100% Latest & Updated Google Professional Machine Learning Engineer Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 120
An ML platform team is resolving 2 independent workstreams. Workstream A: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. Workstream B: A prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. Select exactly TWO actions, one directly addressing each separate workstream.
- Use a supported managed Spark pipeline with skew-aware partitioning and join strategy.
- Build event-time-correct historical feature joins from retained feature records.
- Validate types and mandatory columns during ingestion and quarantine incompatible files.
- Implement idempotent event processing and stable deduplication keys in the pipeline.
- Use rolling-origin or chronological evaluation windows that predict future dates from past data.
Explanation
Correct Answers
B:Build event-time-correct historical feature joins from retained feature records. Check a sample of historical rows for features recorded after their label timestamp. Storing today's latest feature values is not enough to reconstruct the past.
E:Use rolling-origin or chronological evaluation windows that predict future dates from past data. Backtest at several forecast origins and inspect uncertainty by season. A single short validation window may miss changing seasonal demand patterns.
Incorrect Answers
A:Using a supported managed Spark pipeline with skew-aware partitioning and join strategy is relevant when a 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory; it does not establish the result required by this workstream.
C:The approach “Validate types and mandatory columns during ingestion and quarantine incompatible files” is relevant when an upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows; it does not establish the result required by this workstream.
D:The approach “Implement idempotent event processing and stable deduplication keys in the pipeline” is relevant when a pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur; it does not establish the result required by this workstream.
Question 121
An engineer is choosing an appropriate design for this workload: Researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts. Which action best addresses the actual technical requirement?
- Run a Dataflow streaming pipeline using event-time windows and appropriate lateness policy.
- Deidentify or tokenize protected fields before creating the shared training dataset.
- Validate types and mandatory columns during ingestion and quarantine incompatible files.
- Apply correct date partition predicates and select only necessary fields.
Explanation
Correct Answer
B:Deidentify or tokenize protected fields before creating the shared training dataset. Review deidentification findings and who can recover token mappings. Tokenization alone may not prevent reidentification from rare attribute combinations.
Incorrect Answers
A:That measure applies when a mobile-app stream delivers delayed purchase events and processing-time aggregation misassigns revenue to the wrong business hour. The required intervention here is different: Deidentify or tokenize protected fields before creating the shared training dataset. The need is researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts.
C:That measure applies when an upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows. The required intervention here is different: Deidentify or tokenize protected fields before creating the shared training dataset. The need is researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts.
D:That measure applies when a training feature query scans seven years of purchase logs although the model only needs the latest 90 days. The required intervention here is different: Deidentify or tokenize protected fields before creating the shared training dataset. The need is researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts.
Question 122
An ML platform team is resolving 2 independent workstreams. Workstream A: An upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows. Workstream B: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. Select exactly TWO actions, one directly addressing each separate workstream.
- Apply correct date partition predicates and select only necessary fields.
- Use rolling-origin or chronological evaluation windows that predict future dates from past data.
- Use a supported managed Spark pipeline with skew-aware partitioning and join strategy.
- Validate types and mandatory columns during ingestion and quarantine incompatible files.
- Deidentify or tokenize protected fields before creating the shared training dataset.
Explanation
Correct Answers
A:Apply correct date partition predicates and select only necessary fields. Compare bytes processed and output-row equality before and after optimization. A predicate on an incompatible expression may not prune target partitions.
D:Validate types and mandatory columns during ingestion and quarantine incompatible files. Track rejected count, declared schema version and downstream row completeness. Coercing every malformed value may conceal source errors and alter statistics.
Incorrect Answers
B:Using rolling-origin or chronological evaluation windows that predict future dates from past data is relevant when a prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days; it does not establish the result required by this workstream.
C:Using a supported managed Spark pipeline with skew-aware partitioning and join strategy is relevant when a 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory; it does not establish the result required by this workstream.
E:The approach “Deidentify or tokenize protected fields before creating the shared training dataset” is relevant when researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts; it does not establish the result required by this workstream.
Question 123
An ML engineering team must resolve this production problem: A data scientist calculates a median replacement value using all data before creating a train/test split. Which action best addresses the actual technical requirement?
- Run a Dataflow streaming pipeline using event-time windows and appropriate lateness policy.
- Deidentify or tokenize protected fields before creating the shared training dataset.
- Fit the imputer on the training partition only and reuse its frozen parameters.
- Build event-time-correct historical feature joins from retained feature records.
Explanation
Correct Answer
C:Fit the imputer on the training partition only and reuse its frozen parameters. Compare train-fitted transform parameters across validation and production code. A missingness indicator may be important if nulls carry signal.
Incorrect Answers
A:That measure applies when a mobile-app stream delivers delayed purchase events and processing-time aggregation misassigns revenue to the wrong business hour. The required intervention here is different: Fit the imputer on the training partition only and reuse its frozen parameters. The need is a data scientist calculates a median replacement value using all data before creating a train/test split.
B:That measure applies when researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts. The required intervention here is different: Fit the imputer on the training partition only and reuse its frozen parameters. The need is a data scientist calculates a median replacement value using all data before creating a train/test split.
D:That measure applies when a loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The required intervention here is different: Fit the imputer on the training partition only and reuse its frozen parameters. The need is a data scientist calculates a median replacement value using all data before creating a train/test split.
Question 124
An ML platform team is resolving 2 independent workstreams. Workstream A: A prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. Workstream B: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. Select exactly TWO actions, one directly addressing each separate workstream.
- Store images in governed Cloud Storage buckets with versioned manifests and protected labels.
- Build event-time-correct historical feature joins from retained feature records.
- Implement idempotent event processing and stable deduplication keys in the pipeline.
- Use a supported managed Spark pipeline with skew-aware partitioning and join strategy.
- Use rolling-origin or chronological evaluation windows that predict future dates from past data.
Explanation
Correct Answers
C:Implement idempotent event processing and stable deduplication keys in the pipeline. Replay duplicate events and compare final feature counts to the unique-event baseline. Delivery semantics alone do not guarantee exactly-once side effects in every external sink.
E:Use rolling-origin or chronological evaluation windows that predict future dates from past data. Backtest at several forecast origins and inspect uncertainty by season. A single short validation window may miss changing seasonal demand patterns.
Incorrect Answers
A:The approach “Store images in governed Cloud Storage buckets with versioned manifests and protected labels” is relevant when a visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images; it does not establish the result required by this workstream.
B:Building event-time-correct historical feature joins from retained feature records is relevant when a loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier; it does not establish the result required by this workstream.
D:Using a supported managed Spark pipeline with skew-aware partitioning and join strategy is relevant when a 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory; it does not establish the result required by this workstream.
Question 125
A Google Cloud change request describes the following issue: A visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images. Which action best addresses the actual technical requirement?
- Use rolling-origin or chronological evaluation windows that predict future dates from past data.
- Implement idempotent event processing and stable deduplication keys in the pipeline.
- Store images in governed Cloud Storage buckets with versioned manifests and protected labels.
- Build event-time-correct historical feature joins from retained feature records.
Explanation
Correct Answer
C:Store images in governed Cloud Storage buckets with versioned manifests and protected labels. Audit manifest object generations, labels and restricted image access. A bucket path alone cannot guarantee an immutable dataset snapshot.
Incorrect Answers
A:That measure applies when a prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. The required intervention here is different: Store images in governed Cloud Storage buckets with versioned manifests and protected labels. The need is a visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images.
B:That measure applies when a pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. The required intervention here is different: Store images in governed Cloud Storage buckets with versioned manifests and protected labels. The need is a visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images.
D:That measure applies when a loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The required intervention here is different: Store images in governed Cloud Storage buckets with versioned manifests and protected labels. The need is a visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images.
Question 126
An incident review is assessing this observed failure: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. Which observation most directly identifies the cause or requirements before a change?
- Examine temporal boundaries and seasonal coverage of all train and evaluation records.
- Compare feature event timestamps with each original prediction date.
- Inspect Cloud Storage object inventory, metadata schema and data-access permissions.
- Inspect bytes scanned and effective partition filters in the BigQuery query plan.
Explanation
Correct Answer
D:The evidence for this case shows situation: Inspect bytes scanned and effective partition filters in the BigQuery query plan. It directly examines the reported situation: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. The corresponding corrective action is: Apply correct date partition predicates and select only necessary fields.
Incorrect Answers
A:This evidence concerns a separate issue: A prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. It does not establish what is happening in the present case: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. The diagnostic evidence to examine here is Inspect bytes scanned and effective partition filters in the BigQuery query plan.
B:This evidence concerns a separate issue: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. It does not establish what is happening in the present case: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. The diagnostic evidence to examine here is Inspect bytes scanned and effective partition filters in the BigQuery query plan.
C:This evidence concerns a separate issue: A visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images. It does not establish what is happening in the present case: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. The diagnostic evidence to examine here is Inspect bytes scanned and effective partition filters in the BigQuery query plan.
Question 127
An ML platform team is resolving 3 independent workstreams. Workstream A: A mobile-app stream delivers delayed purchase events and processing-time aggregation misassigns revenue to the wrong business hour. Workstream B: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. Workstream C: A data scientist calculates a median replacement value using all data before creating a train/test split. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Compare feature event timestamps with each original prediction date.
- Examine temporal boundaries and seasonal coverage of all train and evaluation records.
- Inspect event timestamps, watermark delays and current window settings.
- Inspect samples for PII and access policies on the raw Cloud Storage locations.
- Inspect preprocessing fit timing and partition membership.
- Inspect Spark shuffle spill and key-frequency distribution on a sample.
Explanation
Correct Answers
A:The evidence for this case shows situation: Compare feature event timestamps with each original prediction date. It directly examines the reported situation: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The corresponding corrective action is: Build event-time-correct historical feature joins from retained feature records.
C:The evidence for this case shows situation: Inspect event timestamps, watermark delays and current window settings. It directly examines the reported situation: A mobile-app stream delivers delayed purchase events and processing-time aggregation misassigns revenue to the wrong business hour. The corresponding corrective action is: Run a Dataflow streaming pipeline using event-time windows and appropriate lateness policy.
E:The evidence for this case shows situation: Inspect preprocessing fit timing and partition membership. It directly examines the reported situation: A data scientist calculates a median replacement value using all data before creating a train/test split. The corresponding corrective action is: Fit the imputer on the training partition only and reuse its frozen parameters.
Incorrect Answers
B:The approach “Examine temporal boundaries and seasonal coverage of all train and evaluation records” is relevant when a prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days; it does not establish the result required by this workstream.
D:Inspecting samples for PII and access policies on the raw Cloud Storage locations is relevant when researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts; it does not establish the result required by this workstream.
F:Inspecting Spark shuffle spill and key-frequency distribution on a sample is relevant when a 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory; it does not establish the result required by this workstream.
Question 128
Before changing the system, the team investigates the following behavior: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. Which observation most directly identifies the cause or requirements before a change?
- Compare feature event timestamps with each original prediction date.
- Inspect message IDs, event IDs and resulting duplicate feature writes.
- Examine temporal boundaries and seasonal coverage of all train and evaluation records.
- Compare observed schema and rejected records with the contract used for training.
Explanation
Correct Answer
B:The evidence for this case shows situation: Inspect message IDs, event IDs and resulting duplicate feature writes. It directly examines the reported situation: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. The corresponding corrective action is: Implement idempotent event processing and stable deduplication keys in the pipeline.
Incorrect Answers
A:This evidence concerns a separate issue: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. It does not establish what is happening in the present case: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. The diagnostic evidence to examine here is Inspect message IDs, event IDs and resulting duplicate feature writes.
C:This evidence concerns a separate issue: A prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. It does not establish what is happening in the present case: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. The diagnostic evidence to examine here is Inspect message IDs, event IDs and resulting duplicate feature writes.
D:This evidence concerns a separate issue: An upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows. It does not establish what is happening in the present case: A pubsub-triggered enrichment job updates a customer feature twice whenever message delivery retries occur. The diagnostic evidence to examine here is Inspect message IDs, event IDs and resulting duplicate feature writes.
Question 129
An ML platform team is resolving 3 independent workstreams. Workstream A: A 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory. Workstream B: An upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows. Workstream C: A visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Inspect preprocessing fit timing and partition membership.
- Inspect Spark shuffle spill and key-frequency distribution on a sample.
- Compare observed schema and rejected records with the contract used for training.
- Inspect Cloud Storage object inventory, metadata schema and data-access permissions.
- Inspect samples for PII and access policies on the raw Cloud Storage locations.
- Examine temporal boundaries and seasonal coverage of all train and evaluation records.
Explanation
Correct Answers
B:The evidence for this case shows situation: Inspect Spark shuffle spill and key-frequency distribution on a sample. It directly examines the reported situation: A 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory. The corresponding corrective action is: Use a supported managed Spark pipeline with skew-aware partitioning and join strategy.
C:The evidence for this case shows situation: Compare observed schema and rejected records with the contract used for training. It directly examines the reported situation: An upstream app changes a numeric purchase amount column into a string and downstream feature training silently drops rows. The corresponding corrective action is: Validate types and mandatory columns during ingestion and quarantine incompatible files.
D:The evidence for this case shows situation: Inspect Cloud Storage object inventory, metadata schema and data-access permissions. It directly examines the reported situation: A visual quality model trains on millions of photos with metadata in inconsistent filenames and some photos contain sensitive images. The corresponding corrective action is: Store images in governed Cloud Storage buckets with versioned manifests and protected labels.
Incorrect Answers
A:Inspecting preprocessing fit timing and partition membership is relevant when a data scientist calculates a median replacement value using all data before creating a train/test split; it does not establish the result required by this workstream.
E:Inspecting samples for PII and access policies on the raw Cloud Storage locations is relevant when researchers need statistics from customer service logs but raw identifiers are restricted from broad analysis accounts; it does not establish the result required by this workstream.
F:The approach “Examine temporal boundaries and seasonal coverage of all train and evaluation records” is relevant when a prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days; it does not establish the result required by this workstream.
Question 130
A platform team needs diagnostic evidence for this condition: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. Which observation most directly identifies the cause or requirements before a change?
- Inspect Spark shuffle spill and key-frequency distribution on a sample.
- Examine temporal boundaries and seasonal coverage of all train and evaluation records.
- Compare feature event timestamps with each original prediction date.
- Inspect bytes scanned and effective partition filters in the BigQuery query plan.
Explanation
Correct Answer
C:The evidence for this case shows situation: Compare feature event timestamps with each original prediction date. It directly examines the reported situation: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The corresponding corrective action is: Build event-time-correct historical feature joins from retained feature records.
Incorrect Answers
A:This evidence concerns a separate issue: A 12 TB dataset requires joins with skewed keys and an in-memory pandas prototype runs out of memory. It does not establish what is happening in the present case: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The diagnostic evidence to examine here is Compare feature event timestamps with each original prediction date.
B:This evidence concerns a separate issue: A prediction service forecasts next week's demand but its accuracy was measured on randomly shuffled days. It does not establish what is happening in the present case: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The diagnostic evidence to examine here is Compare feature event timestamps with each original prediction date.
D:This evidence concerns a separate issue: A training feature query scans seven years of purchase logs although the model only needs the latest 90 days. It does not establish what is happening in the present case: A loan model has inflated test accuracy because it joins the customer's latest balance to examples from six months earlier. The diagnostic evidence to examine here is Compare feature event timestamps with each original prediction date.
Professional Machine Learning Engineer Practice Tests