Renewal2 - Work with Compute Targets

Question 1

The 5 computer types/targets you can use

Answer 1

• Compute Instances
• Computer Clusters
• Kubernetes Clusters
• Attached Compute
• SERVERLESS Compute (new one!) - Spark or Azure Machine Learning Serverless Computes (Managed Compute!)

Question 2

Use these compute targets during experimentation and development, and when you prefer working in a Jupyter notebook. A notebook experience benefits most from a compute that is continuously running.

Answer 2

Compute Instances, Spark Serverless Computer

Question 3

Use these compute targets when moving to production and you want the compute target to be ready to handle large volumes of data i.e. scaling needs

Answer 3

Compute Clusters, Azure Machine Learning Serverless Compute

Question 4

Use these compute targets when generating batch predictions, particularly in a pipeline job in Azure Machine Learning, since you’ll need them on-demand and scalable…

Answer 4

Computer Clusters and AML Serverless Compute

Question 5

You want it light weight and cost efficient…

Use this compute target when generating real-time predictions, where you’d want the compute to run continuously, where you manage the compute

Answer 5

Kubernetes Clusters!

Question 6

Not really a Compute Target…you wanted to do this in our own ML

Use this compute target when generating real-time predictions, where you’d want the compute to run continuously, where you let Azure manage the compute

Answer 6

Managed Online Endpoints with Azure ML managing the Containers

Question 7

Compute Instance requirements

Answer 7

• Unique name across REGIONS
• To work on an instance, it needs to be assigned to your user account
• Can only be assigned to one user at a time (can’t handle Parallel Workloads)

Question 8

Compute Cluster creation and parameters to consider

Answer 8

• size: VM type and compute option (CPU or GPU)
• max_instances: Max nodes per cluster
• tier: low priority or dedicated
• idle_time_before_scale_down: The length of time your cluster’s nodes stay alive before they scale down. Keep this high if you want to reduce startup times and scaling costs

Tier - setting Low Priority can lower costs but no guarantee of availability

Question 9

Three scenarios you would use a Compute Cluster and why

Answer 9

• Running a pipeline job you built in the Designer.
• Running an Automated Machine Learning job.
• Running a script as a job.

Because each scenario allows you to scale up or down as necessary. Clusters also allow you to train multiple models in parallel.