Browse Certification Practice Tests by Exam Family

Microsoft DP-100 Cheat Sheet: Azure Data Scientist

Review the legacy Microsoft Azure Data Scientist (DP-100) route, machine-learning lifecycle concepts, model operations, evaluation, monitoring, and the current AI-300 replacement path.

DP-100 is an older Azure Data Scientist route. Use this cheat sheet to preserve useful model-lifecycle concepts while checking whether AI-300 is now the better Microsoft route for machine-learning operations.

Use this as a route check. Review the older data-science scope, then compare the current AI-300 machine-learning operations page.

Open DP-100 exam page Compare AI-300

Exam snapshot

FieldDetail
IssuerMicrosoft
Legacy routeAzure Data Scientist Associate
Exam codeDP-100
Current statusReplacement guidance
Closest current examAI-300 Machine Learning Operations Engineer Associate
IT Mastery statusExam-selection sample question page

Transition map

Older DP-100 areaWhat still mattersCurrent-route trap
Data preparationData quality, feature preparation, splits, leakage, and repeatabilityTreating notebook work as production-ready operations
Model trainingExperiments, metrics, compute, reproducibility, and model selectionOptimizing a model without deployment and monitoring plans
Model deploymentEndpoints, scaling, security, and release controlsIgnoring rollback and version management
EvaluationMetrics, test sets, drift, bias, and fitness for purposeChoosing accuracy without understanding the objective
OperationsMonitoring, retraining, performance, incidents, and governanceStudying only model creation when AI-300 emphasizes operations

Must-know distinctions

DistinctionHow to decide
DP-100 vs AI-300DP-100 is the older data-science route; AI-300 focuses more on ML operations and GenAIOps.
Training metric vs business metricTraining metrics evaluate model behavior; business metrics decide operational value.
Data drift vs concept driftData drift changes input distribution; concept drift changes the relationship between inputs and outcomes.
Batch inference vs online endpointBatch processes groups of records; online endpoints serve real-time predictions.
Experiment tracking vs monitoringTracking records development; monitoring watches deployed behavior.

High-yield checklist

  • Confirm whether DP-100 is still the exam you can actually schedule.
  • Map old model-development notes to AI-300 lifecycle and operations expectations.
  • Check data quality, leakage, and reproducibility before trusting model metrics.
  • Choose metrics that match the business problem.
  • Plan versioning, deployment, monitoring, and rollback.
  • Include security, identity, and access controls around data and endpoints.
  • Treat evaluation as ongoing, not only pre-release.

Common traps

  • Studying notebooks without deployment operations.
  • Choosing a metric that does not match the error cost.
  • Ignoring drift after release.
  • Treating retraining as automatic without validation.
  • Forgetting endpoint security and access control.

Practice strategy

Use the DP-100 exam page to check old data-science concepts, then move to AI-300 if your target is current ML operations. Classify misses by lifecycle stage: data, training, evaluation, deployment, monitoring, or governance.

Revised on Monday, May 25, 2026