Microsoft retired DP-100, Designing and Implementing a Data Science Solution on Azure, on June 1, 2026. It is no longer available to book; historic price and scheduling advice is not current. This guide preserves educational topics including Azure Machine Learning, experiments, data preparation, AutoML, training, deployment and responsible AI. The Microsoft AI-300 operations route is related but assesses different MLOps and GenAIOps capabilities rather than being a direct rename of DP-100.
Azure Machine Learning Workspace
The Azure Machine Learning workspace is the central organizational unit for all resources, experiments, and assets within an Azure Machine Learning environment. Candidates must understand how to create and configure workspaces, manage associated resources including storage accounts and key vaults, and control access through role-based access control policies. The workspace serves as the container that holds compute resources, datasets, models, pipelines, and environments needed to execute machine learning workflows at scale.
The exam tests detailed knowledge of workspace components including compute instances for interactive development, compute clusters for scalable training jobs, and inference clusters for deploying models as web services. Candidates must understand the differences between these compute types, when to use each, and how to configure them appropriately for specific workload requirements. Knowledge of how workspaces support team collaboration through shared assets, access controls, and linked services is also tested because real-world machine learning projects rarely involve a single practitioner working in complete isolation.
Data Management and Preparation Data management within Azure Machine Learning is a substantial exam topic that covers datastores, datasets, and data ingestion patterns used in professional machine learning workflows. Datastores represent connections to external data storage services such as Azure Blob Storage, Azure Data Lake Storage, and Azure SQL Database, allowing Azure Machine Learning to access training data without moving it into the workspace itself. Candidates must understand how to register datastores, authenticate connections securely, and reference datastore paths within training scripts and pipelines.
Azure Machine Learning datasets provide a higher-level abstraction for working with data that includes versioning, profiling, and lineage tracking capabilities. Tabular datasets and file datasets serve different purposes and are appropriate for different types of machine learning tasks. The exam tests knowledge of how to create, register, and consume datasets within training jobs, how dataset versioning supports reproducible experiments, and how data drift monitoring identifies when the statistical properties of incoming data change in ways that may degrade deployed model performance.
Training Jobs and Experiments
Running training jobs and tracking experiments is one of the most heavily weighted domains in the DP-100 exam. Candidates must demonstrate proficiency with Azure Machine Learning’s job submission system, including how to configure training jobs with appropriate compute targets, environment definitions, and input data references. The experiment tracking capabilities of Azure Machine Learning, which log metrics, parameters, and artifacts from each training run, are central to comparing model performance across different configurations and algorithm choices.
The exam covers both the Python SDK and the Azure Machine Learning CLI for submitting and managing training jobs, reflecting the reality that practitioners use both interfaces depending on their workflow preferences and automation requirements. Candidates must understand how to write training scripts that log metrics using the MLflow tracking integration built into Azure Machine Learning, how to register trained models from completed jobs, and how to retrieve and compare metrics across multiple experiment runs programmatically.
AutoML Capabilities Tested Automated Machine Learning, commonly referred to as AutoML, is a significant topic in the DP-100 exam because it represents one of Azure Machine Learning’s most distinctive and widely used capabilities. AutoML automates the process of algorithm selection, feature engineering, and hyperparameter tuning by running many model training iterations in parallel and identifying the best-performing configuration based on a specified metric. Candidates must understand how to configure AutoML experiments for classification, regression, and time series forecasting tasks.
The exam tests knowledge of AutoML configuration options including primary metrics, exit criteria, featurization settings, and blocked algorithms. Candidates must also understand how to interpret AutoML results, including how to examine the models generated during an AutoML run, retrieve the best model, and register it for deployment. Understanding the guardrails that AutoML applies to detect and handle common data quality problems such as class imbalance and missing values is also tested as part of the broader AutoML knowledge domain.
Pipeline Development Skills
Azure Machine Learning pipelines allow data scientists to define and execute multi-step workflows that chain data preparation, training, evaluation, and registration steps into reproducible, automated sequences. The DP-100 exam places significant emphasis on pipeline development because pipelines are central to how professional machine learning workflows operate in production environments. Candidates must understand how to create pipelines using the Python SDK, how to define pipeline steps with appropriate input and output dependencies, and how to schedule or trigger pipeline runs based on time or data availability conditions.
Pipeline components, which are reusable, versioned building blocks for pipeline steps, are an important topic within the pipeline domain. Candidates must understand how to create and register components, how to compose pipelines from components, and how component-based pipelines improve reusability and maintainability compared to monolithic pipeline scripts. The exam also covers parallel job steps that distribute data processing across multiple compute nodes, which is essential knowledge for handling large-scale data preparation and batch inference workloads efficiently.
Model Deployment Options Deploying trained machine learning models as production services is a core DP-100 exam domain that tests knowledge of multiple deployment targets and configurations available within Azure Machine Learning. Managed online endpoints provide real-time inference capabilities where client applications send individual prediction requests and receive immediate responses. Batch endpoints handle large-scale asynchronous scoring jobs where predictions are generated for entire datasets and stored for later retrieval. Candidates must understand the appropriate use cases, configuration requirements, and monitoring approaches for both endpoint types.
The exam also covers deployment configuration details including environment specification, resource allocation, scaling behavior, and authentication settings for online endpoints. Candidates must know how to deploy models to managed online endpoints using both the Python SDK and the Azure Machine Learning CLI, how to test deployed endpoints, and how to implement blue-green deployment strategies that allow new model versions to be validated against production traffic before full rollout. Understanding traffic splitting between multiple deployments behind a single endpoint is a specific configuration topic that appears in professional-level exam questions.
Responsible AI Considerations
Responsible AI practices are an increasingly prominent topic in the DP-100 exam, reflecting the growing importance of fairness, explainability, and ethical considerations in professional machine learning development. Microsoft’s Responsible AI framework, which encompasses principles including fairness, reliability, privacy, inclusiveness, transparency, and accountability, provides the conceptual foundation for this exam domain. Candidates must understand how these principles translate into specific technical practices within Azure Machine Learning workflows.
The Responsible AI dashboard within Azure Machine Learning is a key tool that candidates must know how to configure and interpret. It integrates multiple analysis components including model error analysis, fairness assessment, data exploration, and model explainability through feature importance calculations. Candidates must understand how to use these tools to identify model behaviors that may be problematic for specific demographic groups or data segments. Knowledge of interpretability techniques including permutation feature importance and SHAP values, and how Azure Machine Learning supports these approaches, is directly tested in the exam.
Hyperparameter Tuning Strategies Hyperparameter tuning is the process of systematically searching for the optimal configuration values for a machine learning algorithm, and Azure Machine Learning provides dedicated tooling for this process through its sweep job functionality. The DP-100 exam tests candidates on how to configure hyperparameter sweep jobs including defining the parameter search space with discrete and continuous distributions, selecting sampling strategies such as random sampling, grid sampling, and Bayesian optimization, and specifying early termination policies that stop underperforming runs to conserve compute resources.
Candidates must understand the trade-offs between different sampling strategies in terms of computational cost and the thoroughness of the search. Random sampling is computationally efficient and often finds good solutions quickly but may miss the optimal configuration. Bayesian optimization uses results from previous runs to guide subsequent sampling decisions, which can find better configurations with fewer total runs on complex problems. Early termination policies including Bandit, Median Stopping, and Truncation Selection each use different criteria to identify and terminate runs that are unlikely to outperform the current best result.
MLflow Integration Knowledge
MLflow is an open-source machine learning lifecycle management platform that Azure Machine Learning has adopted as its primary tracking and model management framework. Deep knowledge of MLflow is essential for DP-100 candidates because the exam assumes that candidates understand how to use MLflow APIs for logging metrics, parameters, and artifacts from training scripts, how to register models in the MLflow model registry, and how to load registered models for inference. MLflow’s integration with Azure Machine Learning allows practitioners to use familiar open-source tooling while benefiting from Azure’s managed infrastructure.
The exam also covers MLflow model flavors, which define how models are serialized and loaded for different machine learning frameworks including scikit-learn, PyTorch, and TensorFlow. Candidates must understand how flavor-specific model logging works and how the MLflow pyfunc flavor provides a framework-agnostic interface for serving models that were trained using different libraries. Knowledge of how to use MLflow for comparing experiment runs across different training sessions and selecting the best model for registration is tested as part of the broader experiment management knowledge domain.
Demonstrate these skills by completing a responsible model lifecycle with versioned data, evaluation evidence, monitored deployment and a documented rollback plan.