⚡ Free Classes and Scholarships Available for Underprivileged Students -

Model prototyping using Jupyter notebooks

Collaborating within and across teams to manage data and models

10 Questions
No time limit
Practice Mode
0%
Score
0
Correct
0
Incorrect
10
Total Questions
Back to Topics
Question 1 of 10
Your team needs to prototype a machine learning model using large-scale data processing on a 10TB dataset stored in Cloud Storage. The data scientists are familiar with PySpark and need interactive notebook capabilities. Which Jupyter backend option is most appropriate?
Explanation
Colab Enterprise connected to a Dataproc cluster is the best choice because: (1) Dataproc provides native Apache Spark support for processing large-scale data efficiently, (2) The Jupyter component on Dataproc enables interactive notebook development with PySpark kernels, (3) It's designed for distributed processing of multi-terabyte datasets. Option A (standard Workbench) lacks distributed processing capabilities for 10TB. Option B would struggle with performance despite increased disk space as it's not designed for distributed computing. Option C (Cloud Shell) has resource limitations unsuitable for large-scale ML prototyping.
Question 2 of 10
You are setting up Vertex AI Workbench for your ML team and need to implement security best practices. Which combination of security measures should you prioritize?
Explanation
Option B represents security best practices: (1) VPC Service Controls (VPC-SC) create security perimeters preventing data exfiltration, (2) Disabling public IP access reduces attack surface, (3) Identity-Aware Proxy (IAP) provides secure, context-aware access without VPN. Option D exposes instances to the internet unnecessarily. Option A uses weak password authentication vulnerable to attacks. Option C involves storing service account keys which violates the principle of using attached service accounts and Workload Identity, creating security risks if keys are compromised.
Question 3 of 10
Your data science team is prototyping in Jupyter notebooks and needs to collaborate effectively using version control. What is the recommended approach for integrating notebooks with Git repositories in Vertex AI Workbench?
Explanation
Option A is correct because Vertex AI Workbench includes JupyterLab with native Git integration that allows: (1) Direct repository cloning within the notebook interface, (2) Committing and pushing changes without leaving the development environment, (3) Viewing diffs and managing branches efficiently, (4) Maintaining notebook format (.ipynb) with full metadata. Option C is inefficient and error-prone. Option B loses the interactive notebook format and metadata. Option D doesn't provide true version control capabilities like branching, merging, or collaboration features that Git provides.
Question 4 of 10
You need to preprocess a 5TB dataset using Apache Spark before training a model. The preprocessing job will run once daily. Which approach provides the best balance of cost-efficiency and performance?
Explanation
Option C is optimal because: (1) Dataproc is designed for large-scale Spark workloads with distributed processing, (2) Autoscaling adjusts resources based on workload demands, reducing costs, (3) Ephemeral clusters (create-use-delete) minimize costs by only charging for actual usage time, (4) Dataproc provides native Spark optimization. Option A wastes resources running 24/7 when only needed once daily. Option D (Colab Enterprise) isn't designed for 5TB production preprocessing jobs. Option B would struggle with performance as Workbench isn't optimized for distributed Spark processing at this scale.
Question 5 of 10
Your team is using Colab Enterprise for ML prototyping and needs to access sensitive data in BigQuery. What is the recommended authentication approach?
Explanation
Option B follows security best practices: (1) Runtime-attached service accounts use Workload Identity for authentication without credential management, (2) Least-privilege principle (BigQuery Data Viewer/Job User) grants only necessary permissions, (3) No credential storage in code prevents exposure, (4) Auditable through IAM policies. Option C risks credential exposure and violates security policies. Option D grants excessive permissions (Owner role) violating least privilege. Option A creates security vulnerabilities through credential sharing and makes auditing difficult. Service accounts with scoped permissions are the GCP-recommended approach.
Question 6 of 10
You're choosing between Vertex AI Workbench managed notebooks and user-managed notebooks for your ML team. Which scenario most strongly favors managed notebooks?
Explanation
Option A highlights key advantages of managed notebooks: (1) Automatic health monitoring reduces operational burden, (2) Auto-shutdown for idle instances optimizes costs, (3) Google handles infrastructure management, patching, and updates, (4) Faster setup with pre-configured environments. Options C, D, and B all describe scenarios requiring user-managed notebooks, which provide: root access for custom packages (C), full VM control (D), and ability to install specific legacy software (B). Managed notebooks sacrifice some customization for reduced operational complexity, making them ideal for teams wanting to focus on ML rather than infrastructure.
Question 7 of 10
Your ML pipeline needs to process streaming data for real-time feature engineering in a Jupyter notebook environment. Which architecture would you recommend?
Explanation
Option D is correct because: (1) Spark Structured Streaming is designed for scalable real-time stream processing, (2) Dataproc provides the distributed computing power needed for streaming workloads, (3) Jupyter integration allows interactive development and testing of streaming pipelines, (4) Can handle backpressure and fault tolerance automatically. Option A doesn't scale well and lacks fault tolerance for production streaming. Option B creates significant latency (minute-level polling) and isn't true streaming. Option C is not scalable or reliable as memory is limited and data would be lost on instance restart.
Question 8 of 10
Your team wants to track ML experiments including hyperparameters, metrics, and model artifacts from Jupyter notebooks. Which approach best integrates with Google Cloud services?
Explanation
Option C is the recommended approach because: (1) Vertex AI Experiments SDK provides native integration with Google Cloud, (2) Automatically tracks parameters, metrics, artifacts, and metadata in a structured way, (3) Enables comparison of experiments through the Vertex AI UI, (4) Integrates with Vertex AI Model Registry for model lineage, (5) Provides APIs for programmatic querying and analysis. Option B requires manual effort and lacks structured querying. Option A doesn't scale and makes experiment comparison difficult. Option D is completely manual, error-prone, and not suitable for production ML workflows.
Question 9 of 10
You need to preprocess data using Spark in a Jupyter notebook and want to optimize performance. The dataset is 2TB stored in Cloud Storage. Which Spark kernel configuration is most appropriate?
Explanation
Option A is optimal because: (1) Dataproc provides distributed Spark processing across multiple workers for 2TB datasets, (2) PySpark kernel connects to the Dataproc cluster's Spark master for distributed execution, (3) Workers can be configured with appropriate CPU/memory for the workload, (4) Data parallelism significantly improves performance over single-machine processing. Option B (local mode) processes data on a single machine, limiting parallelism and performance. Option C (pandas) isn't designed for 2TB datasets and would be extremely slow. Option D (TPU) is for model training acceleration, not data preprocessing with Spark.
Question 10 of 10
Your organization requires that all notebook instances comply with security policies including no internet access and integration with corporate identity. Which Vertex AI Workbench configuration meets these requirements?
Explanation
Option D meets all security requirements: (1) VPC deployment with Private Google Access allows GCP service communication without internet exposure, (2) No external IP prevents internet access while allowing internal connectivity, (3) Identity-Aware Proxy (IAP) can integrate with corporate identity providers (like Cloud Identity or external IdPs via GCIP), (4) User-managed notebooks provide the flexibility needed for this security configuration. Option C violates no-internet requirement with public IP. Option B doesn't meet the no-internet requirement and email sharing isn't enterprise-grade security. Option A uses service account keys (not recommended) and SSH tunneling doesn't integrate with corporate identity properly.