PMLE Automating and Orchestrating ML Pipelines Practice Test 4 — 15 Questions

Question 1

An ML engineer is building a Vertex AI Pipeline that includes a custom training component. The component must run on a single n1-standard-8 machine with one NVIDIA Tesla T4 GPU. The engineer wants to specify these hardware requirements directly in the component definition so that Vertex AI Pipelines provisions the correct machine type and accelerator. Which approach should the engineer use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use the `machine_type` and `accelerator_type` fields within the component's `@dsl.component` decorator or the component's YAML specification.. Vertex AI Pipelines lets you specify per-component resource requirements directly in the component definition using `machine_type` and `accelerator_type`. This ensures the training component gets the required GPU while other components can use smaller machines, optimizing cost and performance. Pipeline-level or CustomJob approaches either over-provision or complicate orchestration.

Answer

Create a separate Vertex AI CustomJob for training and invoke it from the pipeline using a custom Python function.

Answer

Define the hardware requirements in a `WorkerPoolSpec` within the component's container specification and pass it as a pipeline parameter.

Answer

Set the machine type and accelerator in the pipeline's RuntimeConfig when submitting the pipeline job.

Question 2

A team is building a Vertex AI Pipeline that includes a data preprocessing step and a training step. The preprocessing step outputs a processed dataset that the training step consumes. The team wants to ensure that the training step only runs if the preprocessing step completes successfully. How should they define the pipeline?

Accepted Answer

Add a dependency from the training step to the preprocessing step by using the output of the preprocessing step as an input to the training step.. In Vertex AI Pipelines, task dependencies are automatically established when one component consumes an artifact produced by another. By using the preprocessing step's output as an input to the training step, the pipeline ensures that training runs only after preprocessing succeeds. This is the idiomatic way to define sequential dependencies and guarantees that failures in preprocessing prevent training from starting.

Answer

Set the training step's 'after' parameter to the preprocessing step's name.

Answer

Create a separate pipeline that runs the training step and trigger it from the preprocessing step using a Cloud Function.

Answer

Use a conditional expression in the pipeline to check the status of the preprocessing step before running the training step.

Question 3

A data scientist is creating a Vertex AI pipeline that includes a component to preprocess data. The component takes a Cloud Storage URI as input and outputs a preprocessed dataset. The scientist wants to test the component in isolation before integrating it into the full pipeline. Which approach should the scientist take?

Accepted Answer

Run the component as a standalone Vertex AI CustomJob by specifying the container image and arguments.. Running the component's container as a Vertex AI CustomJob allows isolated testing with the same runtime environment as in a pipeline. This approach is direct, efficient, and validates the component's behavior with cloud resources. It is a recommended practice for developing and debugging pipeline components.

Answer

Create a new pipeline that contains only this component and run it.

Answer

Use the Vertex AI Pipelines local execution mode to run the component on the local machine.

Answer

Deploy the component as a Vertex AI endpoint and send test requests to it.

Question 4

An ML engineer is building a Vertex AI pipeline that includes a custom training component. The component must read training data from a BigQuery table and write the trained model to a Cloud Storage bucket. The engineer wants to define the component using the Vertex AI SDK for Python. Which approach should the engineer take to ensure the component can access the required resources?

Accepted Answer

Define the component as a Python function and specify the BigQuery table and Cloud Storage bucket as input parameters, using the google-cloud-bigquery and google-cloud-storage libraries within the function.. The Vertex AI SDK for Python allows defining lightweight Python function components. The engineer can specify inputs such as the BigQuery table and Cloud Storage bucket as string parameters, and the function can use the respective client libraries to access them. This approach is idiomatic, flexible, and leverages the SDK's support for parameterized components, ensuring the component can be reused with different resources.

Answer

Create a custom container that includes the training code and dependencies, and hardcode the BigQuery table and Cloud Storage paths in the container.

Answer

Use a pre-built Google Cloud Pipeline Component for BigQuery and Cloud Storage, and chain them together with the training component.

Answer

Define the component as a Python function and pass the BigQuery table and Cloud Storage bucket as artifacts, using the artifact types for BigQuery and Cloud Storage.

Question 5

A company is using Vertex AI Pipelines to automate their ML workflow. They want to schedule the pipeline to run every day at 2:00 AM. They also want to be able to manually trigger the same pipeline on demand. Which Vertex AI feature should they use to schedule the pipeline?

Accepted Answer

Vertex AI Pipelines' built-in cron scheduler, configured with a cron expression.. Vertex AI Pipelines includes a native scheduling capability that uses cron expressions to run pipelines on a recurring basis. You can create a schedule that runs the pipeline daily at 2:00 AM, and you can still manually submit the pipeline for on-demand runs. The other options require external services or confuse Vertex AI Training with Pipelines. The built-in scheduler is the simplest and most direct way to meet the requirement.

Answer

Vertex AI Training's built-in scheduling feature, configured with a recurring training job.

Answer

Cloud Scheduler job that triggers a Cloud Function which submits the pipeline.

Answer

Cloud Composer (Airflow) DAG with a schedule interval that runs the pipeline via the Vertex AI operator.

Question 6

An ML engineer is creating a Vertex AI Pipeline that includes a component to preprocess data. The component needs to read a CSV file from Cloud Storage and output a transformed CSV file to Cloud Storage. The engineer wants to define the component using the KFP SDK v2. Which parameter types should the engineer use for the input and output file paths?

Accepted Answer

Use InputPath for the input file path and OutputPath for the output file path.. In KFP v2, InputPath and OutputPath are the correct parameter types for file inputs and outputs. They handle the translation between pipeline parameters and local file paths, and they ensure that output directories are created. Using plain strings or other types does not provide these benefits and can lead to runtime errors.

Answer

Use Dataset artifact for both input and output.

Answer

Use google.cloud.storage.Blob for both input and output.

Answer

Use str for both the input and output file paths.

Question 7

An ML engineer is using Vertex AI Pipelines to orchestrate a workflow that includes a custom training component. The component must access a secret stored in Secret Manager to authenticate to an external data source. The engineer wants to avoid hardcoding the secret in the component code or pipeline definition. Which approach should the engineer use?

Accepted Answer

Use the `Vertex AI` secret mounting feature to mount the secret as an environment variable in the component's container.. Vertex AI Pipelines can mount secrets from Secret Manager directly into the component's container as environment variables or files. This avoids hardcoding the secret in code or pipeline parameters and keeps it out of metadata and logs. The component's service account must have the necessary permissions to access the secret. This is the secure and recommended way to provide secrets to pipeline components.

Answer

Mount the secret as a file using the `gcloud secrets versions access` command in the component's container at runtime.

Answer

Store the secret in a Cloud Storage bucket and have the component download it at runtime.

Answer

Pass the secret as a pipeline parameter and set it as an input to the component.

Question 8

A company runs a Vertex AI Pipeline that includes a data preprocessing component and a training component. The preprocessing component outputs a dataset artifact that the training component consumes. The team wants to ensure that if the preprocessing component fails and is retried, the training component does not start until the preprocessing component successfully completes. Which Vertex AI Pipelines feature should they use to enforce this dependency?

Accepted Answer

Define an output artifact in the preprocessing component and use it as an input parameter in the training component.. Connecting an output artifact from the preprocessing component to an input of the training component creates an implicit dependency. Vertex AI Pipelines ensures the training component only runs after the preprocessing component successfully completes and produces the artifact. This is the standard way to enforce data-dependent execution order.

Answer

Configure the training component to poll a Cloud Storage bucket for the preprocessed data before starting.

Answer

Use the `after` keyword in the pipeline definition to explicitly set the execution order of components.

Answer

Set the `retry` policy on the preprocessing component to a high number and rely on pipeline-level retries.

Question 9

An ML engineer is designing a Vertex AI pipeline that must run on a schedule and also be triggered by events. The engineer wants to minimize operational overhead while ensuring the pipeline can be parameterized per run. Which two approaches should the engineer use? (Choose two.)

Accepted Answer

Use Vertex AI Pipelines' built-in scheduling to run the pipeline on a cron schedule with runtime parameters.. Vertex AI Pipelines' native recurring runs provide managed cron scheduling with runtime parameters, and Cloud Scheduler can call the Pipelines REST API with a parameterized payload. Both minimize operational overhead by using managed services and support per-run parameters. Cloud Composer adds an Airflow environment, Cloud Run lacks a built-in scheduler, and a Cloud Function on a storage event is event-driven rather than scheduled.

Answer

Write a Cloud Function that triggers the pipeline and invoke it from a Cloud Storage object change notification.

Answer

Deploy the pipeline as a Cloud Run service and let Cloud Run's built-in scheduler invoke it.

Answer

Use a Cloud Composer environment to run the pipeline as an Airflow task on a schedule.

Question 10

An ML engineer is designing a Vertex AI Pipeline that must run a custom training component and then, based on the evaluation metrics, conditionally deploy the model to an endpoint. The engineer wants to minimize pipeline complexity and leverage built-in Vertex AI features. Which two approaches should the engineer use to implement the conditional deployment? (Choose two.)

Accepted Answer

Use the `dsl.If` expression to compare the evaluation metric against a threshold and conditionally execute the deployment component.. Using `dsl.Condition` or `dsl.If` within the Vertex AI Pipeline allows conditional deployment based on evaluation metrics without external orchestration. Both are built-in constructs that keep the logic inside the pipeline, simplifying maintenance and leveraging native features. External Cloud Functions or monolithic components add unnecessary complexity and security concerns.

Answer

Implement the conditional logic inside a single custom component that performs evaluation and, if the metric passes, calls the Vertex AI endpoint deployment API.

Answer

Configure the pipeline to always deploy the model, but add a post-deployment validation step that rolls back if the metric is poor.

Answer

Create a separate Cloud Function that listens for the pipeline's completion event and triggers deployment if the metric is satisfactory.

Question 11

An ML engineer is designing a Vertex AI pipeline that must orchestrate a complex workflow with parallel training of multiple models, followed by a selection step that picks the best model based on evaluation metrics. The pipeline should also support retries for transient failures. Which two features of Vertex AI Pipelines should the engineer use to implement this? (Choose two.)

Accepted Answer

Use the ParallelFor construct to run multiple training components concurrently.. To run multiple training components in parallel, the ParallelFor construct is used. It allows iterating over a list and executing a component for each item concurrently. For retries, each component can have a retry policy that specifies how many times to retry and the backoff strategy. These two features together enable a robust, parallel training workflow with automatic recovery from transient failures.

Answer

Use the Condition construct to select the best model based on evaluation metrics.

Answer

Use the pipeline's caching mechanism to avoid re-running successful components.

Answer

Use the ExitHandler construct to define a cleanup step that runs after the pipeline completes.

Question 12

A team is building a Vertex AI Pipeline that includes a component for hyperparameter tuning. The tuning component uses Vertex AI Hyperparameter Tuning service and outputs the best hyperparameter set. The team wants to ensure that the tuning component can be retried without duplicating trials if it fails due to a transient error. Which configuration should they apply to the tuning component?

Accepted Answer

Implement custom logic in the tuning component to check for existing trials in a Cloud Storage bucket and skip already completed trials before starting new ones.. Since Vertex AI Hyperparameter Tuning does not automatically resume or deduplicate trials, the team must implement custom tracking. By storing completed trials in a persistent location like Cloud Storage and checking that store on retry, the component can avoid re-running trials, ensuring idempotency and cost efficiency.

Answer

Use a persistent trial store, such as a Vertex AI TensorBoard instance, to record trials and configure the tuning component to resume from the last trial.

Answer

Set the `retry` policy on the component to a fixed number of retries and rely on Vertex AI Hyperparameter Tuning's built-in idempotency.

Answer

Pass a `study_id` to the tuning component and configure it to use the same study across retries, ensuring trials are not duplicated.

Question 13

A machine learning engineer is building a Vertex AI pipeline that uses a pre-built Google Cloud Pipeline Components (GCPC) to train a custom model. Which component should the engineer use to submit a custom training job to Vertex AI?

Accepted Answer

CustomJob. The CustomJob component is the correct choice because it is the pre-built GCPC component specifically designed to submit a custom training job to Vertex AI. It allows the engineer to specify a custom container image or a Python training script, along with machine configuration and hyperparameters, directly within a Vertex AI pipeline. Other components serve different purposes, such as hyperparameter tuning, batch predictions, or model deployment.

Answer

HyperparameterTuningJob

Answer

BatchPredictionJob

Answer

ModelDeploy

Question 14

A team has a Vertex AI pipeline that includes a container component for data preprocessing. The team notices that the component is re-executed every time the pipeline runs, even when the inputs and code haven't changed. They want to leverage pipeline caching to avoid redundant executions. What should they do to enable caching for this component?

Accepted Answer

Ensure that the component does not have 'dsl.cache_options(enable_cache=False)' set.. Vertex AI pipeline caching is enabled by default for all components unless explicitly disabled using `dsl.cache_options(enable_cache=False)`. The component re-executing every time indicates that caching was likely disabled on that specific component. Removing or ensuring this setting is not present will allow the pipeline to reuse cached outputs when inputs and code have not changed.

Answer

Set the 'caching' flag to 'True' in the pipeline definition using 'pipeline.caching = True'.

Answer

Set the environment variable 'ENABLE_CACHE' to 'true' on the pipeline run request.

Answer

Re-compile the pipeline with the '--enable-cache' flag.

Question 15

A data engineer wants to orchestrate a complex workflow that includes running a Vertex AI pipeline, then a BigQuery job, and finally a Dataflow pipeline. The workflow must handle dependencies, retries, and monitoring. Which Google Cloud service is most suitable for this orchestration?

Accepted Answer

Cloud Composer. Cloud Composer (based on Apache Airflow) is the most suitable service for orchestrating a complex workflow with dependencies, retries, and monitoring across Vertex AI, BigQuery, and Dataflow. It provides a managed Airflow environment that natively supports DAG-based orchestration, built-in retry logic, and integration with Google Cloud services via operators like VertexAIPipelineOperator, BigQueryOperator, and DataflowTemplatedJobStartOperator.

Answer

Cloud Tasks

Answer

Cloud Scheduler

Answer

Workflows