Azure Machine Learning Workspace
Central resource for ML development
Core Components
| Component | Purpose | Key Features |
|---|
| Workspace | Top-level resource | Contains all ML assets |
| Compute Instances | Development VMs | Jupyter, VS Code integration |
| Compute Clusters | Training resources | Auto-scaling, low-priority VMs |
| Datastores | Data connections | Azure Storage, ADLS, SQL |
| Data Assets | Versioned datasets | URIs, MLTable, file references |
| Environments | Runtime configs | Docker images, conda specs |
Associated Resources
Storage Account
Default datastore, artifacts, logs
Key Vault
Secrets, connection strings, keys
Container Registry
Docker images for environments
Application Insights
Monitoring, telemetry, logs
Compute Types
- Compute Instance: Single VM for development and testing
- Compute Cluster: Multi-node for distributed training
- Kubernetes: AKS for inference workloads
- Serverless: On-demand compute for training jobs
- Attached: External compute (HDInsight, Databricks)
Exam Focus Areas
- Compute clusters auto-scale to 0 nodes when idle
- Use low-priority VMs for cost savings (may be preempted)
- Datastores abstract connection details from code
- Data assets enable versioning and lineage tracking
