McAfee-Secured Website

Exam Code: AI-300

Exam Name: Operationalizing Machine Learning and Generative AI Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Machine Learning Operations Engineer Associate

100% Microsoft AI-300 Guaranteed Success With Testking Training Materials

Get Real AI-300 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

213 Questions & Answers with Testing Engine

"Operationalizing Machine Learning and Generative AI Solutions Exam", also known as AI-300 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-300 Exam Engine. Your AI-300 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-300 Practice Test 25

Question 240

A new training code version is deployed even though the pipeline passed using an older component cached result. What should be checked?

  1. Verify component version and reuse eligibility when code or environment changes.
  2. Enable appropriate pipeline step reuse or caching for unchanged inputs and configuration.
  3. Align upstream output and downstream input contracts with a compatible data type.
  4. Record pipeline component versions, parameters, inputs, outputs, and job run lineage.

Explanation

Correct Answer

A:

Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs.

Incorrect Answers

B:

Enable appropriate pipeline step reuse or caching for unchanged inputs and configuration. That operation would be relevant if: A preprocessing component is deterministic and expensive; rerunning an identical pipeline wastes time reproducing unchanged outputs. Its corresponding reason or verification is: Reuse can avoid recomputation when the component definition and inputs truly match an eligible prior execution. However the problem in this question is: A new training code version is deployed even though the pipeline passed using an older component cached result. The requirement instead calls for Verify component version and reuse eligibility when code or environment changes; the question-specific check is: Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs.

C:

Align upstream output and downstream input contracts with a compatible data type. A separate source case supports that approach when: A dataset preprocessing step produces a folder, but the trainer expects one tabular file. The supporting rationale or observable result is: Explicit file-versus-folder contracts prevent runtime failures caused by incompatible component output structures. This question describes a different issue: A new training code version is deployed even though the pipeline passed using an older component cached result. Address it through Verify component version and reuse eligibility when code or environment changes, with the required justification or evidence: Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs.

D:

Record pipeline component versions, parameters, inputs, outputs, and job run lineage. This is useful for the distinct issue: The operations team wants to compare end-to-end pipeline versions before rolling out a new model training flow. The relevant technical rationale or evidence for that issue is: Lineage provides a reproducible record of which executable stages and data produced the evaluated model version. Here the reported issue is: A new training code version is deployed even though the pipeline passed using an older component cached result. The relevant response is: Verify component version and reuse eligibility when code or environment changes. The first result cannot verify this different correction: Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs.

Question 241

A dataset preprocessing step produces a folder, but the trainer expects one tabular file. How can the pipeline avoid this mismatch?

  1. Publish a machine-readable metrics output and consume it in a decision step.
  2. Gate the training component on successful completion of the data-validation component.
  3. Add an evaluation-based approval gate before model registration or promotion.
  4. Align upstream output and downstream input contracts with a compatible data type.

Explanation

Correct Answer

D:

Explicit file-versus-folder contracts prevent runtime failures caused by incompatible component output structures.

Incorrect Answers

A:

The alternative, Publish a machine-readable metrics output and consume it in a decision step, belongs to a different technical condition: A model pipeline needs the evaluator's metrics to decide whether to deploy, but it currently relies on a log string. There the source-specific mechanism or proof is: Typed outputs support reliable conditional control without parsing unstructured stdout messages from prior components. The condition asked about here is: A dataset preprocessing step produces a folder, but the trainer expects one tabular file. A suitable correction is Align upstream output and downstream input contracts with a compatible data type; the targeted justification or result is: Explicit file-versus-folder contracts prevent runtime failures caused by incompatible component output structures.

B:

Gate the training component on successful completion of the data-validation component. That operation would be relevant if: A pipeline schedules training before a data-quality check completes, allowing invalid rows into the model. Its corresponding reason or verification is: Declared step dependencies prevent invalid datasets from reaching training when prerequisite validation fails. However the problem in this question is: A dataset preprocessing step produces a folder, but the trainer expects one tabular file. The requirement instead calls for Align upstream output and downstream input contracts with a compatible data type; the question-specific check is: Explicit file-versus-folder contracts prevent runtime failures caused by incompatible component output structures.

C:

Add an evaluation-based approval gate before model registration or promotion. A separate source case supports that approach when: Evaluation metrics show a regression but a subsequent model-registration step still runs. The supporting rationale or observable result is: Models failing documented quality thresholds should not be advanced solely because upstream jobs exited successfully. This question describes a different issue: A dataset preprocessing step produces a folder, but the trainer expects one tabular file. Address it through Align upstream output and downstream input contracts with a compatible data type, with the required justification or evidence: Explicit file-versus-folder contracts prevent runtime failures caused by incompatible component output structures.

Question 242

The operations team wants to compare end-to-end pipeline versions before rolling out a new model training flow. What should be captured?

  1. Create a pipeline DAG with explicit preprocessing, training, and evaluation component links.
  2. Verify component version and reuse eligibility when code or environment changes.
  3. Record pipeline component versions, parameters, inputs, outputs, and job run lineage.
  4. Gate the training component on successful completion of the data-validation component.

Explanation

Correct Answer

C:

Lineage provides a reproducible record of which executable stages and data produced the evaluated model version.

Incorrect Answers

A:

Create a pipeline DAG with explicit preprocessing, training, and evaluation component links. It addresses another case: An ML workflow manually runs preprocessing, training, and evaluation notebooks in order, and missed steps caused bad models. For that case, the distinguishing justification or acceptance check is: Pipeline dependencies ensure downstream work consumes declared outputs only after required upstream stages complete successfully. The present case instead states: The operations team wants to compare end-to-end pipeline versions before rolling out a new model training flow. It calls for Record pipeline component versions, parameters, inputs, outputs, and job run lineage. Showing success for the other case would not establish: Lineage provides a reproducible record of which executable stages and data produced the evaluated model version.

B:

The alternative, Verify component version and reuse eligibility when code or environment changes, belongs to a different technical condition: A new training code version is deployed even though the pipeline passed using an older component cached result. There the source-specific mechanism or proof is: Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs. The condition asked about here is: The operations team wants to compare end-to-end pipeline versions before rolling out a new model training flow. A suitable correction is Record pipeline component versions, parameters, inputs, outputs, and job run lineage; the targeted justification or result is: Lineage provides a reproducible record of which executable stages and data produced the evaluated model version.

D:

Gate the training component on successful completion of the data-validation component. That operation would be relevant if: A pipeline schedules training before a data-quality check completes, allowing invalid rows into the model. Its corresponding reason or verification is: Declared step dependencies prevent invalid datasets from reaching training when prerequisite validation fails. However the problem in this question is: The operations team wants to compare end-to-end pipeline versions before rolling out a new model training flow. The requirement instead calls for Record pipeline component versions, parameters, inputs, outputs, and job run lineage; the question-specific check is: Lineage provides a reproducible record of which executable stages and data produced the evaluated model version.

Question 243

A pipeline includes both a successful quality check and a failed security check; promotion still proceeds because it considers only accuracy. What change is required?

  1. Use isolated retryable components and bounded retries for transient failures.
  2. Make release progression depend on both quality and security policy checks.
  3. Gate the training component on successful completion of the data-validation component.
  4. Verify component version and reuse eligibility when code or environment changes.

Explanation

Correct Answer

B:

A multi-gate release must enforce every mandatory condition rather than treating one passing metric as comprehensive approval.

Incorrect Answers

A:

Use isolated retryable components and bounded retries for transient failures. This is useful for the distinct issue: A feature engineering step crashes intermittently due to a transient storage timeout, forcing a whole pipeline restart. The relevant technical rationale or evidence for that issue is: Step-level retry and isolation avoid unnecessarily repeating successful independent stages when a recoverable operation fails. Here the reported issue is: A pipeline includes both a successful quality check and a failed security check; promotion still proceeds because it considers only accuracy. The relevant response is: Make release progression depend on both quality and security policy checks. The first result cannot verify this different correction: A multi-gate release must enforce every mandatory condition rather than treating one passing metric as comprehensive approval.

C:

Gate the training component on successful completion of the data-validation component. It addresses another case: A pipeline schedules training before a data-quality check completes, allowing invalid rows into the model. For that case, the distinguishing justification or acceptance check is: Declared step dependencies prevent invalid datasets from reaching training when prerequisite validation fails. The present case instead states: A pipeline includes both a successful quality check and a failed security check; promotion still proceeds because it considers only accuracy. It calls for Make release progression depend on both quality and security policy checks. Showing success for the other case would not establish: A multi-gate release must enforce every mandatory condition rather than treating one passing metric as comprehensive approval.

D:

The alternative, Verify component version and reuse eligibility when code or environment changes, belongs to a different technical condition: A new training code version is deployed even though the pipeline passed using an older component cached result. There the source-specific mechanism or proof is: Cache reuse must reflect changes in executable dependencies; a new version should not silently consume stale outputs. The condition asked about here is: A pipeline includes both a successful quality check and a failed security check; promotion still proceeds because it considers only accuracy. A suitable correction is Make release progression depend on both quality and security policy checks; the targeted justification or result is: A multi-gate release must enforce every mandatory condition rather than treating one passing metric as comprehensive approval.

Question 244

Two AML jobs report different validation accuracy but were trained on different dataset versions. What should be aligned before comparison?

  1. Evaluate the candidates on the same held-out dataset and versioned metric definition.
  2. Validate model packaging, runtime compatibility, and endpoint operational behavior.
  3. Compare both models on the same independent evaluation split.
  4. Compare accuracy along with serving latency, resource cost, and throughput requirements.

Explanation

Correct Answer

A:

Consistent test conditions are necessary to attribute observed differences to models rather than changed inputs or measurement rules.

Incorrect Answers

B:

Validate model packaging, runtime compatibility, and endpoint operational behavior. A separate source case supports that approach when: Two experimental jobs record identical validation metrics but use different inference environments, and one fails at deployment. The supporting rationale or observable result is: Offline performance equivalence does not ensure that both artifacts can be served successfully in a production container. This question describes a different issue: Two AML jobs report different validation accuracy but were trained on different dataset versions. Address it through Evaluate the candidates on the same held-out dataset and versioned metric definition, with the required justification or evidence: Consistent test conditions are necessary to attribute observed differences to models rather than changed inputs or measurement rules.

C:

Compare both models on the same independent evaluation split. This is useful for the distinct issue: An analyst compares a model's training accuracy with another model's validation accuracy and declares the former better. The relevant technical rationale or evidence for that issue is: Training measurements can be optimistically biased and cannot be ranked fairly against another model's unseen-data performance. Here the reported issue is: Two AML jobs report different validation accuracy but were trained on different dataset versions. The relevant response is: Evaluate the candidates on the same held-out dataset and versioned metric definition. The first result cannot verify this different correction: Consistent test conditions are necessary to attribute observed differences to models rather than changed inputs or measurement rules.

D:

Compare accuracy along with serving latency, resource cost, and throughput requirements. It addresses another case: Two candidates have similar validation accuracy but one requires ten times the inference cost. For that case, the distinguishing justification or acceptance check is: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives. The present case instead states: Two AML jobs report different validation accuracy but were trained on different dataset versions. It calls for Evaluate the candidates on the same held-out dataset and versioned metric definition. Showing success for the other case would not establish: Consistent test conditions are necessary to attribute observed differences to models rather than changed inputs or measurement rules.

Question 245

An experiment's model has higher accuracy but extremely poor recall for a rare adverse event. Which evaluation view should guide selection?

  1. Compare accuracy along with serving latency, resource cost, and throughput requirements.
  2. Compare operating points under the same permitted false-positive rate and recall objective.
  3. Compute a common approved regression metric on both models' predictions.
  4. Inspect minority-class recall and precision, not accuracy alone.

Explanation

Correct Answer

D:

Class-imbalanced tasks require per-class measures that reveal missed events hidden by a majority-class metric.

Incorrect Answers

A:

Compare accuracy along with serving latency, resource cost, and throughput requirements. That operation would be relevant if: Two candidates have similar validation accuracy but one requires ten times the inference cost. Its corresponding reason or verification is: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives. However the problem in this question is: An experiment's model has higher accuracy but extremely poor recall for a rare adverse event. The requirement instead calls for Inspect minority-class recall and precision, not accuracy alone; the question-specific check is: Class-imbalanced tasks require per-class measures that reveal missed events hidden by a majority-class metric.

B:

Compare operating points under the same permitted false-positive rate and recall objective. A separate source case supports that approach when: A classification model has improved AUROC but its positive-class threshold was not tuned for the organization's false-positive budget. The supporting rationale or observable result is: A rank-order metric alone does not choose the threshold required for a business decision with constrained false positives. This question describes a different issue: An experiment's model has higher accuracy but extremely poor recall for a rare adverse event. Address it through Inspect minority-class recall and precision, not accuracy alone, with the required justification or evidence: Class-imbalanced tasks require per-class measures that reveal missed events hidden by a majority-class metric.

C:

Compute a common approved regression metric on both models' predictions. This is useful for the distinct issue: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. The relevant technical rationale or evidence for that issue is: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection. Here the reported issue is: An experiment's model has higher accuracy but extremely poor recall for a rare adverse event. The relevant response is: Inspect minority-class recall and precision, not accuracy alone. The first result cannot verify this different correction: Class-imbalanced tasks require per-class measures that reveal missed events hidden by a majority-class metric.

Question 246

A team wants to compare training jobs across hyperparameters, but every run uses an ambiguous name and has no logged configuration. What should be captured?

  1. Compare group-specific error metrics and investigate the disparate performance before release.
  2. Assess the selected model on a separate untouched test set before promotion.
  3. Log each run's parameters and canonical evaluation metrics in MLflow.
  4. Inspect the precision-recall tradeoff and business threshold on the minority class.

Explanation

Correct Answer

C:

Consistent experiment metadata allows meaningful comparison instead of relying on job display names alone.

Incorrect Answers

A:

The alternative, Compare group-specific error metrics and investigate the disparate performance before release, belongs to a different technical condition: A fairness audit finds different error rates across protected demographic groups, even though aggregate model accuracy is high. There the source-specific mechanism or proof is: Aggregate averages can conceal unequal outcomes across cohorts; responsible evaluation examines relevant subpopulations. The condition asked about here is: A team wants to compare training jobs across hyperparameters, but every run uses an ambiguous name and has no logged configuration. A suitable correction is Log each run's parameters and canonical evaluation metrics in MLflow; the targeted justification or result is: Consistent experiment metadata allows meaningful comparison instead of relying on job display names alone.

B:

Assess the selected model on a separate untouched test set before promotion. That operation would be relevant if: A team compares candidate models using a validation set that influenced repeated tuning choices. Its corresponding reason or verification is: A repeatedly consulted validation set can become an indirect optimization target, overstating performance on unseen data. However the problem in this question is: A team wants to compare training jobs across hyperparameters, but every run uses an ambiguous name and has no logged configuration. The requirement instead calls for Log each run's parameters and canonical evaluation metrics in MLflow; the question-specific check is: Consistent experiment metadata allows meaningful comparison instead of relying on job display names alone.

D:

Inspect the precision-recall tradeoff and business threshold on the minority class. A separate source case supports that approach when: A classifier's ROC AUC looks strong, but the operational objective is finding a handful of rare positive cases. The supporting rationale or observable result is: ROC performance can hide poor practical precision when positive cases are extremely rare in production. This question describes a different issue: A team wants to compare training jobs across hyperparameters, but every run uses an ambiguous name and has no logged configuration. Address it through Log each run's parameters and canonical evaluation metrics in MLflow, with the required justification or evidence: Consistent experiment metadata allows meaningful comparison instead of relying on job display names alone.

Question 247

Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. Which first step supports a fair ranking?

  1. Inspect minority-class recall and precision, not accuracy alone.
  2. Compute a common approved regression metric on both models' predictions.
  3. Assess the selected model on a separate untouched test set before promotion.
  4. Compare accuracy along with serving latency, resource cost, and throughput requirements.

Explanation

Correct Answer

B:

Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection.

Incorrect Answers

A:

Inspect minority-class recall and precision, not accuracy alone. It addresses another case: An experiment's model has higher accuracy but extremely poor recall for a rare adverse event. For that case, the distinguishing justification or acceptance check is: Class-imbalanced tasks require per-class measures that reveal missed events hidden by a majority-class metric. The present case instead states: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. It calls for Compute a common approved regression metric on both models' predictions. Showing success for the other case would not establish: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection.

C:

The alternative, Assess the selected model on a separate untouched test set before promotion, belongs to a different technical condition: A team compares candidate models using a validation set that influenced repeated tuning choices. There the source-specific mechanism or proof is: A repeatedly consulted validation set can become an indirect optimization target, overstating performance on unseen data. The condition asked about here is: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. A suitable correction is Compute a common approved regression metric on both models' predictions; the targeted justification or result is: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection.

D:

Compare accuracy along with serving latency, resource cost, and throughput requirements. That operation would be relevant if: Two candidates have similar validation accuracy but one requires ten times the inference cost. Its corresponding reason or verification is: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives. However the problem in this question is: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. The requirement instead calls for Compute a common approved regression metric on both models' predictions; the question-specific check is: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection.

Question 248

An analyst compares a model's training accuracy with another model's validation accuracy and declares the former better. What correction is required?

  1. Compare both models on the same independent evaluation split.
  2. Compare operating points under the same permitted false-positive rate and recall objective.
  3. Compute a common approved regression metric on both models' predictions.
  4. Estimate uncertainty using sufficient independent evaluation data or confidence intervals.

Explanation

Correct Answer

A:

Training measurements can be optimistically biased and cannot be ranked fairly against another model's unseen-data performance.

Incorrect Answers

B:

Compare operating points under the same permitted false-positive rate and recall objective. This is useful for the distinct issue: A classification model has improved AUROC but its positive-class threshold was not tuned for the organization's false-positive budget. The relevant technical rationale or evidence for that issue is: A rank-order metric alone does not choose the threshold required for a business decision with constrained false positives. Here the reported issue is: An analyst compares a model's training accuracy with another model's validation accuracy and declares the former better. The relevant response is: Compare both models on the same independent evaluation split. The first result cannot verify this different correction: Training measurements can be optimistically biased and cannot be ranked fairly against another model's unseen-data performance.

C:

Compute a common approved regression metric on both models' predictions. It addresses another case: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. For that case, the distinguishing justification or acceptance check is: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection. The present case instead states: An analyst compares a model's training accuracy with another model's validation accuracy and declares the former better. It calls for Compare both models on the same independent evaluation split. Showing success for the other case would not establish: Training measurements can be optimistically biased and cannot be ranked fairly against another model's unseen-data performance.

D:

The alternative, Estimate uncertainty using sufficient independent evaluation data or confidence intervals, belongs to a different technical condition: A newer model outperforms a baseline on a 50-row holdout by one percentage point. There the source-specific mechanism or proof is: Small noisy samples can produce apparent wins from random variation rather than reliable generalization improvement. The condition asked about here is: An analyst compares a model's training accuracy with another model's validation accuracy and declares the former better. A suitable correction is Compare both models on the same independent evaluation split; the targeted justification or result is: Training measurements can be optimistically biased and cannot be ranked fairly against another model's unseen-data performance.

Question 249

A classifier's ROC AUC looks strong, but the operational objective is finding a handful of rare positive cases. What should receive extra attention?

  1. Evaluate the candidates on the same held-out dataset and versioned metric definition.
  2. Compare group-specific error metrics and investigate the disparate performance before release.
  3. Validate model packaging, runtime compatibility, and endpoint operational behavior.
  4. Inspect the precision-recall tradeoff and business threshold on the minority class.

Explanation

Correct Answer

D:

ROC performance can hide poor practical precision when positive cases are extremely rare in production.

Incorrect Answers

A:

Evaluate the candidates on the same held-out dataset and versioned metric definition. A separate source case supports that approach when: Two AML jobs report different validation accuracy but were trained on different dataset versions. The supporting rationale or observable result is: Consistent test conditions are necessary to attribute observed differences to models rather than changed inputs or measurement rules. This question describes a different issue: A classifier's ROC AUC looks strong, but the operational objective is finding a handful of rare positive cases. Address it through Inspect the precision-recall tradeoff and business threshold on the minority class, with the required justification or evidence: ROC performance can hide poor practical precision when positive cases are extremely rare in production.

B:

Compare group-specific error metrics and investigate the disparate performance before release. This is useful for the distinct issue: A fairness audit finds different error rates across protected demographic groups, even though aggregate model accuracy is high. The relevant technical rationale or evidence for that issue is: Aggregate averages can conceal unequal outcomes across cohorts; responsible evaluation examines relevant subpopulations. Here the reported issue is: A classifier's ROC AUC looks strong, but the operational objective is finding a handful of rare positive cases. The relevant response is: Inspect the precision-recall tradeoff and business threshold on the minority class. The first result cannot verify this different correction: ROC performance can hide poor practical precision when positive cases are extremely rare in production.

C:

Validate model packaging, runtime compatibility, and endpoint operational behavior. It addresses another case: Two experimental jobs record identical validation metrics but use different inference environments, and one fails at deployment. For that case, the distinguishing justification or acceptance check is: Offline performance equivalence does not ensure that both artifacts can be served successfully in a production container. The present case instead states: A classifier's ROC AUC looks strong, but the operational objective is finding a handful of rare positive cases. It calls for Inspect the precision-recall tradeoff and business threshold on the minority class. Showing success for the other case would not establish: ROC performance can hide poor practical precision when positive cases are extremely rare in production.

Question 250

Two candidates have similar validation accuracy but one requires ten times the inference cost. How should deployment selection account for operations?

  1. Compute a common approved regression metric on both models' predictions.
  2. Use the same averaging method and label definitions for every candidate's F1 metric.
  3. Compare accuracy along with serving latency, resource cost, and throughput requirements.
  4. Compare operating points under the same permitted false-positive rate and recall objective.

Explanation

Correct Answer

C:

A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives.

Incorrect Answers

A:

Compute a common approved regression metric on both models' predictions. A separate source case supports that approach when: Two regression models are evaluated on identical examples, but one team reports RMSE and the other reports MAE. The supporting rationale or observable result is: Different loss metrics penalize errors differently; side-by-side ranking needs a consistent criterion agreed before selection. This question describes a different issue: Two candidates have similar validation accuracy but one requires ten times the inference cost. Address it through Compare accuracy along with serving latency, resource cost, and throughput requirements, with the required justification or evidence: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives.

B:

Use the same averaging method and label definitions for every candidate's F1 metric. This is useful for the distinct issue: A model selection pipeline ranks candidates by F1 but the scoring code logs micro-average for one and macro-average for another. The relevant technical rationale or evidence for that issue is: Metrics with the same name can represent different aggregation rules; inconsistent definitions invalidate direct comparisons. Here the reported issue is: Two candidates have similar validation accuracy but one requires ten times the inference cost. The relevant response is: Compare accuracy along with serving latency, resource cost, and throughput requirements. The first result cannot verify this different correction: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives.

D:

Compare operating points under the same permitted false-positive rate and recall objective. It addresses another case: A classification model has improved AUROC but its positive-class threshold was not tuned for the organization's false-positive budget. For that case, the distinguishing justification or acceptance check is: A rank-order metric alone does not choose the threshold required for a business decision with constrained false positives. The present case instead states: Two candidates have similar validation accuracy but one requires ten times the inference cost. It calls for Compare accuracy along with serving latency, resource cost, and throughput requirements. Showing success for the other case would not establish: A slightly better offline score can be unsuitable when its infrastructure requirements violate production service objectives.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests