting. Which approach is most appropriate?

You deployed a model as an OCI Data Science model deployment and need to secure the endpoint so only authorized applications can invoke it. What is the recommended method?

A team wants to ensure their model training is reproducible across notebook sessions and between team members. Which practice best supports reproducibility in OCI Data Science?

You are building a binary classifier with 1% positive class and 99% negative class. Accuracy is high but the model misses most positives. Which metric is the best primary choice to assess performance for the minority class?

After deploying a model, you notice prediction quality degrading over time due to changing input data patterns. You want an operational approach to detect this issue early. What should you implement?

Your data engineering pipeline needs to load raw files from Object Storage, transform them, and write curated outputs back to Object Storage on a schedule. You also want built-in orchestration with retries and dependencies. Which OCI service is the best fit?

You deployed a model that requires GPU acceleration for low-latency inference. Requests intermittently fail with out-of-memory errors during peak traffic. Which design change best addresses both reliability and performance?

Your organization requires that model artifacts, training code, and key parameters are traceable for audit. You want to be able to reproduce exactly which artifacts were deployed, including the environment dependencies. Which approach best meets this requirement in OCI Data Science MLOps practices?

A data scientist wants a repeatable way to run notebooks with the same Python dependencies across teammates in OCI Data Science. Which approach is recommended?

You deployed a model as an OCI Data Science Model Deployment and must call it from a private subnet application without exposing a public endpoint. What is the best solution?

A team wants to store training datasets, feature snapshots, and model artifacts durably and make them accessible to OCI Data Science jobs. Which OCI service is the most appropriate primary storage location?

A data science pipeline trains a model daily. The business wants to detect data drift and model performance degradation over time. Which approach best fits OCI MLOps practices?

Your organization requires that Data Science notebook sessions have no direct internet access, but still need to install approved Python packages. What is the best architecture pattern?

A model deployment intermittently returns HTTP 5xx during peak traffic. Logs show long inference times. Which change most directly improves reliability while keeping latency low?

You need to ensure that only a CI/CD pipeline (not individual users) can register models and create deployments in a compartment. What is the best practice in OCI IAM?

A team is building a churn model where positive churn is rare (highly imbalanced). They want a metric that better reflects performance than accuracy and should guide threshold selection. Which is the best choice?

You must design an MLOps workflow where each model version is reproducible: same code, same dependencies, and traceability to training data. Which combination best supports this requirement?

A regulated workload requires that inference requests and responses are logged for audit, but PII must not be stored in logs. The current model deployment logs full payloads, causing compliance issues. What is the best remediation?

A data science team must run a notebook session that can read training data from Object Storage in a different compartment. Security requires least privilege and no user API keys on the instance. Which approach is recommended?

You are building a time-series demand forecasting model using a 3-year dataset. The model shows excellent offline performance but performs poorly in production. Investigation suggests leakage from future information during feature engineering. Which validation approach best reduces this risk?

A model deployed as an OCI Data Science Model Deployment intermittently returns 5xx errors under bursty traffic. Logs show increased latency when multiple requests arrive at once, and each request triggers model loading from disk. What change is most likely to resolve the issue?

An enterprise requires that all model training runs are reproducible and auditable. The team uses Data Science Jobs. Which combination of actions best supports reproducibility?

A model deployment needs to call an external OCI service (for example, to read features from Object Storage) using instance principal-style credentials. Requests fail with authorization errors even though an IAM policy exists for the dynamic group. Which is the most likely cause?