Challenge 09: Azure Machine Learning Workspace
25-35 min | Cost: Free | Domain: Machine Learning on Azure (15-20%)
Exam skills covered
- Describe capabilities of Automated Machine Learning
- Describe data and compute services for machine learning
- Describe model management and deployment capabilities in Azure ML
Overview
Azure Machine Learning (Azure ML) is a cloud platform for building, training, and deploying machine learning models. The workspace is the central hub — think of it as your ML laboratory that contains all your experiments, data, models, and compute resources in one organized space.
Think of Azure ML workspace like a professional kitchen. The kitchen (workspace) contains ingredients (datasets), cooking equipment (compute), recipes (notebooks/pipelines), finished dishes (trained models), and a serving counter (endpoints for deployment). Everything you need for the entire ML lifecycle is in one place.
Azure ML supports three main approaches: Automated ML (the platform builds models for you), Designer (drag-and-drop visual pipelines), and Notebooks (write code directly). For the AI-900 exam, focus on understanding what each component does rather than how to code with it.
Explore
Task 1: Azure ML workspace components
| Component | Purpose | Analogy |
|---|---|---|
| Workspace | Top-level container for all ML assets | The laboratory |
| Datastores | Connections to data storage (Blob, SQL, etc.) | Ingredient pantry |
| Datasets | Registered, versioned collections of data | Prepared recipe ingredients |
| Compute instances | VMs for development (notebooks) | Your personal workstation |
| Compute clusters | Scalable VMs for training jobs | Industrial kitchen ovens |
| Experiments | Records of training runs with metrics | Lab notebook with results |
| Models | Trained ML models (registered and versioned) | Perfected recipes |
| Endpoints | Deployed models serving predictions | Restaurant serving counter |
| Pipelines | Automated workflows (data prep → train → deploy) | Assembly line |
Task 2: Navigate Azure ML Studio
- Visit ml.azure.com (Azure Machine Learning Studio)
- If you don't have a workspace, explore the interface conceptually:
- Left menu: Author (Notebooks, Automated ML, Designer) | Assets (Data, Jobs, Models, Endpoints) | Manage (Compute)
- Key areas to understand for the exam:
- Automated ML: Upload data → choose target column → Azure builds the best model
- Designer: Visual drag-and-drop pipeline builder
- Models: Registry of all trained models with versions
- Endpoints: Where deployed models serve predictions
Task 3: Understand Automated ML capabilities
Automated ML (AutoML) automates the most time-consuming parts of machine learning:
| Step | What AutoML does automatically |
|---|---|
| Algorithm selection | Tests multiple algorithms (logistic regression, decision trees, gradient boosting, etc.) |
| Hyperparameter tuning | Finds the best settings for each algorithm |
| Feature engineering | Creates and selects the most useful features |
| Model evaluation | Compares all models using appropriate metrics |
| Best model selection | Returns the top-performing model |
AutoML supports three task types:
- Classification — predict categories
- Regression — predict numbers
- Time-series forecasting — predict future values over time
Task 4: Model deployment and endpoints
Once a model is trained, it needs to be deployed so applications can use it:
| Deployment type | Use case | How it works |
|---|---|---|
| Real-time endpoint | Immediate predictions | Application sends data, gets prediction back instantly (REST API) |
| Batch endpoint | Process large datasets | Submit a dataset, get predictions back later (asynchronous) |
Deployment workflow:
- Register the model in the Model Registry
- Create an endpoint (real-time or batch)
- Deploy the model to the endpoint
- Test by sending data and receiving predictions
- Monitor performance and data drift over time
# List Azure ML workspaces in your subscription
az ml workspace list --output table
# List compute instances in a workspace
az ml compute list --workspace-name my-workspace --resource-group my-rg --output table
# List registered models
az ml model list --workspace-name my-workspace --resource-group my-rg --output table
Key Concepts
| Concept | Definition |
|---|---|
| Azure ML workspace | Central hub containing all ML resources: data, compute, models, endpoints |
| Automated ML (AutoML) | Feature that automatically trains and compares multiple models, selecting the best one |
| Compute instance | Single VM for development and testing (notebooks) |
| Compute cluster | Scalable group of VMs that grows/shrinks based on training job demand |
| Model registry | Versioned catalog of trained models for tracking and deployment |
| Real-time endpoint | Deployed model that returns predictions immediately via REST API |
| Batch endpoint | Deployed model that processes large datasets asynchronously |
| ML pipeline | Automated, repeatable workflow for the full ML process |
| Designer | Visual drag-and-drop tool for building ML pipelines without code |
Common Misconceptions
| Misconception | Reality |
|---|---|
| "You need to know how to code to use Azure ML" | Azure ML Designer provides a drag-and-drop visual experience, and Automated ML builds models with minimal configuration. Code (Python) is optional |
| "Automated ML produces production-ready models every time" | Automated ML gives you a great starting point, but production deployment often requires additional evaluation, testing, and monitoring |
| "A compute instance is required to deploy a model" | Compute instances are for development. Deployed models run on managed endpoints — separate infrastructure optimized for serving predictions |
| "Once deployed, a model never needs updating" | Models degrade over time as real-world data changes (concept drift / data drift). Monitoring and retraining are ongoing needs |
| "Azure ML and Azure AI services are the same thing" | Azure AI services provide pre-built, ready-to-use AI APIs. Azure ML is for building your OWN custom models from your data |
Knowledge Check
1. A data scientist wants Azure to automatically try multiple algorithms and select the best-performing model for their dataset. Which Azure ML feature should they use?
2. What is the purpose of a real-time endpoint in Azure Machine Learning?
3. Which Azure ML component provides a visual, drag-and-drop experience for building machine learning pipelines?
4. What is the difference between a compute instance and a compute cluster in Azure ML?
5. What is the primary difference between Azure Machine Learning and Azure AI services?