support@Free-Braindumps.com
Register Login
Cisco
CompTIA
ISC
EC-Council
EXIN
Fortinet
ITIL
Juniper
Microsoft
VMware
Avaya
SAP
Google
NVIDIA
Salesforce
Amazon
Palo Alto Networks
ISC2
Splunk
ServiceNow
All Vendors
  1. Home
  2. Exams
  3. Google
  4. PROFESSIONAL MACHINE LEARNING ENGINEER

Google PROFESSIONAL MACHINE LEARNING ENGINEER Exam Actual Questions
Professional Machine Learning Engineer (Page 6 )

Updated On: 27-Jul-2026
Page 6 of 44
PDF with 375 Questions

You have a demand forecasting pipeline in production that uses Dataflow to preprocess raw data prior to model training and prediction. The only preprocessing step applies Z-score normalization on data stored in BigQuery and writes it back to BigQuery. New training data is added every week. You want to make the process more efficient by minimizing computation time and manual intervention. What should you do?

  1. Normalize the data using Google Kubernetes Engine.
  2. Translate the normalization algorithm into SQL for use with BigQuery.
  3. Use the normalizer_fn argument in TensorFlow's Feature Column API.
  4. Normalize the data with Apache Spark using the Dataproc connector for BigQuery.

Answer(s): B

Explanation:

Option B is correct because translating the normalization into SQL allows BigQuery to perform on-dataset computation directly where data resides, eliminating extra ETL steps and enabling incremental updates as new data arrives. This leverages BigQuery’s SQL engine for scalable, serverless preprocessing integrated with the dataflow pipeline workflow.
A) Incorrect — using GKE adds operational overhead and does not leverage warehouse-optimized SQL processing, increasing maintenance.
C) Incorrect — normalizer_fn is for feature engineering at model input time in TensorFlow, not for preprocessing and storing normalized data back to BigQuery.
D) Incorrect — Dataproc with Spark would reintroduce external processing and data movement, adding latency and cost.



You need to design a customized deep neural network in Keras that will predict customer purchases based on their purchase history. You want to explore model performance using multiple model architectures, store training data, and be able to compare the evaluation metrics in the same dashboard. What should you do?

  1. Create multiple models using AutoML Tables.
  2. Automate multiple training runs using Cloud Composer.
  3. Run multiple training jobs on AI Platform with similar job names.
  4. Create an experiment in Kubeflow Pipelines to organize multiple runs.

Answer(s): D

Explanation:

Option D is correct because Kubeflow Pipelines supports organizing and tracking multiple pipeline runs (experiments) to compare architectures and metrics in a centralized dashboard, aligning with ML lifecycle tooling for experimentation and reproducibility.
A) Incorrect — AutoML Tables is a high-level automated tool not suited for customized neural networks or detailed cross-architecture experimentation dashboards.
B) Incorrect — Cloud Composer is an orchestration service (Airflow) not specialized for organizing and visualizing ML experiments or comparing evaluation metrics within a dedicated dashboard.
C) Incorrect — Running multiple training jobs with similar names in AI Platform lacks structured experiment tracking and integrated comparison dashboards.



You are developing a Kubeflow pipeline on Google Kubernetes Engine. The first step in the pipeline is to issue a query against BigQuery. You plan to use the results of that query as the input to the next step in your pipeline. You want to achieve this in the easiest way possible. What should you do?

  1. Use the BigQuery console to execute your query, and then save the query results into a new BigQuery table.
  2. Write a Python script that uses the BigQuery API to execute queries against BigQuery. Execute this script as the first step in your Kubeflow pipeline.
  3. Use the Kubeflow Pipelines domain-specific language to create a custom component that uses the Python BigQuery client library to execute queries.
  4. Locate the Kubeflow Pipelines repository on GitHub. Find the BigQuery Query Component, copy that component's URL, and use it to load the component into your pipeline. Use the component to execute queries against BigQuery.

Answer(s): D

Explanation:

Option D is correct because Kubeflow Pipelines provides reusable components in its repository, such as a BigQuery Query Component, which can be loaded directly into a pipeline to run a BigQuery query and pass results to downstream steps, aligning with MLOps best practices for modular, reproducible pipelines. A is incorrect because manually exporting results to a new table is not automated within the pipeline workflow and adds manual steps. B is incorrect since building a custom Python script as a first step introduces unnecessary boilerplate when a ready-made component exists. C is incorrect because although valid, it reimplements functionality rather than leveraging an existing, shareable component from the Kubeflow ecosystem.



You are building a model to predict daily temperatures. You split the data randomly and then transformed the training and test datasets. Temperature data for model training is uploaded hourly. During testing, your model performed with 97% accuracy; however, after deploying to production, the model's accuracy dropped to 66%.
How can you make your production model more accurate?

  1. Normalize the data for the training, and test datasets as two separate steps.
  2. Split the training and test data based on time rather than a random split to avoid leakage.
  3. Add more data to your test set to ensure that you have a fair distribution and sample for testing.
  4. Apply data transformations before splitting, and cross-validate to make sure that the transformations are applied to both the training and test sets.

Answer(s): B

Explanation:

Option B is correct because: time-based splitting prevents leakage and data drift between training and production, ensuring evaluation reflects temporal distribution and seasonality important for weather data.
A) Incorrect — Normalizing separately can cause data leakage if computed exclusively on train/test; consistent preprocessing should be applied using a single fit on training data and applied to test data, not two separate steps.
C) Incorrect — Increasing test set size alone doesn’t address temporal leakage or distribution shift; it may still misrepresent production conditions.
D) Incorrect — Applying transformations before splitting can leak information if fit on full data; cross-validation must ensure consistent transformation fitting only on training folds.



You are developing models to classify customer support emails. You created models with TensorFlow using small datasets on your on-premises system, but you now need to train the models using large datasets to ensure high performance. You will port your models to Google Cloud and want to minimize code refactoring and infrastructure overhead for easier migration from on-prem to cloud. What should you do?

  1. Use Vertex AI for distributed training.
  2. Create a cluster on Dataproc for training.
  3. Create a Managed Instance Group with autoscaling.
  4. Use Kubeflow Pipelines to train on a Google Kubernetes Engine cluster.

Answer(s): A

Explanation:

Option A is correct because Vertex AI provides scalable, managed training with minimal code changes, facilitating migration from on-prem to Google Cloud for large datasets and distributed training. It abstracts infrastructure management and integrates with TensorFlow, enabling seamless porting of existing workflows.
B is incorrect because Dataproc focuses on Hadoop/Spark clusters and batch processing, introducing more refactoring and operational overhead for TensorFlow training compared to Vertex AI’s managed training.
C is incorrect because Managed Instance Groups are IaaS/SRE-focused for VM scaling, not a turnkey ML training platform with TensorFlow integration.
D is incorrect because Kubeflow Pipelines on GKE requires more setup and orchestration effort; Vertex AI Training is a higher-level, migration-friendly solution.



You have trained a text classification model in TensorFlow using AI Platform. You want to use the trained model for batch predictions on text data stored in BigQuery while minimizing computational overhead. What should you do?

  1. Export the model to BigQuery ML.
  2. Deploy and version the model on AI Platform.
  3. Use Dataflow with the SavedModel to read the data from BigQuery.
  4. Submit a batch prediction job on AI Platform that points to the model location in Cloud Storage.

Answer(s): A

Explanation:

Option A is correct because exporting the TensorFlow model as a BigQuery ML-supported model enables efficient batch predictions directly within BigQuery, minimizing data movement and overhead.
B) Deploying on AI Platform adds orchestration overhead for batch predictions and isn’t as seamless for in-query BigQuery data.
C) Dataflow with SavedModel introduces additional ETL/streaming complexity and does not natively optimize batch evaluation inside BigQuery.
D) AI Platform batch prediction requires Cloud Storage model assets and a separate job, increasing latency and cost compared to in-place BigQuery ML inference.



You work with a data engineering team that has developed a pipeline to clean your dataset and save it in a Cloud Storage bucket. You have created an ML model and want to use the data to refresh your model as soon as new data is available. As part of your CI/CD workflow, you want to automatically run a Kubeflow Pipelines

training job on Google Kubernetes Engine (GKE). How should you architect this workflow?

  1. Configure your pipeline with Dataflow, which saves the files in Cloud Storage. After the file is saved, start the training job on a GKE cluster.
  2. Use App Engine to create a lightweight python client that continuously polls Cloud Storage for new files. As soon as a file arrives, initiate the training job.
  3. Configure a Cloud Storage trigger to send a message to a Pub/Sub topic when a new file is available in a storage bucket. Use a Pub/Sub-triggered Cloud Function to start the training job on a GKE cluster.
  4. Use Cloud Scheduler to schedule jobs at a regular interval. For the first step of the job, check the timestamp of objects in your Cloud Storage bucket. If there are no new files since the last run, abort the job.

Answer(s): C

Explanation:

Option C is correct because it creates an event-driven, scalable CI/CD workflow: a Cloud Storage trigger emits when new data arrives, Pub/Sub delivers the event, and a Pub/Sub-triggered Cloud Function starts a Kubeflow Pipelines training job on GKE, enabling near-real-time retraining with minimal polling. This aligns with Google Cloud-native triggers, scalable message delivery, and seamless orchestration of Kubeflow on GKE.
A) Dataflow is not the correct trigger mechanism for initiating ML training; it is a data processing service, not a reliable event trigger for model retraining.
B) App Engine polling introduces latency and inefficiency; continuous polling is not event-driven and scales poorly.
D) Cloud Scheduler scheduling with periodic checks can miss bursts of new data and adds unnecessary latency compared to event-driven triggers.



You have a functioning end-to-end ML pipeline that involves tuning the hyperparameters of your ML model using AI Platform, and then using the best-tuned parameters for training. Hypertuning is taking longer than expected and is delaying the downstream processes. You want to speed up the tuning job without significantly compromising its effectiveness. Which actions should you take? (Choose two.)

  1. Decrease the number of parallel trials.
  2. Decrease the range of floating-point values.
  3. Set the early stopping parameter to TRUE.
  4. Change the search algorithm from Bayesian search to random search.
  5. Decrease the maximum number of trials during subsequent training phases.

Answer(s): C,E

Explanation:

Option C is correct because enabling early stopping allows hyperparameter trials to terminate when they stop improving, reducing wasted compute and speeding up the search. Option E is correct because lowering the maximum number of trials directly limits overall search time, accelerating the tuning phase while still seeking good configurations. Incorrect — A: Decreasing parallelism slows wall-clock progress, increasing time to completion. Incorrect — B: Narrowing the value range can reduce search space but risks missing optimal regions and may not preserve effectiveness. Incorrect — D: Replacing Bayesian with random search generally reduces efficiency and likelihood of finding high-quality configs, increasing total tuning time. INSUFFICIENT_KNOWLEDGE



  PDF
Viewing page 6 of 44
Viewing questions 41 - 48 out of 375 questions


Post your Comments and Discuss Google PROFESSIONAL MACHINE LEARNING ENGINEER exam prep with other Community members:

AI Tutor AI Tutor 👋 I’m here to help!