Does this free GCP PMLE practice exam include explanations?

Yes. Each question includes detailed option-by-option explanations to help you review the Google Cloud ML engineering concept being tested.

Free GCP PMLE mock test – Exam 2

Google Professional Machine Learning Engineer

Free Google Professional Machine Learning Engineer practice exam for GCP certification prep.

Q: Is this Google Professional Machine Learning Engineer practice exam free?

Yes. This Google Professional Machine Learning Engineer mock test is free to open, answer, and retake.

Use this free GCP Professional Machine Learning Engineer practice exam to review Vertex AI, Model Garden, BigQuery ML, MLOps pipelines, model monitoring, responsible AI, and production ML design decisions.

10 exam-style questions Free GCP PMLE mock test Detailed option explanations No signup required

Start Practice Exam 2 below. Answer each question first, then review the detailed explanation for every option to understand the Google Cloud ML engineering pattern behind the answer.

Google Professional Machine Learning Engineer Practice Exam 2

Free Google Professional Machine Learning Engineer practice exam 2 with GCP MLOps, Vertex AI, AutoML, BigQuery ML, and training infrastructure scenarios.

1 / 10

Question

A credit card fraud model built with AutoML misses many fraudulent transactions. False negatives are much more costly than false positives, and fraudulent examples are rare in the dataset. Which two changes are most likely to improve detection? (Choose TWO.)

Which option meets the requirement?

A. Increase the score threshold so fewer transactions are classified as fraud.

B. Decrease the score threshold so the model flags more potential fraud cases.

C. Add more representative positive fraud examples to the training set.

D. Add only more negative legitimate-transaction examples to the training set.

E. Reduce the maximum training node hours to shorten training.

2 / 10

Question

You manage a Vertex AI model in production and want to automatically start retraining when real-world performance or data distribution deteriorates. What should you implement?

Which option meets the requirement?

A. Use Vertex Explainable AI alone and retrain whenever feature attributions look different.

B. Collect occasional end-user comments and manually decide when to retrain.

C. Create a Vertex AI Model Monitoring job with alert thresholds and trigger a retraining pipeline when drift or performance metrics exceed the configured limits.

D. Evaluate the model only on a fixed historical validation dataset once per month.

3 / 10

Question

A biotech team trains large TensorFlow models with custom C++ operations, 20 GB of weights and embeddings, very large batches, and high data-transfer needs. Which hardware choice is most appropriate?

Which option meets the requirement?

A. A small CPU-only cluster because custom TensorFlow ops cannot use accelerators.

B. A GPU-optimized A2 cluster with multiple NVIDIA A100 GPUs, high GPU memory, high RAM, and high-bandwidth interconnect.

C. A single older TPU v2 device with limited host memory.

D. General-purpose high-CPU VMs without GPUs.

4 / 10

Question

You are building a Vertex AI pipeline for an XGBoost classifier using a BigQuery table. The workflow must split data 65/35, perform feature engineering, log evaluation metrics, and compare models across pipeline executions. Which approach should you use?

Which option meets the requirement?

A. Use Vertex AI Pipelines components for the split and feature engineering, enable metric autologging in training, and compare pipeline runs with Vertex AI Experiments and ML Metadata lineage.

B. Use only BigQuery table snapshots and compare output files manually in Cloud Storage.

C. Train all models from a notebook and paste metrics into a spreadsheet.

D. Use Cloud Logging text searches as the primary experiment comparison tool.

5 / 10

Question

A custom Python image classifier trains on several hundred thousand small images. Training will run on Vertex AI. You want maximum input throughput and minimal extra training code. How should the images be stored and read?

Which option meets the requirement?

A. Store individual image files in Cloud Storage and read each object directly during training.

B. Store the images in Cloud Storage as sharded serialized records such as TFRecord and read those records during training.

C. Store individual files on Cloud Filestore and read through an NFS mount for all training jobs.

D. Store images in Firestore documents and stream them one at a time.

6 / 10

Question

A company wants to detect training-serving skew for a model deployed to Vertex AI. They have a baseline training dataset and want alerts when production feature distributions differ materially from training. Which service should they configure?

Which option meets the requirement?

A. Vertex AI Model Monitoring with a training baseline and skew/drift thresholds.

B. Cloud Build triggers on every Git commit.

C. Cloud CDN cache invalidation rules.

D. Cloud NAT logs exported to BigQuery.

7 / 10

Question

A team stores raw tabular training data in BigQuery and needs reproducible feature engineering for both training and serving. They want transformations to be part of the model pipeline rather than copied manually into application code. What should they use?

Which option meets the requirement?

A. TensorFlow Transform or an equivalent pipeline transformation component that records and reuses preprocessing logic consistently.

B. Manual spreadsheet formulas that analysts copy into each training run.

C. Only model comments that describe the transformations.

D. A dashboard filter in Looker Studio.

8 / 10

Question

A batch training job repeatedly reads a very large dataset from Cloud Storage. You need to reduce startup time and keep experiments reproducible while avoiding manual dependency installation on each run. Which packaging approach is best?

Which option meets the requirement?

A. Build a custom training container with pinned dependencies and submit it to Vertex AI Training.

B. SSH into a VM before every run and install packages interactively.

C. Install dependencies from the public internet at the start of every training script without pinning versions.

D. Store dependencies in a README and ask each engineer to recreate the environment manually.

9 / 10

Question

A model owner needs to compare two deployed model versions using business KPIs such as conversion rate, not just offline accuracy. The current model runs behind a Vertex AI endpoint. Which production test is most appropriate?

Which option meets the requirement?

A. Use Vertex AI endpoint traffic splitting and monitor KPI differences between the model versions.

B. Choose the model with the lower training loss and skip production validation.

C. Run a single offline batch prediction job and assume user behavior will match it.

D. Deploy both models to unrelated projects and manually alternate DNS records.

10 / 10

Question

An ML platform team wants a central view of datasets, pipeline runs, trained models, metrics, and which artifacts produced each deployed model. Which Vertex AI capability should they rely on?

Which option meets the requirement?

A. Vertex ML Metadata and pipeline lineage tracking.

B. Cloud DNS private zones.

C. Cloud Armor security policies.

D. Cloud Storage object lifecycle rules only.

Your score is

The average score is 0%

What Practice Exam 2 covers

AutoML score thresholds and imbalanced classification
Vertex AI Model Monitoring and automated retraining triggers
GPU, TPU, and training hardware selection
Vertex AI Pipelines, Experiments, and ML Metadata
Cloud Storage serialized records and high-throughput training

Who should take this free mock test

Use this free Google Professional Machine Learning Engineer practice exam if you are preparing for the GCP PMLE certification and want focused practice with detailed answer explanations.

Browse GCP practice exams Visit DevOpsEngine

FAQ

Is this Google Professional Machine Learning Engineer practice exam free?

Yes. This GCP PMLE mock test is free to open and retake for certification study.

Does the free practice exam include explanations?

Yes. Each question includes detailed explanations for the correct and incorrect options so you can learn the service tradeoff, not just memorize an answer.

How should I review missed questions?

Read every option explanation, map the scenario to the relevant Google Cloud service, then revisit the matching Vertex AI, BigQuery ML, MLOps, or model monitoring topic before retaking the free practice exam.

Free Google Professional Machine Learning Engineer Practice Exam 2 | GCP PMLE Mock Test

Free Google Professional Machine Learning Engineer practice exam for GCP certification prep.

Question

Correct Answer: B, C

Explanation

Question

Correct Answer: C

Explanation

Question

Correct Answer: B

Explanation

Question

Correct Answer: A

Explanation

Question

Correct Answer: B

Explanation

Question

Correct Answer: A

Explanation

Question

Correct Answer: A

Explanation

Question

Correct Answer: A

Explanation

Question

Correct Answer: A

Explanation

Question

Correct Answer: A

Explanation

What Practice Exam 2 covers

Who should take this free mock test

FAQ

Is this Google Professional Machine Learning Engineer practice exam free?

Does the free practice exam include explanations?

How should I review missed questions?