Path1.Mod1.g - Explore ML Workspace - Train Models in the Workspace

Question 1

You know this from AI-900

Four options for training models in Azure ML Workspaces

Answer 1

• Azure ML Designer
• Automated ML
• Jupyter Noteooks
• Run a script as a job (python code)

Question 2

When to use Azure ML Designer

Answer 2

• For easy and quick exploration during the initial phase of training ML models.
• Swiftly create Pipelines using your workspace-registered custom Components

Question 3

When to use Automated ML

Answer 3

When you have a training dataset and you need to find the best performing model; you’ll want to experiement with various algorithms and hyperparameters as quickly as possible.

AutoML will iterate through algorithms, hyperparameter tuning and feature selections to find the best performing model for your data.

Question 4

When to use Jupyter Notebooks

Where all files are stored, how to run them and where you can edit them.

Answer 4

When you prefer to develop by running code in notebooks (inline documentation, snippet execution and immediately visible output).

• Files cloned or created are stored in the file share of the workspace’s Storage Account
• Notebook Kernels run on the Compute Instance
• Edit and run in VS Code or in your Workspace via Compute details > Terminal

Question 5

Co Sw Pi

The three types of Jobs for Running a Script as a Job and their execution considerations

Answer 5

• Command - Single script execution
• Sweep - Single script execution with Hyperparameter tuning
• Pipeline - Multi-script/Component execution

The main consideration is when you want to prep your code for production readiness. Easier to automate the process. Run the script as a job in Azure Machine Learning so that all inputs/outputs are stored in the Workspace.

Question 6

The two scenarios that run as a Pipeline Job

Answer 6

• When you submit a pipeline you created with the designer
• When you submit an Automated Machine Learning experiment

Question 7

Restricting access when creating a new ML Workspace

Answer 7

Use advanced options through a private endpoint and specifying custom keys for data encryption

Question 8

Larger VM size vs Smaller VM size (when creating a Compute Instance)

Answer 8

A larger image may incur higher cost and a smaller image may not be sufficient to complete the tasks.

Question 9

IO

Running a script or Pipeline as a Job allows you to define…

Model Tracking with Jobs

Answer 9

Jobs allow you to define inputs and document any outputs

Jobs keep track of different models you train to compare and identify the best model