DP-100: Designing and Implementing a Data Science Solution on Azure

Domains include: (1) Design and prepare a machine learning solution (~20-25%), (2) Explore data and train models (~35-40%), (3) Prepare a model for deployment (~20-25%), and (4) Deploy and retrain a model (~10-15%).

A Compute Instance is a single-node cloud development workstation for running Jupyter notebooks. A Compute Cluster is an auto-scaling multi-node cluster for executing distributed model training jobs and pipelines.

Azure ML supports native MLflow tracking, logging parameters, metrics, artifacts, and models during training runs with zero proprietary lock-in.

AutoML automates algorithm selection, hyperparameter tuning, feature engineering, and cross-validation for tabular classification, regression, and forecasting tasks.

The Python SDK v2 uses `@pipeline` and `@command_component` decorators to define modular, reusable, containerized steps executed on remote compute targets.

Managed Online Endpoints serve real-time predictions with sub-second latency and traffic splitting. Batch Endpoints process asynchronous batch inference over large datasets asynchronously.

Hyperdrive automates hyperparameter sweeping using search algorithms (Random, Grid, Bayesian) and early termination policies (Bandit, Median Stopping) to save compute costs.

Data Assets track versions, lineage, and schemas of datasets stored in Azure Blob or Data Lake Storage, ensuring experiments can be reproduced reliably.

The Responsible AI dashboard provides model explainability (SHAP values), fairness assessment, error analysis, and counterfactual analysis to identify model bias.

DP-100 is valid for one year, renewable annually for free via Microsoft Learn.

DP-100 certifies Data Scientists, Machine Learning Engineers, and MLOps Specialists earning average US salaries between $130,000 and $175,000.